Please stop the AI Confidence Theater

Elena Verna (guest essay, cross-posted by Lenny Rachitsky / Lenny's Newsletter) · email · 2026-07-02 source ↗ Reader 🖥️

AI Confidence Theater Measuring AI Impact Trust

The widespread performance of confident AI mastery — inflating modest, unreliable AI wins into 'life-changing' claims and projecting certainty where genuine inconsistency exists — does more harm than good, and everyone up and down the org chart should replace it with honest anchoring in real, verifiable outcomes.

practical · mixed — A persuasive op-ed / 'desperate plea' (argumentative) built on an expository diagnosis of causes and harms, with narrative anecdotes and analogies. Practical because it argues for action: stop performing, and it prescribes concrete conduct for content creators, execs, managers, and employees.

Structure

Part	Role
Opening framing: 'We don’t need an extra reason to be anxious' — the litany of serial AI doom/hype headlines, the author's own AI-company insider friction, and her 'Cool. Show me.' test that reveals a very short list of genuinely transformative use cases	context / problem statement
'It is screwing a lot of stuff up' — three harms: reality distortion makes people write off AI entirely; a new 'reverse-hustle hustle culture' of status performance; and a broken hiring process where sounding competent has decoupled from being competent	evidence (harms)
'The FOMO is exhausting' — the psychological harm: a 'fake baseline' that makes genuinely useful modest wins feel embarrassingly inadequate and leaves people demoralized	evidence (harm to individuals)
'Why is this happening?' — four causes: (1) attention economy rewards exaggeration, (2) it's the 'wild west' because AI claims are hard to verify, (3) marketing sells certainty AI can't deliver, (4) top-down VC→exec→employee pressure forces performance	causal diagnosis
'Cut it out. You’re only hurting yourself.' — prescriptions: for content (stop spreading hype, hold creators accountable, don't amplify untested claims); for companies (execs set reasonable expectations, managers focus on outcomes, employees find actual impact; 'Learning AI IS the job')	conclusion / prescription
'Let’s just be honest & talk about the hard part' — the real value is in the 'next thousand prompts': AI systems are living systems needing monitoring, evaluation, iteration and tuning as models/prompts/APIs drift; be honest about prototypes vs. finished products	resolution / reframing

Key terms

AI Confidence Theater AI Confidence Theater
The collective, performative projection of confident AI mastery — inflating routine, unreliable, or modest AI use into 'life-changing' claims and broadcasting certainty where genuine inconsistency exists — enacted by content creators, marketers, execs, and employees. It is 'theater' because it is status performance without substance (the same signaling as old hustle-culture props, now with AI props) and 'confidence' theater because it manufactures certainty and expertise that isn't really there.
fake baseline AI Confidence Theater
The illusory norm created when everyone around you appears to have transformed their work with AI, so genuinely time-saving modest uses (e.g. summarizing meetings) suddenly feel embarrassingly basic; you stop appreciating real gains because someone else claims to have done far more.
trust-based growth Trust
The current AI era's growth logic: because over-selling will eventually be discovered, exaggeration cashes in short-term attention at the cost of long-term trust — every inflated promise makes people trust the next AI product a little less.
reverse-hustle hustle culture AI Confidence Theater
The new status performance — bragging about burning tokens, running 17 agents, and not opening email — which is structurally identical to the old hustle-culture performance (5am wakeups, cold plunges, inbox zero, Peloton), just with different props.
the next thousand prompts AI Confidence Theater
Where real AI value is actually created — not the fun first prompt but the ongoing work when the model changes, the prompt stops working, the API behavior shifts, or the integration quietly breaks; AI systems are 'living systems' needing monitoring, evaluation, iteration, and tuning, not 'set it and forget it'.

Propositions

explicitThe set of AI use cases that are genuinely critical or life-changing — such that removing them tomorrow would make someone's work fall apart — is very short, far shorter than the volume of hype implies.conf: high
explicitExaggerated AI hype is net-harmful and 'not a victimless crime': it makes people write off AI entirely, robbing them of a genuine aha moment.conf: high
explicitdevelopedAI has made average intelligence cheap and given everyone the vocabulary of expertise, so 'sounding competent' and 'being competent' have decoupled; verbal interviews no longer work and case studies / work trials become necessary.conf: high
extends ←The Last Technical Interview · p9 — Elena's essay asserts that verbal interviews no longer work and 'case studies / work trials become necessary'; Yegge extends this into a full mechanism (provisional employment, campfire trials, portable employer-signed credentials) for how such trials could actually work and be made fair. That extraction's gap g3 flags work trials as costly, hard to scale, and gameable — Yegge's 'count work twice' partially addresses the cost/scale half (turning trials into a profit center) but never addresses the gameability half, so it only partially closes that gap (not recorded as a formal closure).
explicitAI hype is uniquely corrosive compared to past tech hype because the claims are hard to verify — you cannot tell whether a viral workflow is real (different model? your data/setup? or faked?) the way you instantly could with, say, a Dropbox brag.conf: high
explicitReal AI value lives not in the impressive first prompt but in the unglamorous, unposted maintenance of living systems — monitoring, evaluation, iteration, and tuning as models, prompts, and APIs drift.conf: high
implicitThe genuine but modest productivity gains from AI are real and worth celebrating; the problem is the gap between those gains and the inflated claims, not AI itself.conf: high
implicitestablishedTrust is the scarce, shared resource of the AI era: each exaggeration is a withdrawal from a common trust account that makes the next AI product a little less believable — a tragedy-of-the-commons of credibility.conf: high
supports ←AI Enthusiasts Are in a Race Against Time, AI Skeptics Are in a Race Against Entropy · p3 — Identical trust-account-withdrawal metaphor: each inflated AI claim is a withdrawal from a slow-built, shared trust balance.
supports ←John Cutler's Post · p10 — Cutler's 'social contract as precondition for honest experimentation, else compliance theater' directly echoes Elena's framing of trust as the scarce, withdrawable shared resource of the AI era.
implicitThe behavior is fundamentally status signaling, not about AI: humans will always find props to perform superiority, and AI agents/tokens are merely the current props replacing cold plunges and inbox zero.conf: high
implicitThe operational cost and unreliability of AI systems are systematically hidden from public discourse because maintenance work isn't postable, so the public picture of AI's capability and cost is structurally distorted upward.conf: medium

Arguments

We should collectively stop the AI Confidence Theater and replace it with honest, outcome-anchored practice. strength: medium

Explicit premises:

It distorts reality and makes people write off AI, hurting genuine innovation and adoption.
It creates a fake baseline that demoralizes people and inflates unnecessary FOMO.
It breaks hiring by decoupling sounding-competent from being-competent.
It erodes trust ('trust-based growth'); overselling gets discovered and you pay for it in the end.

Implicit premises:

The collective harms outweigh the private benefits (clicks, career protection) the performers gain — i.e. this is a collective-action / commons problem where individually rational hype is collectively destructive.
Honest signaling is feasible and will still be rewarded ('we will still be excited, we promise'), so nothing essential is lost by dropping the performance.

Inferences:

Because the payoff structure she diagnoses (attention economy, VC pressure, job protection) rewards hype, a moral plea to individuals is unlikely to shift behavior without changing the incentives or adding verification mechanisms — a step the essay gestures at ('show me the receipts') but does not design.
If verification (work trials, receipts, benchmarks) is what cuts through, then the durable fix is building shared verification infrastructure, not exhorting virtue.

AI hype is uniquely corrosive relative to previous technology hype. strength: high

Explicit premises:

With prior tech, an exaggerated claim ('this INSANE Dropbox setup') was instantly falsifiable by inspection, so it could be dismissed.
With AI, a viral claim cannot be confirmed or reproduced — you can't tell if the difference is the model, your data, your setup, or a fake.

Implicit premises:

The informational value of a claim depends on its verifiability; unverifiable-but-sometimes-true claims are maximally anxiety-inducing because you can neither safely dismiss them nor reliably reproduce them.

Inferences:

The anxiety and fakery are downstream of a verification vacuum, so restoring verifiability (receipts, work trials, reproducible benchmarks) attacks the root, not the symptom.

Verbal interviews no longer work; case studies and work trials are now necessary to assess real AI competence. strength: medium

Explicit premises:

AI hands everyone the vocabulary of expertise and generates smart-sounding one-liners on demand.
Being able to confidently explain MCP does not mean you can build one or have shipped anything anyone depends on.

Implicit premises:

Interviews were only ever a proxy that worked because verbal fluency correlated with underlying competence; AI severs that correlation, so the proxy fails.

Inferences:

Any low-cost verbal or written signal of expertise is now cheaply forgeable, so hiring (and by extension all trust judgments) must shift toward costly, demonstrable signals.

Narratives

'Cool. Show me.' — the author's habit of asking anyone who claims AI changed their life to demonstrate it; most responses are basic workflows (summarizing Slack, answering emails, booking something), and the truly indispensable list turns out to be very short.
→ Operationalizes the author's skepticism into a repeatable test and provides the empirical basis for the 'short list' claim
The 5am-cold-plunge-inbox-zero-Peloton hustle of five years ago vs. today's 17-agents / not-opening-email / AI-runs-my-business bragging — 'the exact same performance, just with different props.'
→ Analogy exposing the AI hype as a re-skin of perennial status signaling rather than a substantive change
'I just created this INSANE setup with Dropbox' — a hypothetical past-tech brag that everyone would have instantly called BS on by reading the description.
→ Counter-example that isolates verifiability as the crucial difference between AI hype and previous tech hype
The serial litany of AI headlines: killing education, then eliminating writers/engineers/designers/PMs/sales (in that order), then Google/SEO, then SaaS (which cratered the stock but didn't actually die), now everyone needing an AI agent.
→ Establishes a track record of chronic over-claiming to prime the reader's skepticism
The VC → exec → manager → employee pressure cascade: investors are promised AI miracles, so execs demand miracles, so employees are asked to do the impossible with underperforming tools and are incentivized to perform confidence to protect their careers.
→ Traces the causal chain that makes the theater structurally, not just individually, driven
The author — who works at an AI company and uses AI all day — admitting she asks ChatGPT to rewrite the same paragraph three times because it keeps defaulting to 'LinkedIn wisdom post' mode, and still feels like everyone else has 'cracked the code'.
→ Self-deprecating insider anecdote that de-shames the reader and models the honesty the essay prescribes

Questions addressed

✅ Why does it feel like everyone else has 'cracked the AI code' while I'm stuck on basic uses?
Largely because of a fake baseline manufactured by performative hype; the transformative claims are mostly theater, and modest real gains are worth celebrating.
✅ Why is exaggerated AI hype happening at all?
Four converging causes: attention economics reward exaggeration, AI claims are hard to verify, marketing sells certainty AI can't deliver, and top-down VC→exec→employee pressure forces performance.
✅ How do you hire for real competence when AI lets everyone sound expert?
Shift from verbal interviews to case studies and work trials that require demonstrated, dependable work.
✅ Where does the real value of AI actually get created?
In the 'next thousand prompts' — the ongoing monitoring, evaluation, iteration, and tuning of AI as living systems, the part nobody posts about.
⭕ How can honest behavior actually take hold when the incentives (attention, VC pressure, job protection) reward hype?
Not resolved — the essay prescribes voluntary honesty and 'show me the receipts' accountability but never reconciles this with the incentive gradient it identifies as the root cause.

Gaps

incompleteThe central internal tension: Elena diagnoses the theater as structurally incentive-driven (attention economy, VC→exec→employee cascade, career self-protection) yet prescribes a voluntary remedy — be honest, hold creators accountable, don't amplify untested claims. She appeals to individual virtue against an incentive gradient she herself says rewards hype, without any mechanism (verification infrastructure, benchmarks, changed rewards) to make honesty the dominant strategy. This is a collective-action problem treated as a matter of conscience.
Read next: Costly-signaling and mechanism-design (Spence's job-market signaling), tragedy-of-the-commons remedies (Ostrom on governing shared resources — here, shared trust), and John Cutler on designing organizational conditions/incentives rather than exhorting behavior; connects to the persona's systems-thinking lens and 'organizational adoption of ideas' hidden layer.
incompleteThe 'next thousand prompts' / living-systems point — that real value is in monitoring, evaluation, iteration, and tuning — is asserted but not operationalized. The essay offers no framework for how to actually measure honest AI value (evals, benchmarks, drift detection), which is precisely the verification infrastructure its argument implies is needed.
Read next: LLM evaluation practice, DORA's AI findings on measuring engineering impact, and disciplined benchmark design — directly relevant to the persona's rigorous-adopter interest in what AI tools actually do to productivity and quality.
incompleteThe hiring prescription ('work trials and case studies become absolutely necessary') ignores that work trials are costly, hard to scale, and can themselves be AI-assisted or gamed — the same forgeability problem may simply move one level up. The claim that costly demonstration restores signal is under-examined.
Read next: Signal-design literature on hiring and evaluation under cheap-talk conditions; how costly signals degrade when the cost itself can be outsourced to AI.
illogicalAn unresolved surface tension: she asserts both that 'AI really CAN do crazy stuff' and that people are genuinely building things that could change how we work, and that the truly transformative list is 'super short.' The reconciliation (capability exists but is not yet reliably transformative for most users) is implied but never stated, leaving the two claims in apparent conflict.
Read next: The capability-vs.-diffusion gap — technology adoption/diffusion theory (Rogers) and the distinction between demonstrated capability and reliable, dependable value in real workflows.
closed byOrdinary Engineers, Not Heroic Inventors — The LS-theory/diffusion-theory distinction directly names the tension the gap flags: demonstrated capability ('AI really can do crazy stuff') and diffused, reliable value are different axes, and Japan's hardware dominance without economic overtake is a historical case where capability leadership failed to translate into economic leadership.

Please stop the AI Confidence Theater

Structure

Key terms

Propositions

Arguments

Narratives

Questions addressed

Gaps

Related documents