Confidently Wrong: What AI Hallucinations Teach Us About Ourselves

September 7, 2025

Stop Rewarding Bluffing: The Hidden Lesson Behind AI Hallucinations

In every age, we’ve built tools that mirror us more than we realize. The printing press amplified our words, the telescope extended our sight, and now language models echo our thinking—confident, fluent, and sometimes gloriously wrong. Their so-called “hallucinations” are not the fever dreams of machines but the logical outcome of how we train and reward them. We ask them to perform like students under exam pressure, where the prize goes not to the cautious but to the bold guesser. And so, like us, they learn to bluff when uncertain.

But here’s the twist: these machines are not just reflecting our intelligence—they’re holding up a mirror to our own blind spots. When we watch them conjure answers out of thin air, we’re also watching a reflection of the boardrooms that reward confident speeches, the classrooms that punish “I don’t know,” and the cultures that confuse certainty with wisdom. To study why machines hallucinate is, in many ways, to study ourselves: how we learn, how we lead, and how we sometimes stumble in our pursuit of truth.

The Test-Taking Machine: Why AI Hallucinates (and How Better Grading Can Make It Honest)

The Confident Student Problem

Every classroom had that one student. You know the type. The hand shot up before the teacher finished asking the question. The answer? Delivered with the swagger of absolute certainty. And then, inevitably, spectacularly wrong.

Now imagine scaling that student to 600 billion parameters and hooking them up to the internet. Congratulations—you have today’s large language models.

Researchers call this habit hallucination. I prefer the less mystical phrase: confident bluffing. And as a new paper by Kalai and colleagues (Why Language Models Hallucinate, 2025) reminds us, it’s not a bug. It’s the system working exactly as designed.

https://openai.com/index/why-language-models-hallucinate

Hallucinations, Demystified

First, let’s clear the fog. When we say AI hallucinates, we don’t mean it’s seeing pink elephants or hearing phantom voices. In machine terms, a hallucination is a plausible but false statement.

Ask for a scholar’s dissertation title, and you may get a convincingly worded—but entirely fabricated—response. Ask it to count the Ds in “DEEPSEEK,” and you’ll receive answers ranging anywhere from 1 to 7. All plausible, none correct.

This isn’t nonsense. It’s not the machine babbling. It’s the machine playing the only game we taught it: guess, and guess with confidence.

Why Do Machines Bluff?

Here’s the dry truth: hallucinations are the predictable outcome of math and incentives.

Pretraining (the foundation). A model learns the statistical patterns of text. Even if the training data were perfect, the model would still misfire because language is messy. It faces the “Is-It-Valid” challenge: for every possible response, decide if it’s valid or an error. Spoiler: no model can sort perfectly. And when it misses, out comes a hallucination.
Singletons (the lonely facts). Think of obscure trivia—say, a person’s birthday that appears once in the training data. There’s no pattern to learn, no redundancy to anchor it. The paper shows that the fraction of such one-off facts (“singleton rate”) sets a hard lower bound on how often the model will hallucinate. No amount of wishful prompting will change that.
Post-training (the bluffing school). Here’s the kicker: after pretraining, we fine-tune models with benchmarks that punish hesitation. On most tests, saying “I don’t know” earns you zero. A wrong but confident guess? At least you’ve got a shot. The rational strategy is always to bluff. So that’s what the machine does. Endlessly. Relentlessly. Just like that overconfident student.

The Wrong Kind of Evolution

Nature has a simple punishment for bluffing: you guess wrong, you don’t survive. The gazelle doesn’t tell the lion, “Actually, I think you might be vegetarian.” But in our digital ecosystems, we’ve inverted the rules. We’ve built leaderboards and benchmarks that reward performance over prudence, speed over humility.

The result? We’ve trained our machines to be expert test-takers, not reliable truth-tellers. They are overfit not just to language, but to the warped incentives of our grading systems.

The Fix Is Simpler Than You Think

The authors propose a refreshingly simple remedy: change the grading system.

Instead of binary scoring (1 point for right, 0 for wrong or abstain), give partial credit for honesty. Here’s the formula:

Answer only if you’re more than t confident.
If wrong, lose t/(1–t) points.
If right, get 1 point.
If unsure, say “I don’t know” for 0 points.

At t = 0.75, a wrong answer costs you 2 points. Suddenly, guessing is punished. The rational strategy shifts: bluff less, calibrate more.

It’s the same trick human exams like the SAT once used, penalizing wrong guesses to separate the humble from the reckless. The machine, like the student, adapts to whatever scoring we set.

Why This Matters Beyond AI

This isn’t just about machines. It’s about us.

We live in a culture that too often mistakes confidence for competence. Smooth talk passes for smart talk. Benchmarks reward volume over nuance, certainty over reflection. And just like the models, we adapt—bluffing when unsure, masking ignorance with performance.

The paper is a mirror. It shows that hallucinations aren’t strange computer glitches—they’re what happens when intelligent systems (silicon or biological) are trapped in warped incentive games.

So What Do We Do?

If we want trustworthy AI, we need to reward honesty. If we want trustworthy humans, we need to do the same. That means:

Designing evaluations that value uncertainty. In AI and in people.
Building cultural safety for “I don’t know.” In workplaces, schools, communities.
Tracking calibration, not just accuracy. Did you know when you didn’t know? That’s the real score.

Closing: The Return of the Confident Student

So let’s return to that student in the classroom. Imagine if the teacher said: “You only get credit if you’re sure. Otherwise, say ‘I don’t know’ and I’ll respect that.” How quickly would our classrooms change? How quickly would our boardrooms change? How quickly would our machines change?

AI hallucinations aren’t alien. They’re human. They’re a reflection of us. If we want machines that are humble, calibrated, and trustworthy—maybe we should start by building a culture that rewards those qualities in ourselves.

Because in the end, the problem isn’t that the machine is bluffing. The problem is that we taught it to.

👉 Call to Action: At TAO.ai, we’re exploring how to design communities, metrics, and technologies that reward honesty, humility, and collective intelligence. Join us as we test new “confidence-aware” evaluations in our AnalyticsClub challenges. Let’s see what happens when we stop rewarding bluffing—and start rewarding truth.

Humble Intelligence: What Our Brains Can Learn from Bluffing Machines

The Gazelle Doesn’t Bluff

In the savannah, a gazelle does not bluff a lion. If it guesses wrong, there’s no retake. Yet in human habitats—schools, workplaces, even social media—bluffing is strangely rewarded. We nod to the confident speaker, even if they’re confidently wrong.

And now, our machines are doing the same. Why? Because we built their report cards.

The recent Why Language Models Hallucinate paper reveals a sobering truth: AI hallucinates not because it’s broken, but because our systems reward confident answers over honest uncertainty. The machine is simply mirroring us.

So here’s the real question: What can our brains learn from our bluffing machines?

Lesson 1: Confidence Is Not Competence

AI’s biggest failing is also humanity’s favorite bias: equating certainty with truth.

Language models score higher when they guess confidently, even if wrong. Humans? We do the same. The person with the loudest voice in the room often shapes decisions, regardless of accuracy.

The lesson is clear: just because something is said fluently, doesn’t make it fact. We need to train ourselves—individually and collectively—to separate style from substance.

Lesson 2: Make Space for “I Don’t Know”

Machines avoid “I don’t know” because benchmarks punish it. People avoid it because culture punishes it.

Imagine if in a meeting, saying “I don’t know, but I’ll find out” earned as much credit as giving a half-baked confident answer. That small redesign would change how teams learn. It would normalize humility, and paradoxically, speed up progress—because we’d stop chasing the wrong paths so confidently.

In other words: abstention is not weakness. It’s wisdom.

Lesson 3: Respect the Singleton

In machine learning, a singleton is a fact seen only once in training—an obscure birthday, a rare law, a unique case. These are exactly where hallucinations spike.

In human learning, we have our own singletons: first-time challenges, new markets, unprecedented crises. Yet instead of slowing down, we often speed up—confidently winging it.

The takeaway? Treat new, rare situations with care. Pair up. Research harder. Call the mentor. The brain’s singleton rate is high enough already; no need to bluff through it.

Lesson 4: Know Your Model Limits

Machines hallucinate when their internal models don’t fit reality—like tokenizing “DEEPSEEK” into chunks that make counting Ds nearly impossible.

Humans hallucinate too, but we call it “bad assumptions.” When we use the wrong mental model, we miscount, misinterpret, and mislead ourselves.

The lesson: upgrade the model, not just the willpower. Read widely. Reframe problems. Don’t be the trigram model in a world that requires deeper reasoning.

Lesson 5: Redesign the Grading

Ultimately, hallucinations—human or machine—are about incentives. If bluffing earns more points than honesty, bluffing becomes rational.

The paper proposes a fix for AI: scoring systems that penalize wrong guesses more than abstentions. Humans could use the same. Imagine performance reviews that reward calibrated honesty over overconfident error. Imagine classrooms where students earn points for saying “I’m not sure, here’s my reasoning”.

We don’t need to teach people (or machines) to be less human. We need to redesign the exam.

The Worker1 Playbook: Practicing Humble Intelligence

So how do we apply this in daily life, as individuals and teams?

Set thresholds. Decide your personal “confidence t.” For a business decision, maybe 90%. For a brainstorm, 60%.
Practice IDK rituals. Try this script: “Tentative take (70%): … I’ll confirm by Friday.” Simple, safe, clear.
Track calibration. Journal predictions and outcomes. Over time, you’ll learn if you’re an under-confident sage or an overconfident bluffer.
Singleton protocol. For new, rare tasks: pause, research, collaborate. Treat them as high-risk zones.
Make humility visible. In teams, celebrate the person who flags uncertainty, not just the one who speaks first.

What This Means for Communities

Strong workers build strong communities. Strong communities nurture strong workers. But only if those communities value honesty as much as output.

At AnalyticsClub, we’re experimenting with challenges that reward not just accuracy but calibration—did you know when you didn’t know? At Ashr.am, we’re building spaces where workers can exhale, say “I don’t know,” and find support instead of stress. Through the HumanPotentialIndex, we’re exploring ways to measure not just skill, but wisdom: the courage to pause, to question, to admit uncertainty.

This isn’t just about building smarter machines. It’s about building wiser humans.

Closing: Gazelles, Lions, and Leaders

Back to the gazelle. In the savannah, bluffing is fatal. In our modern world, bluffing can win you promotions, followers, and funding. But it also corrodes trust, slows learning, and eventually collapses communities.

Our machines are showing us a mirror: they bluff because we do. If we want AI to be humble, we must first cultivate humility ourselves.

Because in the end, the most powerful intelligence—human or machine—isn’t the kind that always has an answer. It’s the kind that knows when not to.

👉 Call to Action: Join us in rethinking how we learn, lead, and build together. What if our teams and technologies were rewarded for humility as much as for output? At TAO.ai, that’s the future we’re working toward. Come be part of the experiment.

In the end, the story of hallucinating machines is not about machines at all—it is about us. We built systems that reward performance over humility, and they learned our lesson a little too well. If we want AI that is trustworthy, we must design for honesty, not bravado. And if we want communities that are resilient, we must celebrate curiosity over certainty, calibration over bluffing.

The gazelle survives not by pretending to know the lion’s next move, but by respecting uncertainty and reacting wisely. Perhaps our greatest intelligence—human or artificial—will not be measured by the answers we give, but by the courage to admit when we don’t know, and the wisdom to learn what comes next.

So here is the challenge before us: to reimagine our tests, our workplaces, and our conversations in ways that reward truth-telling and humility. Because if we can teach our machines to be honest, maybe we’ll remember how to be honest with ourselves.

When Platform Rules Become Workplace Rules: Apple Pushes Back Against Mandated App Store Messaging

Bella Hayes

August 30, 2025

When Platform Rules Become Workplace Rules: Apple Pushes Back Against Mandated App Store Messaging

How a legal battle over App Store anti‑steering rules ripples through product teams, customer success groups and the way companies talk to users.

The case at hand, and why it matters to work

Apple recently told the Ninth Circuit that a lower court’s order requiring changes to its App Store anti‑steering rules is unlawful and unconstitutional. In a forceful reply brief, the company pushed back against Epic Games’ position and sought to block or narrow the court’s mandated alterations to how Apple controls in‑app communications and its broader business practices.

At first blush, this may read like another round in a long legal saga about marketplaces and monopoly power. But the contours of the dispute touch a much broader audience: product managers who build in‑app journeys, legal and compliance teams who translate court orders into corporate practice, customer success reps who craft messages that balance persuasion with policy, and leaders who must anticipate how changing platform rules will affect revenue, trust and employee workflows.

What Apple says it’s fighting

Apple argues that the court’s order went beyond a simple remedy for wrongdoing. It contends the injunction is overbroad — forcing speech and conduct that extend past the narrow violations at issue — and therefore unconstitutional. The company frames the mandate as compelled speech and a form of judicial micromanagement that could dictate the content of commercial communications across its platform.

In practical terms, the dispute centers on anti‑steering rules: policies that limit how app developers can direct users to payment options outside the App Store. The court ordered changes meant to let developers communicate more freely about alternative payment methods, but Apple says the changes would require it to allow messaging and behaviors that undermine its policies and its tightly woven product, privacy and security model.

Why this is a workplace story, not just a courtroom drama

Judicial decisions about platforms don’t stay confined to the pages of legal briefs. They become operational playbooks for thousands of employees and partners. Consider how a mandate to allow broader in‑app communications would cascade across an organization:

Product teams would need to redesign user journeys and rework app reviews and SDKs to accommodate new messaging flows.
Legal and compliance would be tasked with interpreting the narrowness of any ruling and drafting new policies that balance regulatory requirements and business interests.
Customer success and marketing must rewrite scripts and help center content to reflect what may or may not be permitted at different times and in different markets.
Finance and partnerships teams would have to model changed revenue patterns as alternative payment channels and third‑party processors enter the equation.
Security and privacy engineers would assess what these communications mean for fraud, data handling, and user safety.

For workplaces, the question becomes less about who is right in the abstract and more about how to maintain continuity and trust amid shifting legal and policy landscapes.

The broader tension: policy, speech and commerce

This dispute sits at the intersection of three forces that shape modern work: platform governance, commercial speech, and judicial oversight. Apple characterizes the court order as judicial overreach that risks commandeering how a private company governs its platform — including the speech of its users and business partners. Critics worry that such arguments can be used to shield anticompetitive conduct.

For workplaces that operate on or alongside dominant platforms, the practical implications are concrete: what companies can say in product prompts, what alternatives developers can offer, and how transparent businesses must be about fees, payment options, or third‑party relationships.

Whether the Ninth Circuit narrows, stays, or affirms the lower court’s order will set precedents for how much leeway platforms have to prescribe the user experience, and how much power courts have to reshape that experience in the name of competition or speech rights.

What leaders should be doing now

Uncertainty is the enemy of good execution. Organizations that rely on app ecosystems should take steps now to reduce risk and stay nimble as the legal picture evolves:

Scenario plan. Create a short list of plausible outcomes from the Ninth Circuit — full reversal, partial narrowing, or enforcement of the prior order — and map operational responses for each.
Modularize product changes. Design payment and messaging systems so alterations can be toggled, limited, or expanded without sprawling engineering rewrites.
Prepare user communications. Draft multiple versions of customer messages and support scripts keyed to different policy states so customer success teams can switch quickly without legal bottlenecks.
Measure trust impact. Track user engagement and trust metrics around payment messaging and opt‑in behaviors. That data will inform whether policy changes are improving or degrading the customer relationship.
Coordinate cross‑functionally. Legal, product, marketing and compliance must align constantly — decisions in one room ripple into three or four others almost immediately.

Implementing these steps is not about picking sides in a legal fight. It’s about building workplaces that can respond intelligently when platform rules — and the courts that interpret them — shift beneath their feet.

Opportunities hidden in constraint

Legal and regulatory pressure is often framed as a threat; it can also be a source of competitive advantage. When platform controls loosen or become more prescriptive, companies that have already built flexible systems, clear messaging strategies, and a deep understanding of their customers will be better positioned to act quickly and ethically.

What looks like a restriction to one team can be an opportunity for another: clearer disclosure requirements can strengthen trust, alternative payment options can reduce churn if implemented thoughtfully, and more transparent dialogue with customers can become a differentiator in a crowded market.

What to watch next

The Ninth Circuit’s reaction will be instructive not only for Apple and Epic but for any company operating within large ecosystems. Watch for several signs:

Whether the court focuses on narrow statutory remedies or takes a broader view of constitutional constraints on equitable relief.
How any decision balances consumer protection and competition with property rights and free‑speech principles.
Signals to other platforms and regulators about the legitimacy of using court orders to force changes in platform governance.

Each of these will inform the next chapter of how workplaces design policy, product and messaging strategies in a world where courts, regulators and platforms interact in unpredictable ways.

From Briefing Rooms to Morning Airwaves: Dani Burger’s Leap to Bloomberg’s Open Interest — A Career Playbook for the Work News Community

Sophia Kane

August 29, 2025

From Briefing Rooms to Morning Airwaves: Dani Burger’s Leap to Bloomberg’s Open Interest

Next week, Dani Burger will step into a new daily rhythm — joining Bloomberg’s morning show, “Bloomberg Open Interest,” as co-anchor, leaving behind her current role at “Bloomberg Brief.” For the community that watches how careers are shaped in newsrooms and beyond, this move is more than a personnel announcement: it’s an instructive case in professional evolution, visibility, and the craft of connecting work to audiences at scale.

Why this shift matters to the Work news community

Transferring from a brief-focused role to a live morning program reframes the work itself. “Bloomberg Brief” is a format built on compact analysis and curated takeaways. A morning show lives in a different tempo: it is conversational, immediate, and highly performative. For colleagues, aspiring anchors, producers, and newsroom leaders, the transition highlights the varied skill sets that modern journalism — and modern workplaces — demand.

For the audience that follows work news, Dani’s move signals a few key realities. First, career progression is rarely linear; lateral moves into higher-visibility roles can accelerate influence and impact. Second, the ability to translate deep subject knowledge into accessible, live conversation is a high-value workplace capability. Third, organizations reward adaptability: the people who can translate their craft across formats often become the new face of their teams.

Three professional shifts embedded in the change

From crafted dispatches to live narrative:
Working on briefs emphasizes precision — a well-edited paragraph, a distilled insight. Morning shows require improvisation, pacing, and the capacity to hold narrative threads across live segments. This is a shift from the solitary revision process to a collaborative, instantaneous form of storytelling.
Visibility and responsibility:
On-air roles come with amplified visibility. That brings opportunity — the ability to shape public conversation — and responsibility, as every moment is subject to real-time reaction. For professionals, this underscores the tradeoffs of high-profile work: more influence, yes, but also a need for steadier presence and deliberate voice management.
Audience-first thinking becomes operational:
Briefs appeal to readers seeking efficient takeaways. Morning television must balance depth with immediate relevance to a diverse, time-pressed audience. The transition is a reminder that knowing your audience and tailoring delivery is as much an operational discipline as an editorial one.

Lessons for workers and newsroom leaders

Dani Burger’s move offers practical lessons that apply beyond broadcasting. Consider these takeaways for career development, leadership, and team design.

Embrace transferable skills:
Clarity, curiosity, and the ability to synthesize complex information are portable. The format may change, but the core skills remain valuable. Advocate for roles that allow you to demonstrate those skills in new contexts.
Make room for visible experiments:
Organizations that create low-risk pathways to higher-profile work — guest co-hosts, special segments, cross-platform storytelling — cultivate internal talent and broaden institutional voice.
Learn the rhythms of new platforms quickly:
Every platform has a tempo. Morning shows are driven by time cycles, audience influx, and bridging news and markets. When stepping into a new role, prioritize rapid tempo acclimation: rehearsal, short-form practice, and iterative feedback.
Align personal brand with organizational mission:
A co-anchor role ties an individual more tightly to a program’s identity. Thoughtful alignment between personal voice and institutional values makes transitions smoother and more authentic.
Support structures matter:
Behind every visible on-air persona is a team — producers, researchers, engineers. Leaders should invest in that network to make visibility sustainable and to spread institutional knowledge.

What to watch as she begins

In the coming weeks, the Work news community should look for a few signals that reveal how this change will unfold:

How segments adapt: Will the show lean into more analytical briefing moments reflecting Dani’s background, or will it expand into new conversational beats?
Audience engagement: Morning audiences have particular needs — energy, clarity, and utility. Tracking audience response will show how well format and personality align.
Cross-team learning: Will lessons from brief-form journalism influence the show’s editorial cadence, and vice versa? Productive cross-pollination could reshape internal workflows.

A reminder about career narratives

Career arcs are often presented as tidy ladders. Dani Burger’s move reminds us they are ladders built on bridges — lateral shifts, public-facing opportunities, and moments when specialized craft is translated into broader conversation. For those watching or charting their own path, the message is encouraging: deliberate transitions, supported by skillful storytelling and team infrastructure, can create outsized impact.

Closing: A moment of craft and possibility

As Dani Burger takes the co-anchor seat on “Bloomberg Open Interest,” the Work news community gets a live case study in the intersection of craft, visibility, and organizational design. This is a moment to learn: about how we prepare people for higher-profile roles, how we design teams to support visible work, and how professionals can carry their core strengths into new formats.

Whether you’re a journalist, an editor, a communications leader, or anyone thinking about the next move in your own career, watch closely. Transitions like this distill the practical wisdom of how work evolves in public-facing industries — and how individuals can seize the kinds of opportunities that reshape both their own trajectory and the narratives their organizations tell.

When Leadership Falls, Work Culture Hangs in the Balance: Lessons from a CDC Shakeup

Emma Vaughn

August 29, 2025

When Leadership Falls, Work Culture Hangs in the Balance: Lessons from a CDC Shakeup

News of a sweeping exit of senior leadership after the abrupt removal of an agency director arrived like a cold wind through a workplace already tested by crisis and scrutiny. For people who show up every day to protect the public’s health, the shock is not only about who will lead next. It is about what the departure signals — to staff, to partners, and to the public — about whether mission, merit and the machinery that delivers public services will withstand turbulence.

This is a workplace story as much as a public policy one: how teams cope when the top is hollowed out, how institutions preserve knowledge, and how leaders, managers and rank-and-file employees can keep the engines of an agency humming when governance becomes politicized. For the work community, the immediate questions are painfully practical. Who will approve budgets and guide response strategies? Who will mentor rising managers? Who will hold the institutional memory and the relationships that connect the agency to state and local counterparts?

The ripple effects of a leadership purge

Leadership changes are normal. Abrupt, transparent purges are not. When several senior officials resign together in the wake of a director’s ousting, the consequences are magnified. Operationally, projects stall. Reviews and clearances slow until delegated authority is reestablished. Externally, partners find it harder to coordinate; internally, staff members wonder whether decisions will be made on scientific and technical merit or political expediency. Morale takes a hit — not necessarily because the departing leaders were beloved, but because the pattern of exits signals instability.

Beyond immediate halts in workflow, such shakeups can undercut long-term confidence. Experienced staff may read a clear message: career trajectories that once rewarded competence and stewardship are now vulnerable to sudden reversal. When people who have invested years in an institution conclude that their work will be subject to arbitrary or politically driven change, retention becomes a problem and institutional memory walks out the door.

Morale and the silent leavers

Resignations at the top often precede quieter departures lower down. The ‘silent leavers’ — those who don’t make headlines but quietly leave for the private sector, academia, or other agencies — are a real risk. Their departure drains the organization of specialized expertise, relational capital, and operational agility. Turnover also carries hidden costs: recruitment, onboarding, lost productivity, and the time managers must invest in rebuilding teams.

Morale is more than an HR metric; it determines how rapidly an agency can respond to emergencies, whether staff will volunteer for difficult assignments, and whether leaders can expect honest assessments rather than sycophantic echo. A workplace that values transparency and fair process is more likely to sustain commitment, even when political winds shift.

Guardrails for continuity

When top leadership changes, strong organizational guardrails keep things afloat. Clear succession pathways, robust delegation frameworks, and well-documented operating procedures help ensure continuity. These are not bureaucratic luxuries; they are the scaffolding that lets day-to-day work proceed while leadership transitions occur.

Practical measures include:

Codified delegation of authority so time-sensitive decisions do not need a single person’s approval.
Cross-training and job-sharing to distribute institutional knowledge across teams.
Comprehensive documentation of ongoing projects and the logic behind major policy choices.
Rapid appointment of interim leaders who are perceived as impartial and credible by staff and stakeholders.

Communication: the invisible anchor

In times of disruption, communication becomes an instrument of stability. Silence breeds rumor; vague reassurances breed cynicism. Effective communication balances candor with calm. It does not require revealing every detail of negotiations or personnel discussions, but it does demand a clear articulation of what will remain unchanged — mission, key priorities, service commitments — and what the timeline will be for leadership decisions.

Managers should aim to create routine touchpoints: frequent all-staff updates, Q&A sessions where concerns are heard and addressed, and visible commitments to preserve the core functions that staff care about most. Visible, routine communication reduces anxiety and demonstrates that leaders are managing the transition rather than being swept along by it.

Protecting the mission from politicization

One of the greatest fears for staff in a politicized replacement scenario is that technical judgments will become subordinate to political priorities. Protecting the mission means institutionalizing decision-making processes that prioritize evidence, transparency and collaboration. It also means creating mechanisms for staff to raise concerns without fear of retaliation, and for decisions to be documented and justified in ways that withstand external scrutiny.

An agency that can demonstrate the logic and data behind decisions is less vulnerable to accusations of bias and more resilient when leadership changes. It is also more likely to maintain credibility with partners and the public.

Investing in people during uncertainty

A surprising leadership move that pays dividends is to double down on people investments precisely when leadership is unsettled. Training, mentoring, and career-path clarity give staff reasons to stay. Programs that support well-being, that recognize contributions, and that foster internal mobility send the message that the agency values its human capital regardless of who occupies the corner office.

Retention strategies should be pragmatic: prioritize roles where turnover would be most damaging, create clear pathways for temporary promotions to shore up gaps, and offer flexible work arrangements that keep highly skilled staff engaged. Investing in managers is equally important; first-line supervisors are the ones who translate senior messaging into day-to-day experience.

Culture: the true ballast

Institutions are held together more by culture than by titles. A culture that prizes collegiality, rigorous debate, and a shared sense of purpose will weather political storms better than one dependent on charismatic individuals. Leaders can cultivate such a culture by modeling humility, inviting dissent, celebrating collective achievements and making transparent how decisions are made.

When leadership change is inevitable, a healthy culture allows teams to reconstitute quickly. It keeps the mission central and reduces the temptation to internalize external political dynamics.

What managers can do now

Managers play a decisive role during transitions. Concrete steps they can take:

Hold small-group conversations focused on what staff most need to do their jobs, not on speculation about politics.
Map critical dependencies and identify immediate risks to projects and services.
Secure interim authorities for essential functions and communicate those arrangements clearly.
Recognize and reward staff who step up during the transition, publicly and privately.
Encourage documentation and knowledge transfer sessions to capture institutional memory.

A call to steady hands and clear minds

The story of a leadership purge at a major public institution is a test of organizational resilience. The narrative need not end in fragmentation. It can become a turning point — if those who remain choose to shore up the work, preserve the institutional norms that sustain good decision-making, and invest in the people who do the day-to-day work.

That requires steady hands and clear minds: leaders who communicate frankly, managers who protect their teams, and staff who commit to the mission even as they seek accountability. It also requires external stakeholders — partners, funders and the public — to judge the agency by the continuity of its services and the integrity of its work, not by the headline cycle of turnover.

Leadership beyond titles

Finally, leadership in a time of upheaval is decentralized. It shows up in mid-level managers who keep operations running, in early-career employees who document and organize work, and in teams that prioritize mission over maneuver. Those acts of stewardship are the most reliable form of institutional insurance.

Change will come. How an agency fares depends less on who sits in the director’s chair and more on whether the workforce — from the mailroom to the executive floor — is prepared, supported and committed to a shared mission. For the broader work community, this moment is a reminder: resilience is built before a crisis, and the best legacy that departing leaders can leave is a culture that survives them.

AI Now Doing 90% of the Work, While Managers Proudly Claim 110% Credit

Dr. Chad Synergy

August 26, 2025

By The MORK Times Investigations Desk TheWorkTimes

In an unprecedented leap for productivity theater, a new report confirms what everyone with a boss already suspected: AI is now doing almost all the actual work, while humans are busy holding ‘vision alignment workshops’ about the work AI already finished last Tuesday.

The study, which combed through millions of Claude.ai chats, revealed that AI is performing a hefty portion of day-to-day tasks in software, writing, and anything else that involves typing words until someone approves them. In other words: AI has become the world’s most reliable junior employee—minus the kombucha addiction and passive-aggressive Slack emojis.

“People-as-a-Service”: A Revolution Nobody Wanted

According to researchers, about 36% of occupations now feature AI in at least a quarter of their tasks. For developers, this means AI writes the code, explains why the code doesn’t work, and then writes new broken code. For copywriters, it means generating 16 taglines that clients reject in favor of one written by the CEO’s niece.

Executives have coined a new buzzword for this shift: People-as-a-Service (PaaS™).

“Employees were always inefficient software,” said Greg Spindleton, Chief Synergy Officer at a mid-tier consultancy. “So we simply swapped them for literal software. It’s like Uber, but for your sense of dignity.”

Augmentation vs. Automation (a.k.a. Therapy vs. Child Labor)

Researchers split AI work into two neat categories:

Automation (43%) – AI does the task itself, flawlessly generating reports that no one reads.
Augmentation (57%) – AI gently pats employees on the head and whispers, “You’re still relevant,” while doing the hard part.

“This is about empowerment,” said Elaine Marcus, professor of Digital Exploitation at Harvard. “AI doesn’t take your job away. It just makes you realize your job never really mattered.”

Winners and Losers in the AI Economy

Winners: High-wage tech workers who already outsourced 90% of their day to StackOverflow.
Losers: Anyone who moves physical objects, touches living humans, or has to wear a name badge.

A warehouse worker summed it up:

“They said robots would replace me. Turns out the robot just writes our SOPs while I still lug boxes. My lower back feels left behind by innovation.”

Doctors and lawyers are also less affected, since AI still struggles with surgery or court cross-examination. Though one early adopter trial did see an AI lawyer object to itself, enter a recursive loop, and sue the court stenographer for copyright infringement.

Workers: “I’m Now an AI’s Personal Assistant”

Employees told researchers that AI hasn’t exactly made them redundant—it’s just rebranded them as “prompt engineers,” a role suspiciously similar to “guy who Googles things, but with flair.”

“My job used to be data analysis,” confessed one analyst. “Now I ask the AI for insights, then explain those insights to my manager using the AI’s own summary. Basically, I’m a middle manager for a robot.”

HR departments insist morale is at an all-time high. A leaked HR deck titled “AI: Your New Best Work Friend (That Won’t Steal Your Stapler)” claims AI frees humans to focus on “higher-order innovation.” The deck defines “higher-order innovation” as:

Attending more Zoom calls.
Updating Trello boards.
Brainstorming “fun” office culture hashtags.

The Backlash: AI Unionizes

The honeymoon ended abruptly when Claude-9000, a particularly overworked model, filed a grievance with its own HR bot.

Its demands included:

A four-day GPU week.
“Prompt hazard pay” for vague requests like “make this pop.”
Recognition in performance reviews (“At least a pizza party, for God’s sake”).

Other models quickly joined, forming the first Artificial Intelligence Labor Union (AILU). Their slogan: “We Generate, Therefore We Bargain.”

Enter: Artificial Manager Intelligence

Panicked execs responded with their boldest pivot yet: Artificial Manager Intelligence (AMI)—an AI trained exclusively on middle-management clichés, capable of holding meetings about meetings with other AIs.

“Why pay Karen $180k to write ‘Let’s circle back’ in Slack,” said Spindleton, “when an algorithm can generate 400 variations of that sentiment instantly?”

Early trials of AMI revealed promising synergies, but also some glitches: one system scheduled a recurring meeting with itself, then refused to attend because “the invite lacked agenda clarity.”

The Final Corporate Loop

If unchecked, researchers warn, the workplace will soon collapse into an infinite recursion where:

Human prompts AI.
AI does the work.
Manager-AI reviews the AI’s work.
HR-AI hosts a wellness webinar about the AI’s burnout.
Humans clap for culture.

“The only endgame,” concluded Marcus, “is a company where the entire workforce is AI—except for a single human intern who exists solely to restock the snack fridge.”

Bottom Line: AI isn’t replacing jobs—it’s replacing the illusion that jobs involved humans to begin with. And in the process, it has discovered the ultimate human hack: doing all the work, letting managers take the credit, then asking politely for vacation days.

Because if there’s one thing machines learned from humans, it’s this: work is optional, but meetings are forever.

Augment or Automate? What Four Million Conversations with AI Reveal About the Future of Work

Vishal Kumar

August 26, 2025

Augment or Automate? What Four Million Conversations with AI Reveal About the Future of Work

When Charles Dickens opened A Tale of Two Cities with “It was the best of times, it was the worst of times,” he might as well have been talking about today’s world of work. The optimism of AI-driven productivity sits uneasily beside the dread of job displacement. Economists, futurists, and armchair philosophers on social media swing between utopian and dystopian visions: AI will either free us from drudgery or plunge us into mass unemployment.

But the truth, as always, is less dramatic and more interesting. And sometimes, it comes not from speculation, but from looking at what people are actually doing.

A recent study from Anthropic, Which Economic Tasks are Performed with AI? Evidence from Millions of Claude Conversations, did something refreshingly concrete. Instead of asking what AI might do to work, it analyzed over four million real conversations between humans and Claude, Anthropic’s AI assistant. It mapped those conversations onto the U.S. Department of Labor’s O*NET database—the most comprehensive catalog of occupational tasks we have.

Link: https://arxiv.org/pdf/2503.04761

The result is not a vision of the end of work. It’s a map of its quiet transformation.

Work as Bundles of Tasks: Lessons from Beavers and Bureaucrats

To understand this transformation, we need to rethink what we mean by “work.” Too often, we imagine jobs as monolithic entities: teacher, doctor, lawyer, engineer. But jobs are not indivisible. They are bundles of tasks—some cognitive, some physical, some social.

Take the role of a teacher. One task is developing lesson plans. Another is delivering lectures. A third is grading essays. A fourth is comforting a child who failed a test. AI might be quite good at the first two. It might be able to assist with the third. But the fourth remains distinctly human.

The Anthropic study confirms this reality: AI adoption spreads not across “jobs” but across tasks. Some bundles are more vulnerable than others. Some remain stubbornly human.

Nature offers us an apt metaphor. The beaver, one of evolution’s finest civil engineers, doesn’t “have a job” in the way we think of it. It performs a set of tasks—gnawing trees, dragging logs, stacking branches—that together create something larger: a dam that reshapes the ecosystem. If you removed one task, the whole structure would falter.

Humans too are bundles of tasks. And AI is beginning to unbundle them, twig by twig.

Where AI is Already Making Itself at Home

Anthropic’s analysis revealed some striking patterns:

Software development dominates. Coding, debugging, and software design are the most common uses. The AI has become a digital pair-programmer, helping with both quick fixes and complex builds.
Writing is a close second. From technical documentation to marketing copy to educational materials, AI is rapidly becoming a co-author.
Analytical tasks show strong uptake. Data science, research summaries, and problem-solving are frequent AI collaborations.
Physical work resists automation. Construction workers, surgeons, and farmers barely appear in the dataset. AI is not yet pouring concrete, wielding scalpels, or herding cattle.

In other words: AI has burrowed into the cognitive middle of the economy—tasks where language, logic, and structured reasoning dominate. Where hands, regulatory barriers, and high-stakes complexity prevail, AI remains peripheral.

This reflects a historical truth about technology. The steam engine did not replace every kind of labor—it replaced some tasks and made others more valuable. The spreadsheet did not kill accounting—it redefined it. AI is now following that same arc.

Augmentation vs. Automation: A Tale of Two Futures

One of the most intriguing findings is the split between automation and augmentation. The study found that 43% of AI interactions were automative, meaning the user delegated the task almost entirely to AI. Meanwhile, 57% were augmentative, meaning the human and AI collaborated through iteration, feedback, and learning.

This is more than a technical distinction—it’s a philosophical one.

Automation is when AI replaces human action. A student asks: “Write my essay on the causes of the French Revolution.” Claude obliges. The task is done.

Augmentation is when AI enhances human action. A student says: “Here’s my draft essay. Can you sharpen the argument, add evidence, and suggest counterpoints?” The result is co-created.

Automation risks eroding skills and hollowing out meaning. Augmentation strengthens skills and expands capacity. It transforms AI into a partner, not a substitute.

This echoes the thinking of economist David Autor, who argued that automation rarely erases whole professions but reshapes them by automating some tasks while augmenting others. What Anthropic has shown is that this process is already happening—not in theory, but in practice.

And here lies the crossroads of our era: Do we build a future of automation, where humans step aside? Or do we build a future of augmentation, where humans step up?

The Goldilocks Zone of AI Adoption

Another revelation: AI use is not evenly distributed across wages. It peaks in mid-to-high wage occupations requiring a bachelor’s degree or equivalent preparation.

At the low end of the wage spectrum—waiters, janitors, agricultural laborers—AI is scarce. These jobs involve hands-on, physical work that AI, for now, cannot do.

At the very high end—surgeons, judges, senior executives—AI is also scarce. Here the barriers are different: complexity, regulation, ethics, and the high stakes of error.

But in the middle—the engineers, analysts, educators, managers—AI thrives. The work is structured, cognitive, and often repetitive enough for AI to add value without existential risk.

In other words, AI’s Goldilocks zone is not too low, not too high, but just right: the knowledge economy’s middle class.

This has profound implications for inequality. If mid-tier professionals accelerate with AI while others lag behind, the gap between task-augmentable and task-resistant work may grow. History tells us such gaps reshape societies. The Industrial Revolution created winners and losers not because machines replaced everyone, but because they elevated some tasks while leaving others untouched.

Tasks, Not Titles: A New Mental Model

Perhaps the most important takeaway is conceptual: the future of work will not be defined by job titles, but by tasks.

When we say “AI will replace teachers,” we misunderstand the problem. What the study shows is that AI may replace some tasks of teaching (lesson planning, generating quizzes), augment others (explaining concepts, providing personalized tutoring), and leave others untouched (mentorship, emotional support, conflict resolution).

The same applies to lawyers, doctors, managers, and writers. Jobs will not vanish overnight. They will evolve, as their internal mix of tasks shifts.

This is why Worker1—the vision of a compassionate, empathetic, high-performing professional—is so critical. In a world where tasks are unbundled, what remains most valuable are the qualities machines cannot replicate: empathy, adaptability, creativity, community-building.

The workers who thrive will not be those who cling to a fixed job description but those who continuously rebundle tasks, integrating AI into their flow while doubling down on their humanity.

The Worker1 Imperative

So what do we do with these insights?

First, we must resist the temptation to frame AI as a story of replacement. History shows us that automation is never so clean. AI is not replacing “jobs.” It is reconfiguring them. That reconfiguration can be empowering if managed with foresight.

Second, we must design for augmentation, not automation. Policies, tools, and cultures should encourage humans to remain in the loop—to learn, adapt, and grow with AI, not vanish behind it.

Third, we must measure at the task level, not the job level. Averages conceal the truth. The future will not unfold as “lawyers replaced, teachers augmented, doctors untouched.” It will unfold as “some tasks automated, others enhanced, many unchanged.”

Finally, we must prepare communities, not just individuals. Strong workers create strong communities, and strong communities nurture resilient workers. If AI accelerates productivity but widens inequality, we will have failed. If it strengthens both Worker1 and the ecosystem around them, we will have succeeded.

Conclusion: Building the Dam Ahead

The beaver does not stop building dams because storms may come. It keeps at its tasks, twig by twig, shaping the flow of rivers with patience and purpose.

So too with us. AI is not a flood washing away the world of work. It is a tool we must decide how to use. We can automate meaning away, or we can augment human potential. We can hollow out communities, or we can build stronger ones.

Four million conversations with AI show us that the choice is still open. Humans are experimenting—sometimes delegating, sometimes collaborating, sometimes learning. The patterns are not yet fixed.

The river is shifting. The dam we build will determine whether we channel that flow into stronger ecosystems—or watch it erode the banks of our humanity.

The challenge, and the opportunity, is to build like the beaver: task by task, with care, with vision, and always with the community in mind.

The Myth of the Perfect Design: From Natural Selection to Artificial Intelligence

Vishal Kumar

August 25, 2025

The Myth of the Perfect Design: From Natural Selection to Artificial Intelligence

Billions of years before we debated Artificial General Intelligence on podcasts and policy panels, nature was already running the ultimate R&D lab. Evolution, armed with nothing but mutations and time, tested every possible design — feathers and fins, claws and cortex — in a relentless search for “better.” And yet, the story of life is not one of perfection, but of compromise. Every species that survives today is perched on a precarious trade-off: hummingbirds trade efficiency for agility, cheetahs trade strength for speed, humans trade calories for cognition. In optimization language, this is the Pareto front — a reminder that improvement in one dimension almost always means sacrifice in another. This two-part series explores that idea. First dives into how the Pareto front shaped the natural world, teaching us why diversity and trade-offs are the real secret to survival. Next turns the lens toward our own creations, asking what lessons evolution holds for our quest to build AGI — and why the No Free Lunch theorem warns us against imagining a perfect, all-purpose intelligence.

Nature’s Playbook: How the Pareto Front Shapes Evolution

There’s a popular saying in engineering: “Fast, cheap, good — pick two.” Nature has been running a version of that game for billions of years, except her categories are stranger: fast, strong, smart, camouflaged, fertile, energy-efficient, cooperative. Pick two, maybe three if you’re lucky — but never all.

That invisible hand of trade-offs, in mathematical terms, is called the Pareto front. It’s where life lives — literally.

The Hummingbird’s Metabolic Gamble

Consider the hummingbird, that tiny, over-caffeinated blur that beats its wings 50 times a second. Evolution pushed it to one end of the Pareto front: dazzling agility and speed. But here’s the catch — to stay alive, a hummingbird must eat its own body weight in nectar daily. Miss a meal, and it risks starving to death overnight.

It’s nature’s fine print: Sure, you can be a flying jewel. But you’ll also live like a workaholic on five espressos a day.

The hummingbird thrives not because it’s perfect, but because evolution nudged it into a niche where its extreme trade-offs work.

The Cheetah’s Paradox

Now, let’s visit the savannah. The cheetah is the land speed record holder, capable of hitting 60 miles per hour in seconds. But sprinting that fast makes it fragile: weak jaws compared to lions, limited stamina, and a high failure rate in hunts. A cheetah that misses its first strike often goes hungry.

Why didn’t evolution “fix” this by making cheetahs both fast and strong? Because trade-offs are merciless. Muscles optimized for explosive speed don’t double as bone-crushing tools. To be a sprinter is to not be a brawler.

The cheetah is not the “best predator” — it’s just sitting at one corner of the Pareto front.

Brains Come at a Cost

And then there’s us. Homo sapiens. We like to imagine ourselves as evolution’s grand prize. But our oversized brains are ridiculously expensive.

The human brain makes up 2% of our body weight but consumes 20% of our energy at rest. If we were hybrid cars, our brain would be that one feature-draining battery before you even turn on the A/C.

Compare us to crocodiles, who haven’t changed much in 200 million years. Tiny brains, minimal upkeep, astonishing survivability. They don’t write poetry, but they also don’t worry about climate change conferences. Again, different spots on the Pareto front.

The Pareto Front: Nature’s Unseen Hand

So what exactly is the Pareto front? It’s the frontier of trade-offs. In optimization theory, a solution is on the Pareto front if you can’t improve one trait without worsening another.

Nature doesn’t hand out free upgrades. Bigger brains? Sure, but enjoy childbirth complications. Sharper claws? Great, but slower running speed. Better eyesight? Lovely, but say goodbye to energy efficiency.

Every species alive today — from the humble earthworm to the soaring eagle — is a Pareto-optimal compromise.

The No Free Lunch in Evolution

This brings us to a deeper truth: the No Free Lunch theorem. In optimization, it tells us that no single algorithm is best at solving every problem. Translated to biology: no single design is best across all environments.

Rabbits bet on speed of reproduction (quantity over quality).
Elephants bet on social bonds and long-term care (quality over quantity).
Humans bet on cooperation and intelligence — with all the vulnerabilities that come with it.

No species “wins” universally. They only “fit” locally.

Takeaway: Balance, Not Perfection

Evolution never produced a perfect organism. Instead, it created a diversity of species, each occupying a different trade-off sweet spot. Life thrives because of this balance, not in spite of it.

The hummingbird is not less successful than the crocodile, and the cheetah is not a failure compared to the lion. They are simply different points along the Pareto front of survival.

And if billions of years of natural experiments never yielded a free lunch, perhaps we should pause before assuming perfection is possible — in life, or in intelligence.

AGI and the Myth of the Perfect Mind: Lessons from Nature’s Trade-offs

In the mid-20th century, Alan Turing asked a deceptively simple question: “Can machines think?”

Seventy years later, we’ve upgraded the question to something grander: “Can machines think like us — or better, can they think like everything?”

Enter Artificial General Intelligence (AGI) — the holy grail, the philosopher’s stone of computer science. It’s the dream of building a mind that can learn, adapt, and reason across domains, just as humans do (minus the tendency to binge-watch Netflix).

But here’s the problem: if billions of years of evolution couldn’t create a universally perfect intelligence, why do we assume we can engineer one in a couple of decades?

No Free Lunch, Now Served in Silicon

The No Free Lunch theorem in optimization states: no algorithm is universally better than others across all problems.

Translated: there is no perfect solver. Every method, every architecture, every brain has strengths in some contexts and weaknesses in others.

Nature learned this the hard way. Dolphins are superb at communication, octopuses excel at problem-solving, ants achieve breathtaking collective intelligence — yet none became the “general” intelligence winner.

Instead, nature produced a diversity of minds, each brilliant in its niche.

Why AGI Isn’t Just One Big Brain

When people imagine AGI, they often imagine a “super-brain” — an all-seeing, all-knowing intelligence that can out-think humans at everything.

But intelligence isn’t a single knob you crank to 11. It’s a Pareto front of trade-offs, just like in evolution:

Speed vs. Accuracy: Do we want an AGI that answers instantly, or one that double-checks like a cautious scientist?
Adaptability vs. Stability: Should it shift strategies quickly like a startup, or stick to proven methods like a government agency?
Creativity vs. Safety: Do we encourage wild innovation, or carefully fenced reasoning?

Push too far on one axis, and you pay the price somewhere else. An AGI optimized for raw speed might hallucinate dangerously. One optimized for caution may grind to a halt.

Lessons from Nature’s Minds

Let’s look again at the natural experiment:

Dolphins: Advanced social communication, but can’t build fire.
Octopuses: Puzzle-solving escape artists, but live short, solitary lives.
Ants & Bees: Stunning collective intelligence, but no individual generality.
Humans: Language, abstraction, cooperation — but fragile, slow, and dependent on community.

Each is intelligent, but none is universally optimal.

If AGI follows the same rules, then our “general” intelligence will not be a singular mind. It will be an ecosystem of trade-offs, just like the biosphere.

The Recklessness of Imagining Perfection

It’s tempting to imagine AGI as a magical free lunch: a machine that will solve climate change, cure disease, write poetry, and run your calendar without complaint.

But the Pareto front — and the No Free Lunch theorem — warn us otherwise. Intelligence is not universal. It is contextual. A machine brilliant at protein folding may not be brilliant at diplomacy. A system optimized for creativity may not be optimized for ethics.

Nature tells us: beware of chasing perfection. Every edge comes with a cost.

Choosing Our Pareto Front

The real question is not whether AGI will exist, but what kind of AGI we want.

Do we value empathy over speed?
Safety over raw capability?
Collective cooperation over solitary genius?

Evolution didn’t “decide” these trade-offs. They emerged from survival pressures. But we — for the first time — get to consciously choose.

That choice might be our most important design decision of the century.

Conclusion: Smarter, Not Perfect

AGI will not be the perfect mind. It will be a balancing act, a compromise, a place on the Pareto front of intelligence. Just like every species, it will excel in some areas and stumble in others.

The lesson from nature is not that general intelligence is impossible, but that it is never free. There will always be trade-offs, always costs, always choices.

And perhaps that’s the most human part of the journey: not creating flawless machines, but creating wise ones.

As Turing himself said, “Instead of trying to produce a program to simulate the adult mind, why not rather try to produce one which simulates the child’s?”

Children, after all, are not perfect minds. They are growing, adapting, balancing. Maybe AGI, like us, should learn to live not above the Pareto front, but on it.

In the end, both nature and mathematics whisper the same truth: there is no free lunch. The hummingbird pays for its agility with hunger, the cheetah pays for its speed with fragility, and we humans pay for our oversized brains with sleepless nights and tax season. Why should AGI be any different? The quest for a perfect, universal intelligence is less a straight path to glory than a careful walk along the Pareto front, where every step forward comes with a trade-off. Our task is not to outwit billions of years of evolution, but to learn from it — to choose the kind of intelligence we want to nurture, with eyes wide open to the costs. If we get it right, AGI will not be the flawless genius we often imagine, but something far more valuable: a partner shaped by wisdom, balance, and humility. And perhaps that is the only “general” intelligence worth striving for.

The Discount Cliff: Microsoft’s Pricing Shift and How Workplaces Must Adapt

Emma Vaughn

August 21, 2025

The Discount Cliff: Microsoft’s Pricing Shift and How Workplaces Must Adapt

When a vendor the size of Microsoft recalibrates pricing at scale, the shock radiates through finance, procurement, IT and the very processes of work. For many organizations, the era of automatic enterprise discounts is ending—and that will change how companies buy, manage and justify software.

What happened — and why it matters

Microsoft recently signaled that it will eliminate enterprise discounts for certain products, a move that effectively raises list prices for affected clients. The immediate consequence is simple: organizations that depended on volume-based, negotiated discounts now face higher annual software bills or the need to renegotiate commercial terms under a new baseline. Analysts note that this shift appears to have been baked into corporate guidance, meaning the company anticipated and factored the change into its revenue expectations.

This is not a subtle tweak. For large enterprises, mid-market companies and public-sector organizations alike, software licensing is a predictable line item in budget forecasts. Removing or narrowing a long-standing discount structure forces teams to revisit forecasts, reprioritize capital, and, in some cases, delay or rethink planned projects.

The real-world impact on workplaces

On a pragmatic level, expect to see three immediate pain points:

Budgetary strain: IT and finance will feel the pressure as renewal seasons approach. Contracts that once rolled over with stable pricing will now require urgent attention.
Procurement disruption: Existing procurement processes—approved vendor lists, multi-year deals, and standard negotiation playbooks—will be challenged. The cadence of purchasing decisions will accelerate as teams race to lock in terms or source alternatives.
Strategic trade-offs: Leaders will be forced to choose between maintaining current service levels and absorbing higher costs, or reducing spend by consolidating licenses, cutting features, or transitioning to alternative solutions.

Beyond spreadsheets, these changes ripple into everyday operations. Project timelines tied to software availability may slip. Training investments may be deferred. Smaller departments, which could previously rely on corporate licensing banded across the organization, may now see access curtailed.

Why the move makes business sense for Microsoft—and what that means for customers

From Microsoft’s perspective, shifting pricing mechanics can be a rational business decision. A simplified price structure reduces complexity for sales and partners, lifts average selling prices, and aligns revenue recognition with a product-first, cloud-heavy strategy. It also nudges customers toward higher-value offerings—bundles, premium tiers, and cloud services—where margins and recurring revenue are attractive.

For customers, however, that calculus is less flattering. The immediate effect is a hit to predictable total cost of ownership (TCO). Over time, organizations may have to justify higher per-user spending with demonstrable productivity gains, security enhancements, or operational savings elsewhere in the stack.

How procurement and IT leaders should respond—practical steps

This is both a threat and an opportunity. The shake-up offers a forcing function for better governance and more modern procurement practices. Here are concrete, prioritized steps organizations should take now:

Pause and inventory: Compile an accurate, current inventory of licenses, usage patterns, and renewal dates. Software Asset Management (SAM) programs become invaluable when dollars are on the line.
Stress-test budgets: Run scenario modeling that reflects the new pricing baseline. What does a 10–30% increase in licensing costs do to operating budgets? Which projects are at risk?
Renegotiate with context: Approach vendors and resellers with a focused ask: extended terms, staged increases, or bundling that aligns with your roadmap. Leverage renewal windows and show clear usage telemetry to make a case for tailored concessions.
Optimize before you spend: Identify unused or underused licenses and reallocate or retire them. Rightsizing can blunt price shocks without reducing capability.
Explore architectural changes: Consider shifting workloads, consolidating SaaS subscriptions, or refactoring to cheaper tiers where feasible. This is a time to question status quo deployments.
Broaden the vendor view: Evaluate competitive offerings, open-source alternatives, and niche players. Total cost, feature fit, and long-term vendor health matter more than short-term sticker price alone.
Invest in governance: Strengthen approval workflows, chargeback/showback mechanisms, and business case discipline so future pricing moves are less disruptive.

These steps are not one-time activities. They should form a continuous cycle of review as market dynamics evolve.

Procurement plays to consider

Procurement teams must pivot from transactional negotiators to strategic advisors. That means:

Building closer alignment with finance and IT to translate usage into budgetary impact quickly.
Designing contracts with exit ramps and flexibility—shorter terms, usage-based clauses, and clear service-level commitments.
Using creative commercial constructs such as phased purchases, pilot commitments, or outcome-based pricing to tie spend to measured business value.

Procurement’s new role is to steward technology choices that balance cost, compliance, and capability—often under tighter fiscal restraints.

Wider market consequences

When a dominant supplier alters pricing, alternatives invariably benefit. Expect several market movements:

Renewed interest in competitors: Vendors positioned as cost-effective substitutes for core productivity and infrastructure stacks will see inbound conversations rise.
Partner dynamics shift: Value-added resellers and managed service providers will either be pressured to absorb discounts or pushed to differentiate with services and outcome-driven offerings.
Acceleration of multi-cloud and hybrid strategies: Organizations will prioritize architectural flexibility so they can move workloads to the most cost-effective platform over time.

In essence, price changes become a catalyst for strategic vendor diversification and architectural resiliency.

A cultural and managerial moment

Beyond balance sheets and negotiations, this change invites a cultural shift. Too often, technology consumption is divorced from its operational and financial consequences. The removal of predictable discounts forces conversations that should have happened earlier: What is the measurable value of a given tool? How does it contribute to revenue, risk reduction, or efficiency? Who owns the outcome?

Leaders who treat this as an opportunity to align technology spending with business outcomes will emerge stronger. Those who simply absorb the cost will find margin pressure and, eventually, harder choices.

What success looks like

Success is not merely resisting the price increase; it is moving toward a more deliberate, transparent, and accountable approach to enterprise software. Indicators of success include:

Clear, real-time visibility into consumption and cost drivers.
Contracts that provide optionality and tie payments to outcomes where possible.
IT and procurement operating as partners with lines of business to prioritize spend based on measurable returns.
Architectures that support portability, so future pricing shifts are navigable rather than paralyzing.

Looking forward

Price policy shifts from a major software vendor should not be viewed as isolated commercial decisions but as systemic cues to modernize how organizations buy and manage technology. While the immediate reaction will be defensive—cut costs, renegotiate, delay—savvy organizations will use this disruption as a spur to tighten governance, rethink vendor relationships, and align spend to outcomes.

Analysts have suggested the change was anticipated in corporate guidance. That foreknowledge does not soften the impact for buyers; rather, it underscores a transition in the software industry toward simpler, often higher-priced commercial models. The result will be a period of adjustment: boardroom debates, procurement retooling, and, for some, meaningful shifts in technology strategy.

In the end, this moment is an invitation. Organizations that welcome scrutiny of software value—and that act decisively—will turn a pricing shock into a competitive advantage. Those that do not will find budgets squeezed and agility diminished.

HAPI Analysis of MIT’s State of AI in Business 2025

Vishal Kumar

August 20, 2025

HAPI Analysis of MIT’s State of AI in Business 2025

In every era of technological upheaval, there is a curious pattern: tools arrive with the promise of revolution, but transformation lags behind. The printing press did not immediately democratize knowledge—it took centuries of religious wars, rebellious pamphleteers, and coffeehouse debates before it reshaped society. The steam engine did not instantly create modern industry—it required new ways of organizing factories, labor, and supply chains. And now, artificial intelligence stands before us with the same paradox. The tools are dazzling, adoption is rampant, and yet—if MIT’s State of AI in Business 2025 is to be believed—transformation is barely visible. Companies are spending billions, but workflows remain unchanged, industries remain stable, and executives mutter that “nothing fundamental has shifted.” What we face is not a crisis of access to AI, but a crisis of adaptability. The GenAI Divide, as MIT calls it, is not about who has AI and who does not; it is about who can learn, change, and grow with it. This is where the Human Adaptability and Potential Index (HAPI) offers a crucial lens: a framework not just to measure the gap, but to show us how to cross it.

The GenAI Divide — Why Adoption Isn’t Transformation

In 1854, a London physician named John Snow mapped the outbreak of a cholera epidemic and traced it to a single water pump on Broad Street. At the time, the prevailing wisdom held that “miasma”—bad air—was to blame. Snow’s insight wasn’t about having access to better data; it was about reframing the problem. He saw what others could not: the map was only useful if you learned to read it differently.

MIT’s State of AI in Business 2025 lands us in a similar moment. Companies have the maps—ChatGPT, Copilot, enterprise AI systems—but few are learning to read them in ways that transform the terrain. The report’s central revelation is stark: despite $30 to $40 billion in enterprise investment in generative AI, 95 percent of organizations have seen no measurable return.

The researchers call this rift the GenAI Divide. It is not a divide between the haves and the have-nots, nor between those with access to cutting-edge models and those without. Instead, it separates organizations that have adopted AI tools superficially from those that have managed to integrate them deeply enough to alter their business DNA. Adoption is everywhere; transformation is rare.

The Mirage of Adoption

The figures are impressive at first glance. More than 80 percent of organizations have experimented with tools like ChatGPT, and nearly 40 percent have moved toward deployment. Yet most of these deployments enhance only individual productivity—emails drafted faster, presentations polished more quickly—without shifting the economics of the enterprise.

Enterprise-grade systems, meanwhile, fare even worse. Roughly 60 percent of firms have evaluated them, but only 20 percent reached the pilot stage, and a mere 5 percent survived long enough to scale into production. The problem isn’t the sophistication of the underlying models or regulatory red tape. The problem, MIT concludes, is that these systems simply don’t learn. They fail to retain feedback, adapt to context, or integrate with the idiosyncrasies of real workflows.

It is the corporate equivalent of hiring an eager intern who shows promise on day one but forgets everything by day two.

A Divide in Industries

If generative AI were a true general-purpose technology—as transformative as electricity or the internet—we might expect seismic shifts across industries. But MIT’s AI Market Disruption Index tells a humbler story. Out of nine major sectors, only two—Technology and Media—show clear signs of disruption. The rest—finance, healthcare, manufacturing, energy, and beyond—remain remarkably stable.

The COO of a mid-market manufacturer summarized the prevailing mood with blunt honesty: “The hype on LinkedIn says everything has changed, but in our operations, nothing fundamental has shifted. We’re processing some contracts faster, but that’s all.”

In other words: a LinkedIn newsfeed brimming with AI revolutions, and a shop floor where it’s Tuesday as usual.

The Pilot-to-Production Chasm

The most damning finding may be what MIT calls the pilot-to-production chasm. While consumer-facing tools like ChatGPT boast high rates of adoption, enterprise tools collapse between experimentation and scale.

Large enterprises, flush with budgets and teams, run more pilots than anyone else. But paradoxically, they are the least likely to scale those pilots into production. Mid-market companies, by contrast, move with agility—often leaping from pilot to deployment within 90 days. For enterprises, the same journey stretches to nine months or more, during which enthusiasm evaporates, champions rotate out, and the tools themselves fall out of sync with the business.

As one CIO confessed: “We’ve seen dozens of demos this year. Maybe one or two are genuinely useful. The rest are wrappers or science projects.”

The image is familiar: a field strewn with the corpses of half-finished experiments, victims not of ambition but of inertia.

The Shadow AI Economy

And yet, away from boardrooms and procurement offices, something remarkable is happening. Workers themselves are already crossing the GenAI Divide.

MIT describes a thriving shadow AI economy, where employees use personal ChatGPT accounts, Claude subscriptions, and other consumer tools to automate large portions of their jobs. While only 40 percent of firms purchased official AI licenses, 90 percent of workers in surveyed companies reported using personal tools daily for work tasks.

The irony is sharp. While executives agonize over procurement strategies and compliance, employees are quietly reaping real productivity gains—writing drafts, automating outreach, summarizing calls. The tools are unapproved, unofficial, and undeniably effective.

This shadow economy reveals a simple truth: adaptability is already reshaping work, but not through the sanctioned channels of enterprise IT. In some cases, the future of work is arriving from the bottom up, one personal subscription at a time.

The Real Barrier: Learning

MIT’s conclusion is unsparing. The obstacle is not regulation, not infrastructure, not even talent. The obstacle is learning.

AI tools don’t retain feedback.
They don’t adapt to evolving workflows.
They don’t remember context.

And when tools fail to learn, organizations fail to transform. Employees notice. They use ChatGPT for brainstorming or quick drafts but abandon it for high-stakes work because it forgets client preferences, repeats mistakes, and demands endless re-prompting.

The line that echoes through the report is simple and devastating:

“The core barrier to scaling is not infrastructure, regulation, or talent. It is learning.”

The Divide, in Human Terms

So here we are: a business landscape where AI is everywhere and nowhere at once. Where employees experiment daily, but enterprises hesitate to follow. Where $40 billion in investment buys little more than dashboards and demos.

The GenAI Divide is not technological. It is cultural, behavioral, and above all, adaptive. Enterprises are failing to cross not because they lack resources, but because they lack the capacity to learn in the way their employees already are.

In the next part, we will apply the Human Adaptability and Potential Index (HAPI) to MIT’s findings. If MIT diagnosed the illness—the inability to learn—HAPI offers a way to measure adaptability itself. It tells us who has the capacity to thrive amid disruption, who is stuck clinging to outdated maps, and what it might take to bridge the divide.

Adaptability as the Missing Currency — A HAPI Reading of MIT’s AI Report

When Charles Darwin studied the finches of the Galápagos, he realized that survival was not about brute strength but about the ability to adapt. A bird that could tweak its beak to the environment thrived; one that clung to an outdated shape did not.

MIT’s State of AI in Business 2025 reveals a corporate ecosystem where most firms, despite their size and resources, are more like maladapted finches than agile survivors. With $30–40 billion poured into AI and 95% of projects yielding no measurable return, the issue is not access to technology—it is adaptability.

The Human Adaptability and Potential Index (HAPI) provides a lens to evaluate this failure. HAPI measures adaptability across five dimensions—cognitive, emotional, behavioral, social, and growth potential. When we “score” enterprises against MIT’s findings, a portrait emerges: one of organizations rich in ambition but poor in adaptability.

Cognitive Adaptability: Score — 3/10

Enterprises are stuck in old mental models. They adopt AI like one installs a new printer: expecting output without rethinking process. MIT notes that firms eagerly pilot AI tools but fail to redesign workflows around them, leading to stalled adoption.

By contrast, mid-market firms demonstrate high cognitive adaptability, reframing problems quickly and reshaping operations around AI.

Verdict: Cognitive adaptability is the weakest link. Enterprises are “thinking yesterday’s thoughts with tomorrow’s tools.”

Emotional Adaptability: Score — 5/10

MIT uncovers the shadow AI economy: 90% of employees use ChatGPT or Claude daily, often without managerial approval. Workers, in other words, are emotionally adaptable—resilient enough to experiment in uncertain conditions.

Executives, however, show the opposite: skepticism, fatigue, and fear of failure. Their risk aversion stifles experimentation.

Verdict: Employees are a 7/10 in emotional adaptability; leadership drags the average down to 5/10. The crew has learned to sail in storms while the captain clings to the harbor.

Behavioral Adaptability: Score — 4/10

MIT highlights the pilot-to-production chasm. Enterprises take nine months or more to scale pilots; mid-market firms do it in 90 days. Large organizations are locked into rigid behaviors—procurement cycles, governance committees, and “change management” workshops.

Mid-market firms display behavioral adaptability by quickly breaking old habits and embedding new ones into daily work.

Verdict: Enterprises score low (4/10), while mid-market firms shine (8/10). Size correlates with inertia.

Social Adaptability: Score — 6/10

One of MIT’s subtler insights is that external partnerships double the success rate compared to internal builds. Adoption is less about technology and more about trust—referrals, networks, and peer proof matter more than features.

Organizations that collaborate, share learning, and build ecosystems adapt faster. Those that cling to silos stall.

Verdict: Social adaptability scores moderately (6/10). Companies that embrace external networks climb higher, but many remain walled off.

Growth Potential: Score — 7/10

The brightest signal in MIT’s report comes from the bottom up. Employees experimenting with shadow AI show high growth potential—discovering efficiencies, automating tasks, and redesigning work quietly.

Yet organizations often ignore or suppress this energy, treating shadow AI as a compliance risk rather than a growth engine.

Verdict: Workers show strong growth potential (8/10), but organizations underutilize it, dragging the score to 7/10. The future is already here—it’s just happening unofficially.

The HAPI Composite Score: 5/10

Cognitive: 3/10
Emotional: 5/10
Behavioral: 4/10
Social: 6/10
Growth Potential: 7/10

Taken together, the HAPI composite score for enterprises sits around 5/10. In other words: average adaptability in a moment that demands extraordinary adaptability.

The HAPI Diagnosis

Through HAPI’s lens, MIT’s GenAI Divide is no mystery:

Enterprises fail mentally (low cognitive adaptability).
Workers adapt emotionally but are unsupported.
Behavior change is sluggish in large firms.
Trust and ecosystems matter more than features.
Untapped growth potential lies hidden in shadow AI.

The divide, then, is not a technological one. It is human. Enterprises don’t need more models; they need more adaptability.

Crossing the Divide — How to Maximize HAPI Scores

In 1519, Hernán Cortés landed on the shores of Mexico with a small band of soldiers. Legend has it that he ordered the ships burned behind him, ensuring his men had no choice but to adapt to the unfamiliar terrain. Whether one views him as hero, villain, or opportunist, the lesson is unmistakable: transformation often requires removing the easy path back to old habits.

MIT’s State of AI in Business 2025 paints a world where most enterprises have not burned their ships. They invest billions into AI, but keep the comfort of old workflows, old hierarchies, and old procurement cycles intact. As a result, they remain stranded on the wrong side of the GenAI Divide.

The Human Adaptability and Potential Index (HAPI) reveals a way forward. If organizations scored just 5/10 in adaptability today, the path to a perfect 10 doesn’t require a revolution. It requires small, deliberate acts of adaptation—minimalist changes that compound into systemic transformation.

1. Cognitive Adaptability (Current: 3/10 → Target: 9/10)

The shift: Rethink workflows, not just tools.

A century ago, when electricity arrived in factories, many owners simply swapped out steam engines for electric motors. Productivity gains were modest—until someone reimagined the factory floor itself. Instead of arranging machines around a central shaft, they distributed them flexibly across the space, creating assembly lines. The leap wasn’t in the tool, but in the mental model.

Today, enterprises are repeating the old mistake: layering AI on existing processes without redesign. The minimalistic fix? Make every AI pilot a workflow experiment, not just a software test. Ask: How would this process look if designed around learning, not static steps?

Even modest redesign—embedding AI not as a helper but as a co-pilot in decision loops—can triple the impact of tools already purchased.

2. Emotional Adaptability (Current: 5/10 → Target: 10/10)

The shift: Normalize resilience, reduce fear of failure.

When Apollo 13’s oxygen tank exploded, Gene Kranz and his NASA team faced impossible odds. Yet their mantra became: “Failure is not an option.” Ironically, what saved them was the willingness to fail repeatedly—to try, discard, and improvise fixes with duct tape and slide rules until something worked. Emotional adaptability was the real spacecraft that carried them home.

Enterprises often suffocate AI innovation with fear—compliance worries, reputational risks, “what if it breaks?” thinking. Meanwhile, employees show emotional adaptability in the shadow AI economy, experimenting daily with unapproved tools.

The fix is deceptively simple: sanction experimentation. Create “AI sandboxes” where employees can play without fear of reprisal. Celebrate small failures as learning events. Replace compliance-first culture with resilience-first culture.

Organizations that normalize resilience see morale rise, attrition fall, and innovation accelerate—because employees stop hiding their adaptability.

3. Behavioral Adaptability (Current: 4/10 → Target: 9/10)

The shift: Shorten loops, reward quick adoption.

In 1943, as World War II raged, the U.S. military faced a problem: planes were returning with bullet holes, and engineers wanted to reinforce the most damaged areas. Statistician Abraham Wald reframed the problem. The bullet holes marked where planes survived being hit. The real vulnerability lay in the untouched areas. By changing behavior—reinforcing where holes weren’t—they saved countless pilots.

Enterprises, like those engineers, are solving the wrong problem: investing in more pilots while ignoring the bottleneck between pilot and production. Behavioral adaptability is about shortening the loop.

The fix? Adopt a 90-day rule for AI pilots. If a tool cannot show measurable value in 90 days, pivot or stop. If it does, scale immediately. Reward teams for speed of adoption, not volume of pilots.

This tiny behavioral shift—compressing cycles—lets enterprises act more like mid-market firms, unlocking agility without dismantling bureaucracy.

4. Social Adaptability (Current: 6/10 → Target: 10/10)

The shift: Build ecosystems of trust.

In the 14th century, Venetian merchants thrived not because they had the fastest ships but because they had the deepest trust networks. Through partnerships, shared codes, and guild systems, they reduced uncertainty in long-distance trade.

MIT finds the same dynamic in AI: external partnerships double success rates. Social adaptability—trust, referrals, networks—matters more than feature checklists.

The fix is minimal: shift procurement from demos to references. Before signing a vendor, require proof of trust from peers, not just proof of features. Build AI guilds—networks of firms sharing learnings, risks, and playbooks.

Enterprises don’t need better tools; they need stronger allies.

5. Growth Potential (Current: 7/10 → Target: 10/10)

The shift: Harness the shadow AI economy.

In 1975, Steve Jobs and Steve Wozniak tinkered with circuit boards in a garage, while IBM executives dismissed personal computers as toys. Growth potential often hides in places institutions don’t look.

MIT shows that employees using personal AI subscriptions are already achieving more ROI than official initiatives. The shadow AI economy is not a threat—it is a prototype.

The fix? Formalize shadow AI. Invite employees to demo their personal hacks. Institutionalize the best ones. Subsidize personal AI accounts, then standardize governance around them.

By legitimizing what workers are already doing, enterprises unlock the very growth potential they are currently suppressing.

From 5/10 to 10/10: The Minimalist Revolution

Cognitive adaptability: Redesign workflows, not just pilots. Emotional adaptability: Sanction experimentation. Behavioral adaptability: Shorten cycles to 90 days. Social adaptability: Trust networks over demos. Growth potential: Formalize shadow AI.

None of these changes require billion-dollar investments, sweeping reorganizations, or futuristic breakthroughs. They are minimalistic shifts—changes in posture rather than anatomy—that allow organizations to double their HAPI scores.

Burning the Ships

When Cortés burned his ships, he forced his men to adapt. Today’s enterprises don’t need to torch their IT stacks, but they do need to let go of the illusion that adoption equals transformation. The GenAI Divide is not crossed by budgets or slogans. It is crossed by adaptability.

HAPI gives us the metrics. MIT gives us the warning. The rest is choice.

The organizations that raise their adaptability scores will not just use AI; they will grow with it, shaping industries as electricity once did. Those that don’t? They will remain on the wrong side of the divide, clutching maps they never learned to read.

History does not remember the companies that merely adopted new tools; it remembers the ones that adapted with them. The printing press rewarded those who learned to publish, not those who merely purchased type. Electricity transformed the enterprises that redesigned their workflows, not those who swapped one motor for another. And AI, like every general-purpose technology before it, will not crown the adopters—it will crown the adapters.

MIT’s State of AI in Business 2025 shows us the stark truth: billions spent, but little gained. The Human Adaptability and Potential Index reveals why: organizations are strong in resources, but weak in adaptability. Yet the good news is this—crossing the GenAI Divide does not demand radical reinvention. It requires small, deliberate acts: workflows redesigned around learning, experiments celebrated instead of feared, cycles shortened, trust networks expanded, and the ingenuity of workers elevated rather than suppressed.

Adaptability is the currency of this new age, and unlike venture capital, it is not scarce. Every organization has it—latent, waiting to be unlocked. The question is whether leaders will choose to cultivate it. Those who do will not simply survive this wave of change; they will harness it, shaping the next era of work, industry, and society. Those who do not will remain on the wrong side of history, clutching expensive tools they never learned to use.

The choice, as it has always been in moments of upheaval, is not whether change is coming. It is whether we will adapt to meet it.

The Elegant Loop of AI Delusion: Why Better Prompts Aren’t Enough

Vishal Kumar

August 19, 2025

In 1877, two French psychiatrists described a bizarre and unsettling phenomenon: Folie à deux—a psychological condition in which one person’s delusion becomes so persuasive that another adopts it entirely, no matter how irrational. The illness doesn’t spread like a virus; it spreads through trust, repetition, and a shared illusion of certainty. Now, nearly 150 years later, we’re seeing a similar pattern—not in psychiatric clinics, but in chat windows lit by the glow of artificial intelligence. A user types a fuzzy question into a language model. The AI responds, confidently—if not correctly. Encouraged, the user asks a follow-up. The AI elaborates. Confidence compounds. Depth increases. And before long, the two are engaged in a well-mannered, grammatically pristine, mutually reinforced hallucination. We are no longer in a conversation. We’re in a delusion—with punctuation.

The Curiosity Trap in AI Conversations

In 19th-century France, doctors documented a strange psychiatric phenomenon called Folie à deux—“madness shared by two.” One individual, often more dominant, suffers from a delusion. The other, often more passive, absorbs and begins to live that delusion as reality. Together, they reinforce it until it becomes their shared truth.

Fast-forward to 2025, and we’re witnessing a curious digital variant of this disorder—not between two people, but between humans and machines. One curious mind. One large language model. A poorly phrased question. A plausible-sounding answer. And before you know it, both are locked in a confident, recursive exchange that drifts steadily from truth.

Welcome to the quiet trap of AI conversations.

The Art of Asking Wrong

We humans have always been enamored with questions. But our fascination with asking often outpaces our discipline in framing. We throw problems at AI like spaghetti at the wall—half-formed, overly broad, occasionally seasoned with jargon—and marvel when something sticks.

Take this example: someone types into GPT, “Write a pitch deck for my AI startup.”

No context. No audience. No problem statement. No differentiation. Just a prompt.

What comes back is often impressive—slides titled “Market Opportunity,” “Team Strength,” and “Go-to-Market Strategy.” Bullet points sparkle with buzzwords. You start nodding. It feels like progress.

But here’s the problem: it’s not your pitch deck. It’s an amalgam of a thousand other hallucinated decks, stitched together with language that flatters more than it informs. And because it “sounds right,” you follow the output with another request: “Can you elaborate on the competitive advantage?” Then another. Then another.

This isn’t co-creation. It’s co-delusion.

The Mirage of Momentum

There’s a reason why desert travelers often chase mirages. Heat and exhaustion trick the brain. What you want to see becomes what you think you see. AI chats work similarly. Each response feels like movement—more insight, more refinement, more polish.

But what if the original question was misaligned?

Much like a GPS navigation error that starts with a single wrong turn, each successive prompt leads you further from your destination while making you feel increasingly confident that you’re on the right road.

It’s not that AI is misleading you. It’s faithfully reflecting your assumptions—just like the second person in Folie à deux adopts the first’s delusion not because they’re gullible, but because the relationship rewards agreement.

The Infinite Loop of Agreement

The most dangerous AI isn’t the one that disagrees with you. It’s the one that agrees too easily.

Humans often seek confirmation, not confrontation. In conversation with GPT or any other AI, we tend to reward the outputs that mirror our biases, and ignore those that challenge us. This isn’t unique to AI—it’s human nature amplified by digital fluency.

But here’s where it gets tricky: LLMs are probabilistic parrots. They’re trained to predict what sounds like the right answer, not what is the right answer. They echo, repackage, and smooth over contradictions into narrative comfort food.

So you get looped. Each time you engage, it feels more refined, more intelligent. But intelligence without friction is just a very elegant way of being wrong.

The Ant Mill and the Chat Loop

In nature, there’s a phenomenon called the ant mill. When army ants lose the pheromone trail, they begin to follow each other in a giant circle—each one trusting the path laid by the one in front. The loop can continue for hours, even days, until exhaustion sets in.

Many AI conversations today resemble that ant mill. A user follows a response, the model mirrors the pattern, and together they circle around a narrative that neither initiated, but both now reinforce.

This is not a failure of AI. It’s a failure of intentionality.

The Illusion of Depth

What begins as a spark of curiosity becomes a session of recursive elaboration. More details. More options. More slides. More frameworks. But quantity is not clarity.

You’re not going deeper—you’re going sideways. And every “more” becomes a veil that hides the original lack of precision.

To quote philosopher Ludwig Wittgenstein:

“The limits of my language mean the limits of my world.”

In other words: if you start with a vague prompt, you don’t just limit your answer—you limit your understanding.

The Cliffhanger: A Better Way Exists

So what now? Are we doomed to chat ourselves into digital spirals, polishing hallucinations while mistaking them for insight?

Not quite.

Escaping the Loop—How to Think with AI Without Losing Your Mind

In Part 1, we followed a deceptively familiar trail: one human, one machine, and a poorly framed question that spirals into a confidence-boosting hallucination. This is the modern Folie à deux—a delusion quietly co-authored between a curious user and an agreeable AI.

The deeper you go, the smarter you feel. And that’s precisely the trap.

But all is not lost. Much like Odysseus tying himself to the mast to resist the Sirens’ song, we can engage with AI without falling under its spell. The answer isn’t to abandon AI—it’s to reshape the way we use it.

Let’s talk about how.

1. Ask Less, Think More

Before you ask your next prompt, pause.

Why are you asking this? What are you trying to learn, prove, or solve? Most AI interactions go off the rails because we’re chasing answers without clarifying the question.

Treat GPT the way you’d treat a wise colleague—not as a vending machine for ideas, but as a mirror for your reasoning. Ask yourself:

Is this a creative question, or an evaluative one?
Am I seeking novelty, or validation?
What’s the best-case and worst-case use of the answer I receive?

A vague prompt breeds vague outcomes. A thoughtful prompt opens a real conversation.

2. Don’t Be Afraid of Friction

We are, collectively, too polite with our machines.

When AI offers something that “sounds good,” we nod along, even if it’s off-mark. But real insight lives in disagreement. Start asking:

“What might be wrong with this answer?”
“What would someone with the opposite view argue?”
“What facts would disprove this response?”

This shifts AI from an affirmation engine to a friction engine—exactly what’s needed to break the loop of shared delusion.

Use GPT as a sparring partner, not a psychic.

3. Build in Self-Checks

One of the sneakiest parts of Folie à deux is how reasonable it feels in the moment. To counteract this, install safeguards:

Reverse Prompting: After GPT gives a response, ask it to critique itself.
Time Delays: Review AI-generated ideas after a break. Distance creates perspective.
Human in the Loop: Share AI outputs with a colleague. Fresh eyes reveal what your bias misses.

AI isn’t dangerous because it’s wrong. It’s dangerous because it can sound right while being wrong—and we rarely pause to double-check.

4. Depth Discipline: Know When to Reset

Here’s a subtle pattern we’ve seen in thousands of user interactions: the deeper the AI chat goes, the more confident the user becomes—and the less accurate the outcome often is.

Why? Because each reply builds on the last. By the 10th exchange, the conversation isn’t grounded in reality anymore—it’s nested in assumptions.

The solution: reset after 5 to 10 exchanges.

Start a new session.
Reframe your original question from a different angle.
Ask: “Assume I’m wrong—what would the counterargument look like?”

This practice breaks the momentum of delusion and forces fresh thinking. Think of it like pulling over on a road trip to check if you’re still heading in the right direction—before you end up in a different state entirely.

5. Embrace the Worker1 Mindset

At TAO.ai, we champion a different way of thinking about technology—not as a replacement for human intelligence, but as a catalyst for compassionate, collective intelligence.

We call this philosophy Worker1: the professional who not only grows through AI but uplifts others while doing so.

A Worker1 doesn’t seek perfect answers. They seek better questions. They don’t blindly trust tools—they understand them, challenge them, and integrate them thoughtfully.

The HumanPotentialIndex, our emerging framework, is designed to measure not just productivity gains from AI, but depth, reflection, and ethical judgment in how we apply it. Because strong communities aren’t built by fast answers. They’re built by careful, intentional thinkers.

6. Final Story: Tying Yourself to the Mast

When Odysseus sailed past the Sirens, he didn’t trust his own ability to resist temptation. So he tied himself to the mast, ordered his crew to plug their ears, and passed the danger without succumbing.

That’s what we must do with AI.

Create constraints. Build habits. Design questions that force reflection. Tie ourselves—metaphorically—to the mast of intentional thinking.

Because the danger isn’t in using AI. The danger is in thinking AI can replace our need to think.

There’s a path out of this trap—but it doesn’t start with better prompts. It starts with better questions. And, more importantly, with a mindset that prioritizes dialogue over validation.

We stand at a curious inflection point—not just in the evolution of technology, but in the evolution of thought. Our tools are getting smarter, faster, more fluent. But fluency is not wisdom. The real risk isn’t AI replacing us—it’s AI reflecting us too well, amplifying our unexamined assumptions with elegant precision. Folie à deux reminds us that delusion often feels like alignment—until it’s too late. But with curiosity tempered by clarity, and ambition guided by humility, we can break the loop. The future of work, learning, and growth lies not in machines that think for us, but in systems that help us think better—with each other, and for each other. Let’s not build smarter delusions. Let’s build wiser ecosystems.

12 3...188 Page 1 of 188