5
THE LEVERAGE ACCOUNTABILITY DESK131 CLAIMSV2 · APRIL 21, 2026
Read The Leverage ↗Is Evan full
of shit?
Ever read a newsletter and wonder, “Is this guy a moron?” Good news! You can finally answer that question. The Leverage periodically examines all of its forecasts and then publishes a grade. This website is a record of how it has done so far.
131claims published
108mature claims
86essay grades
Not a victory lap.
4
Mostly right
49 claims · 45%3
Mixed
8 claims · 7%2
Mostly wrong
3 claims · 3%1
Wrong
0 claims · 0%T
Too recent / not a forecast
23 claims · 18%The misses
Three calls that deserve the red ink.
“Mostly wrong” means subsequent events contradicted the central direction of the claim. Click any card to find the full record.
Methodology · v2
A grade is an argument, not an oracle.
Claude (Anthropic) performed the grading on April 21, 2026 from the full post corpus and public evidence. Evan directed an adversarial second pass when v1 returned zero wrong calls.
Receipts, please.
- 01
Fetch
Pulled the full archive from The Leverage sitemap and downloaded the post bodies.
- 02
Classify
Separated essays, Weekend Leverage sections, housekeeping posts, and too-recent work.
- 03
Extract
Selected the central claim from each essay and split roundup editions into distinct calls.
- 04
Grade
Scored each claim 1–5 or T against public evidence available after publication.
- 05
Attack
Re-read the most generous results adversarially; six grades moved down in the v2 pass.
- 06
Publish
Released every claim, rationale, confidence level, limitation, and source export—not just the score.
The rubric
Six possible outcomes.
5Right — clearly vindicated by subsequent events
4Mostly right — directionally correct; minor errors or magnitude off
3Mixed — real partial validity; significant aspects wrong or unproven
2Mostly wrong — directionally off; subsequent events contradict the central claim
1Wrong — falsified by events or fundamentally misjudged
TToo recent / not a forecast — insufficient elapsed time, or interpretive content with no falsifiable claim
Adversarial regrade
Where v2 got harsher.
Hire Misfits, Not Missionaries
Mission-coded labs dominated; the exceptions did not support the prescriptive rule.
The Future Machine
Prediction markets grew, but the truth-seeking thesis did not survive their sports-heavy reality.
Profit is for Chumps
Rising token consumption and pricing pivots contradicted the original gross-margin logic.
The AI Promise Gap
Productive AI uses—not porn and companions—became the dominant commercial story.
OpenAI Released an Agent, XAI Released an e-girl
One of three calls landed; the two larger platform predictions did not.
AI Education
Re-reviewed for a downgrade and held at mixed.
Still an upper bound
The claims were extracted from Evan’s own framing. A skeptic or domain expert could select harder claims and produce more downgrades.
Self-vindication risk
The grader could see Evan’s follow-up essays. Those were treated as evidence, never proof, but the contamination risk is real.
A T-heavy tail
Twenty-three claims were too recent or insufficiently falsifiable. The percentages exclude them rather than treating uncertainty as success.
Nothing hidden behind the average
The claim ledger.
Showing 131 of 131 records
001Don't Sell The WorkHigh confidenceTToo recent / not a forecast
Claim extracted for grading
Sequoia/a16z 'sell work, not software' thesis is wrong — only AI customer support has worked. Three reasons: (1) production opacity gone (buyers can see token costs); (2) scissors problem (token costs falling but token consumption per task rising faster); (3) quality specification is hard. Winning AI companies (Cursor, Harvey, Copilot) sell software access, not work. Context layer is where pricing power lives.
Why it received a T
Strong contrarian piece against industry consensus. Cited data (KPMG forcing 14% audit discount, ~92% AI cos use hybrid/subscription pricing, only 7% pure outcome-based) is fresh and supports the thesis. Builds on your Feb 12 Context is King essay. Too soon to grade but argument is well-constructed and timely. T leaning 5.
Read the original essay ↗002Guys, Aliens Might Be Real (WL-essay)High confidenceTToo recent / not a forecast
Claim extracted for grading
Silicon Panopticon is real (Meta facial recognition glasses, Border Patrol Clearview, Amazon Ring). Brockman/Anthropic pick political sides ($20M Anthropic PAC). Anthropic 10x revenue 3 years straight ($380B valuation). Bytedance Seedance 2.0 is best video model — used mostly for IP infringement slop. Eggers' The Circle no longer feels farcical.
Why it received a T
Silicon Panopticon predictions all materializing. Anthropic numbers stunning. Sloppening framework continues to be predictive. T because most events are very recent (within 60 days at edge of cutoff).
Read the original essay ↗003Context is King (Notes on the SaaSacre)High confidenceTToo recent / not a forecast
Claim extracted for grading
$300B SaaSacre justified — software splitting into 3 layers: systems of record (databases), point solutions (interface), and a new context layer (institutional knowledge that directs agents). Context layer captures payroll budget, not IT budget. Compounds as agents execute workflows. Battle: ServiceNow/Notion/Glean vs OpenAI Frontier/Anthropic Cowork.
Why it received a T
Outstanding analytical frame; quickly being adopted by other writers. 'Context layer' has become a recognized term in enterprise software discourse. Too soon to grade winners but the diagnosis of where value migrates is well-supported. T leaning 5.
Read the original essay ↗004AI Takes Over the Super Bowl (WL-essay)High confidenceTToo recent / not a forecast
Claim extracted for grading
AI Super Bowl ads (15+ this year, including Anthropic, OpenAI, ai.com) signal distribution crunch. Crypto did this before its bubble popped — could signal same for AI. Adobe spending $1.4B on ads. Apple killed health coach in favor of Siri integration — vindicating tech-of-interiority thesis. Waymo using Genie 3 world models for expansion ($16B raise at $126B post-money).
Why it received a T
Most claims are about events that just happened. The Super-Bowl-as-bubble-signal is a hypothesis that needs more time. Apple Health/Siri integration confirms your Jan 25 prediction. Waymo world-models call validates your Jan video essay. T leaning 5.
Read the original essay ↗005The Idiot Index for Code (a16z Speedrun guest)Medium confidenceTToo recent / not a forecast
Claim extracted for grading
Coding agents make building cheap, but founder time is the new bottleneck. Founders should default to 'do without' before 'build vs buy'. Apply Musk's 5-step algorithm. Idiot index has inverted: code is no longer expensive, attention is.
Why it received a T
The framework is durable and well-articulated. The contrarian advice ('don't replace your CRM with vibe-coded software') is backed by your follow-up Feb 8 piece ('Beware Claude Code'). Hard to grade as forecast — more of a how-to-think. T leaning 4.
Read the original essay ↗006TikTok Has a New Challenger (WL-essay)High confidence4Mostly right
Claim extracted for grading
Maintenance review. Zuckerberg to spend $115-135B on AI in 2026. OpenAI ads launched but lack steerability and conversion data — repeats Snapchat's brand-ads mistake. Institute for Progress directing $25B in AI philanthropy. Upscrolled (TikTok competitor) hit #1 — still controlled by algorithms. Books rule.
Why it received a 4
Datacenter spend numbers confirmed by Q4 earnings. The OpenAI ads critique is sharp and well-supported (your follow-up Feb essay on Adobe $1.4B ads spend confirms distribution pressure). Strong 4.
Read the original essay ↗007On Stewart Brand's Maintenance (Tech Canon)Low confidenceTToo recent / not a forecast
Claim extracted for grading
Brand's Maintenance is a meandering but spiritually enriching book — reads it as 'reflecting' rather than sweeping or scattered. Worth reading not for argument but for connection-making (Ezra Klein quote). Belongs in Tech Canon.
008YouTube Declares War on AI Slop (WL-essay)High confidence5Right
Claim extracted for grading
Apple making Siri ChatGPT competitor (necessary but late). TikTok deal done — 80.1% American JV but algo control still effectively ByteDance. Runway 4.5 indistinguishable from reality (1043-person test). YouTube CEO targeting AI slop in 2026 letter. Anthropic constitution is fascinating ethical artifact. Code is downstream of secrets — speed of building doesn't replace need for customer insight (Casado).
Why it received a 5
Apple Siri overhaul confirmed in subsequent reporting. TikTok 80/20 deal correctly described and called weird. Video model indistinguishability story continues to play out (Sora, Veo 3, Seedance). YouTube AI slop fight is real. Strong 5.
Read the original essay ↗009AI, GLP-1s, and MeMedium confidenceTToo recent / not a forecast
Claim extracted for grading
GLP-1s and AI health are 'technologies of interiority' — they modify inner currents (appetite, attention, motivation). Combination is the future. Will reshape what we expect from companies like Apple. (Paywalled stack/diet/regime details.)
Why it received a T
Personal experiment is genuinely working for the user. Macro thesis on tech-of-interiority is coherent and probably right; explicit Apple Health integration prediction was confirmed within 2 weeks (your Feb 8 essay 'Apple is realizing AI is existential' validates). T because the longer-term societal-reshaping claim is years out.
Read the original essay ↗010OpenAI Is Doing Ads While Anthropic Is Doing Unemployment (WL-essay)High confidence4Mostly right
Claim extracted for grading
OpenAI launched ads ('this had to happen' — Evan called it). Microdramas (TikTok PineDrama, My Drama) are next slop format. Anthropic's Economic Index confirms productivity gains overstated by ~50% when you account for bottleneck/multiplicative tasks (Gans-Goldfarb O-Ring framing). Apply 50% discount to all AI productivity projections.
Why it received a 4
OpenAI ads launched as predicted. Microdramas continue to grow. The Anthropic Economic Index findings are a direct, valuable insight. The bottleneck-task discount is a sharp analytical move. Strong 4.
Read the original essay ↗011Google's Attempt To Commoditize Everyone (UCP)High confidenceTToo recent / not a forecast
Claim extracted for grading
Google's UCP is classic 'commoditize your complement' — attempts to standardize commerce so AI agents can shop, pushing margin out of execution and up to the decision layer where Google lives. Execution stops being the moat; advertising pressure moves upstream.
Why it received a T
Excellent strategic frame; UCP analysis is sharp. Too soon to grade adoption outcomes — protocol launched Jan 2026 and competing standards (Anthropic MCP, OpenAI agent protocols) are jockeying. The diagnosis of Google's strategic intent is well-supported. T leaning 4.
Read the original essay ↗012Dude, I Just Fired Half My Sales TeamMedium confidenceTToo recent / not a forecast
Claim extracted for grading
AI is automating SDR jobs — unit of advertising changing from impression to conversation. Built AI-salesperson experiment for The Leverage. Future: personal AI agents shopping with retailer AI agents pitching them.
Why it received a T
The SDR-replacement story is borne out (your prior FDE essay framed this trajectory). AI-as-salesperson experiment is too new to grade outcomes meaningfully. Personal-AI-agent commerce future likely but not yet realized. T leaning 4.
Read the original essay ↗013New Year, New AI, New MeLow confidenceTToo recent / not a forecast
Claim extracted for grading
Resolutions fail because you remain the same person. Foreign-language psychological-distance effect can be replicated via AI conversations. Built and shipped 'Personal Strategy System' ($10 Notion + LLM integration) as a Foucauldian 'technology of the self'.
Why it received a T
Self-help / experimental piece. The product launched and your Jan 2026 followup ('Dude, I Just Fired Half My Sales Team') reused the AI persuasion thesis. Mark T because the underlying claim about AI as technology-of-the-self is too soon to grade.
Read the original essay ↗0142025 is Winding Down, Nuclear is Blowing Up (WL-essay)High confidence4Mostly right
Claim extracted for grading
Models are getting better predictably (METR time horizon doubling every 7 months — Opus 4.5 hit 4h49m). Streamers losing commercial-free subscribers — sloppening subsidized by consumer choice. Chai Discovery $130M for AI molecular design. AI bringing back nuclear power (Last Energy $100M, X-energy $700M; 14 nuclear startup deals in 2025 totaling $2.9B+).
Why it received a 4
Models continue improving — Anthropic's Feb 2026 chart confirms 10x revenue. Streaming ad-tier shift continues. AI molecular design (AlphaFold/Isomorphic Labs) continues. Nuclear capital cycle continues — Oklo, Last Energy. Strong 4.
Read the original essay ↗015How To Make Taste Worth $100 BillionMedium confidence4Mostly right
Claim extracted for grading
DFW's prophecy coming true via The Sloppening. 'Taste Ladders' are new type of company that can hit $100B outcomes by training tasteful consumption. Three economic forces explain why slop won (paywalled). Productive friction can fight algorithmic stupidity at scale.
Why it received a 4
Taste Ladders framing has been picked up; FWB Fest invited you to speak on it (April 2026 reference in user memory). Whether a $100B taste ladder company will emerge is unproven. Framework has currency.
Read the original essay ↗016The Most Important AI Paper of the Year (WL-essay)High confidence5Right
Claim extracted for grading
Google's video-simulator-trains-robots paper is most important AI paper of year. Runway/Luma type companies are now cheaper-way-to-make-robots, not just cheaper-way-to-make-video. Will accelerate robotics deployment dramatically. GPT-5.2 release fine but doesn't change much. xAI/El Salvador edu deal will fail. Medra and Excelsior signal AI-accelerated science wedge. Skild raising $1B+ for robot brain.
Why it received a 5
Sim-to-real for robots continued to be major theme in 2026 (your Feb 2026 piece on Waymo using Genie 3 confirms). Runway/Luma valuations reflected this. Skild raise materialized. Strong 5.
Read the original essay ↗017The Anti-Slop ListLow confidenceTToo recent / not a forecast
Claim extracted for grading
Slop can be fought with new companies AND new habits. Reading challenging media is sloppening antidote. List of ~15 books/films as worthy diet.
018Uh Oh, AI Is Influencing Elections (WL-essay)High confidence5Right
Claim extracted for grading
Nature/Science papers prove AI is most persuasive tech ever invented. Robotics make society weird. Sloppening framing. Stunt marketing is bad strategy. Korean tariffs (15%) will hit memory supply chain hard. Aaru ($50M raise) bets on AI agent population simulation.
Why it received a 5
AI persuasion findings only amplified through 2026. Sloppening is recognized term. Memory crunch story continues. Population sim still early but interesting. Strong 5.
Read the original essay ↗019THE SLOPPENINGMedium confidence5Right
Claim extracted for grading
Netflix buying Warner Brothers proves attention always beats craft. (Body is paywalled — implied: future of media is dopamine-optimized AI slop, Netflix wins).
Why it received a 5
Sloppening as a coined term has gained currency. Netflix-WBD confirmed. Your April 2026 'Slop Factory Needs Line Inspectors' shows you're still riding this thesis.
Read the original essay ↗020A Robot Walks Into a Coffee ShopMedium confidence4Mostly right
Claim extracted for grading
Specialized coffee robots ($50K, 350-700 drinks/hr) crush humanoids ($20K BOM, ~80 drinks/hr at best). Economics flip when you walk through your front door (humanoids may win in unstructured home environments). Future cafes will be operated by supply chain pros, not coffee snobs. Teece's 'complementary assets' framework explains who actually captures value.
Why it received a 4
Specialized robotics has continued to scale. Humanoids continue slow/expensive (1X NEO teleop debacle). The home/work dichotomy now widely discussed.
Read the original essay ↗0219 Advanced Prompts to Master Nano BananaMedium confidence4Mostly right
Claim extracted for grading
Nano Banana raises baseline acceptable visual quality just as ChatGPT raised baseline writing quality. Disrupts existing visual intelligence software. Educated users have temporary arbitrage window.
Why it received a 4
Aging well. Visual AI tools have proliferated and become integrated into many workflows.
Read the original essay ↗022Can AI Advance Science? (WL-essay)High confidence5Right
Claim extracted for grading
Trump 'Genesis Mission' is signal that US gov treats AI infra for science as Manhattan-Project-scale push. Anthropic vs OpenAI is a real strategic divergence. Ads in AI are necessary, not just acceptable. Nano Banana is future of visual AI. Enhanced Games (juiced athletes) is media + DTC vertically integrated. Gravis Robotics (excavator brain attachment) shows ops-first robotics.
023Are Ads the Answer to AI's Problems? (Sourcegraph interview)Medium confidence4Mostly right
Claim extracted for grading
Quinn Slack at Sourcegraph generated $5-10M ARR in a month from ads in coding tools. Ads might be the demand floor that justifies datacenter buildouts. Coding agents could fuse Google-intent with Instagram-engagement.
Why it received a 4
OpenAI launched ads (Jan 2026), Sourcegraph AMP Free continues. The 'ads as demand floor' thesis alive and being tested. Your Jan 2026 critique of OpenAI ads not having steerability validates Quinn's deeper thesis about trust.
Read the original essay ↗024You Don't Need to Own the Pipes If You're the Water (Opus 4.5)High confidence5Right
Claim extracted for grading
Anthropic is winning by being neutral 'water' — not trying to own application stack. Opus 4.5 more capable AND cheaper per task. Anthropic added $4B+ revenue in last year — historically unprecedented. OpenAI's 'own everything' strategy is not the only path.
Why it received a 5
Anthropic Switzerland strategy continued to pay off through 2026 — Feb 2026 raised $30B at $380B valuation, revenue chart shows another 10x year. Coding remains dominant revenue driver. Strong 5.
Read the original essay ↗025Data Centers are Gobbling Up All the Capital (WL-essay)High confidence4Mostly right
Claim extracted for grading
Anthropic $15B raise + Google's '1000x infra' + Nvidia earnings = bubble inflating not deflating. xAI valuation makes no sense ($230B at maybe $500M-$1B revenue burning $1B/month). DoorDash thesis playing out. Wispr dictation app shows transcription as new computing primitive. Suno AI music a true consumer breakout. Physical Intelligence robot brain raise.
Why it received a 4
Bubble continued to inflate. Aggressive capex strategies have proliferated. xAI valuation skepticism feels increasingly justified. Wispr-style voice interfaces continue to grow. Suno continued to scale.
Read the original essay ↗026Google's New Image Model is Incredible (Nano Banana Pro)High confidence4Mostly right
Claim extracted for grading
Nano Banana Pro raises floor on acceptable knowledge work like pepper raised floor on cooking. Image AI finally good enough for B2B SaaS integration. Every visualizable workflow will become visual.
Why it received a 4
Nano Banana Pro widely acknowledged as breakthrough image model. Many tools (Notion, Canva) have integrated similar models. The 'all info becomes visual' prediction on track.
Read the original essay ↗027The Future of Software isn't Software (DoorDash)Medium confidence4Mostly right
Claim extracted for grading
DoorDash exemplifies AI-era thesis: AI lowers creation costs, raises distribution costs. Aggregated demand is uncopyable — ultimate AI-era asset. DoorDash moving from delivery marketplace to full restaurant infrastructure.
Why it received a 4
DoorDash strategy on track. The 'demand aggregation > tech' thesis adopted in subsequent SaaSacre commentary.
Read the original essay ↗028AI Job Disruption is Here (WL-essay)High confidence5Right
Claim extracted for grading
Stanford/ADP payroll paper shows AI is hitting young workers hardest (-16% relative employment for 22-25 year olds in most exposed jobs). Robotics report most important essay in months. Worldcoin worth taking seriously. Data center supply chain breaking down (Samsung memory price hikes, Majestic startup).
Why it received a 5
Junior tech jobs continue to be cut disproportionately. Robotics report did indeed get validated by adjacent funding within days. Memory supply chain crunch continued. Strong 5.
Read the original essay ↗029Who Actually Makes Money When Robots Work?Medium confidence4Mostly right
Claim extracted for grading
1X NEO's launch shows gap between marketing and reality (teleop in disguise). Robotics is at same point LLMs were in 2021 — clear direction, no escape velocity company. Liability is unsung axis of who deploys autonomous systems. 4-layer bet framework (body, senses, brain, ops). 3 ways to win (paywalled).
Why it received a 4
Robotics has continued to attract massive funding (Skild's billion-dollar raise within days, Physical Intelligence at $5.6B). The liability framing is original and prescient. The 'no escape velocity company yet' call still right as of April 2026. The 4-layer framework is a useful synthesis.
Read the original essay ↗030Can Crypto Fix AI Slop? (Worldcoin interview)Low confidenceTToo recent / not a forecast
Claim extracted for grading
Proof-of-human is a real problem with no good answer; Worldcoin's orb might be best shot. The irony of OpenAI-affiliated company solving AI-caused problem is real. Worldcoin has 17M active users, $375M raised — worth taking seriously.
Why it received a T
Worldcoin continues to operate but hasn't become canonical proof-of-human. Problem remains unsolved. T leaning 3.
Read the original essay ↗031Should Meta Let Drug Dealers Buy Ads? (WL-essay)High confidence4Mostly right
Claim extracted for grading
DoorDash $20B selloff for $300M extra spend is market mispricing strategic logic. Algorithmic predation paper supports DoorDash hardcore investment strategy. Meta charges scam ads higher prices rather than banning them. 3 academic papers on social media corruption.
Why it received a 4
DoorDash thesis published as full essay 10 days later — strategy seems holding. Meta scam ads issue continues unaddressed. The 'social media corrupts even LLMs' point is increasingly mainstream.
Read the original essay ↗032What Revenue Multiple Does Jesus Get You?Medium confidence3Mixed
Claim extracted for grading
Companies can build 'theological monopolies' by Jesusifying existing categories. Hallow ($105M raised, #1 app) shows demand. Angel and Gloo are public-market test cases. Christian market has demographic tailwinds (Gen Z attendance climbing).
Why it received a 3
Hallow continues to grow. Angel Studios mixed results. Gloo IPO performance muted. Thesis interesting but not proven by breakout success in 6 months.
Read the original essay ↗033Will You Listen to AI Music? (WL-essay)High confidence4Mostly right
Claim extracted for grading
OpenAI/MS deal reveals OpenAI in early-stage startup mode taking massive risk. Google does $100B in a quarter — diversification staggering. AI-generated R&B artist (Xania Monet) hitting charts. Bending Spoons applying PE playbook to tech. Mercor expert network valuation faulty comparison.
Why it received a 4
All claims directionally validated. AI-generated music continued to chart. Google's diversification unmatched. Mercor critique was sharp and aged well — its customer concentration risk is real.
Read the original essay ↗034Why Is The Internet Bad Now? (Doctorow interview)Medium confidence4Mostly right
Claim extracted for grading
Doctorow's 'enshittification' is real but framework overreaches. Apps as 'websites with handcuffs' under DMCA. Author skeptical of surveillance ad ineffectiveness, agrees on data portability. Trump's trade war might accidentally fix some of this. (You push back substantively on Doctorow.)
Why it received a 4
Enshittification entered mainstream discourse. Specific predictions on regulators partially borne out. Trade war angle still speculative. Your skepticism of Doctorow's most extreme claims (about ads) has aged well — your own Oct/Nov pieces show ads are working.
Read the original essay ↗035Meet the New OpenAI, Same Problems As The Old OneMedium confidence4Mostly right
Claim extracted for grading
OpenAI/Microsoft restructuring shows OpenAI under pressure, not in control. Three details (paywalled) reveal AGI timelines, abandoned business lines, Microsoft's true priorities.
Why it received a 4
OpenAI continues to ship hits but financial complexity has only grown ($852B private valuation, $115B 2026 spend). Microsoft tensions continue to surface (your Nov 2025 'OpenAI loses $11.5B in a quarter' confirmed).
Read the original essay ↗036What Happens When Amazon Replaces Its Workforce with Robots? (WL-essay)High confidence5Right
Claim extracted for grading
Amazon plans to replace 500K jobs with robots over 10 years. Historical analog: Luddites/Rust Belt — automation freed labor but not from scarcity. Populism and economic rage will worsen. Snap's AR glasses spinoff is shiny-object syndrome. OpenAI browser launches, AI agent browsers limited by security/utility.
Why it received a 5
Amazon robotics plan progressing. Political/social implications increasingly visible. Snap AR continues to disappoint. AI browsers stayed niche through Apr 2026. Strong 5.
Read the original essay ↗037What's the Bet: Sublime (sponsored)Low confidenceTToo recent / not a forecast
Claim extracted for grading
Sublime's bet: notes should DO things, not just store things. AI + embeddings make personal note-taking searchable by feel. Personal context is what AI tools lack — personal library has competitive edge.
Why it received a T
Sublime continues operating but hasn't broken out. The 'personal context as moat' idea is generally right (informs your Context Layer thesis). Hard to grade specific bet.
Read the original essay ↗038The Future of SaaS Is Already Here — EliseAIHigh confidence5Right
Claim extracted for grading
EliseAI proves the AI displacement playbook works in practice. Vertical SaaS incumbents accepting AI 'partnerships' are being eaten. AI-first vertical players will need to give away SaaS features as survival move.
Why it received a 5
The SaaSacre (Feb 2026) is the market pricing in exactly what EliseAI was demonstrating. The give-away-the-CRM playbook now visible across multiple verticals (Harvey, Sierra, Decagon). Strong 5.
Read the original essay ↗039Why Sam Altman Has to Let ChatGPT Get Horny (WL-essay)High confidence5Right
Claim extracted for grading
OpenAI must allow erotica because monetization pressure is extreme (Double Bind Theory). Apple Vision Pro re-release is 'crushing disappointment'. Deel/$300M raise despite espionage allegations shows growth papers over sins. $12M Armstrong dishwasher robot economics.
Why it received a 5
ChatGPT did launch erotica mode for verified adults in late 2025. Apple Vision Pro continued to be a flop. Hardware AI spending wild. Double Bind Theory framework holds up well. Strong 5.
Read the original essay ↗040This Email is Now 10x Better (WL-essay)High confidence5Right
Claim extracted for grading
Big Tech 'AI agent' announcements are mostly chatbots-with-RPA, not real agents. 1 in 5 teens has had AI romance. OpenAI/XAI 'creative financing' (chips-for-equity, AMD swap) escalating risk. Polymarket-style sports betting blurs into real prediction markets.
Why it received a 5
Big Tech 'AI agent' marketing continued to be largely fake. Teen AI romance numbers continued to climb. Creative financing only got weirder (Oracle bond raise). Polymarket sports betting did mainstream. Strong 5.
Read the original essay ↗041Which AI Companion Actually Works?Medium confidence4Mostly right
Claim extracted for grading
Friend.com pendant is uncomfortable failure. AI companions face privacy-usefulness paradox: not invasive enough to be useful = not useful. Hundreds of billions pouring into category despite paradox.
Why it received a 4
Friend.com remained niche/derided. The companion category continues facing same trade-off. Hundreds of billions claim roughly right (Meta + OpenAI hardware investments, ChatGPT companion mode).
Read the original essay ↗042The New King of the App Store (WL-essay)High confidence5Right
Claim extracted for grading
Sora became #1 app — validates inevitable trajectory of AI video. AI agents real and immediate (Lindy interview). OpenAI 7-trillion essay generated grumpy emails but launch validated take.
Why it received a 5
All claims confirmed in real-time by Sora #1 ranking, Lindy continued growth. Strong 5.
Read the original essay ↗043The AI Agent Era Is Here (Lindy interview)High confidence4Mostly right
Claim extracted for grading
Lindy's progress from 'pretty good' to 'amazing' over 6 months is the agent inflection point. Computer Use + horizontal agents finally generalize. Coordination costs dropping inside firms (Coase shift). Within ~2 years, Lindy will spend more on tokens than payroll.
Why it received a 4
Agent era has arrived — Sierra, Decagon, Cognition, Cursor's agent mode all confirm. Computer Use has gotten dramatically better. Token-vs-payroll prediction on track for some companies but still 12-18 months from being verifiable at scale.
Read the original essay ↗044First Access to OpenAI's Video AppHigh confidence5Right
Claim extracted for grading
Sora app is a real product, not just slop — better than Meta's Vibes. Cameos/consistency is meaningful technical bet. OpenAI took 'legally ambitious' choices on IP. Feels empty to scroll AI videos but tech is real.
Why it received a 5
Sora became #1 app in App Store within days (your own Oct 5 follow-up confirmed). IP issues with Disney/etc became a major story. The 'feels empty' critique picked up by mainstream media. Strong 5.
Read the original essay ↗045The Case for Spending Seven Trillion on AIMedium confidenceTToo recent / not a forecast
Claim extracted for grading
OpenAI's plan to build 250GW of datacenters by 2033 ($7T) is mathematically aggressive but might be defensible. Required: $14T in revenue, more than all of big tech combined currently does. Worth taking the upside case seriously.
Why it received a T
Data center boom continued through 2026 — Oracle's $156B commit, Microsoft/Google/Meta all matching. Whether the math justifies $7T is still TBD (target is 2033). Took the right contrarian risk. T because time horizon is long.
Read the original essay ↗046AI's Strange Paradox (WL-essay)High confidence5Right
Claim extracted for grading
AI education needs more entrepreneurs. Notion's Agent bet is interesting. Meta's 'Vibes' AI video feed is grim, but this kind of product WILL happen. Amazon's $2.5B settlement for dark patterns highlights universal problem. TikTok deal getting weirder.
Why it received a 5
Vibes-style AI feeds did proliferate — Sora launched Oct 1 and hit #1 in App Store within days, exactly your prediction. Amazon dark patterns industry-wide. TikTok deal kept getting weirder (your Jan 2026 followup). Strong 5.
Read the original essay ↗047What's the Bet: Notion (sponsored)Medium confidence3Mixed
Claim extracted for grading
Notion is betting one Agent for cross-app workflows beats specialized agents. Block-based architecture gives Notion a structural agent advantage.
Why it received a 3
Notion's Agent has been real but hasn't dominated — Glean, ChatGPT, Slack agents all competed for similar territory. Bet alive but not clearly winning.
Read the original essay ↗048The Most Exciting Software Category I've Seen in Years (AI Edu)Low confidence3Mixed
Claim extracted for grading
AI education will scale to billions in revenue AND make society smarter. Market is wide open; existing tutor startups still small. Talented operator/investor entering now will win big.
Why it received a 3
As of April 2026, AI tutoring has grown but no clear billion-dollar winner emerged in 7 months. Speak.com, Khanmigo, Alpha School are growing. Bull case alive but not validated yet. Note Evan's own Dec 2025 'Chatbots aren't going to fix education' WL section walked back some of this enthusiasm by noting xAI/El Salvador deal misunderstands education.
Read the original essay ↗049Cheers to VCs Taking Insane Risks (WL-essay)High confidence4Mostly right
Claim extracted for grading
Robotics, scientific research, new chips are getting real risk capital — sign of healthier startup ecosystem. Lila Sciences ($235M Series A) and Dyna Robotics ($120M) as examples. Nvidia/Coreweave incestuous deals notable.
Why it received a 4
Robotics funding boom continued through 2026 (your Nov 2025 robotics report validated by adjacent funding within days). Scientific research deals (Medra, Excelsior Sciences, Chai Discovery) continued. Nvidia ecosystem incest only got more pronounced.
Read the original essay ↗050BREAKING: A Questionable Deal for TikTokHigh confidence5Right
Claim extracted for grading
TikTok deal is canonical example of tech being downstream of politics. Yass donations + ByteDance equity directly shaped Trump's TikTok stance. Oracle won bid largely because of Ellison's Trump alliance. AI industry is following crypto in launching political lobby SPACs ('Leading the Future').
Why it received a 5
TikTok deal progressed exactly along these lines (your Jan 2026 followup confirmed: 80.1% American JV with Oracle/Silver Lake/MGX, ByteDance retains algorithm). Crypto Fairshake → AI Leading the Future arc is now visible everywhere — Anthropic donated $20M Feb 2026, Brockman is one of Trump's largest donors. Strong 5.
Read the original essay ↗051So, is AI Gonna Kill Us All? (Soares interview)Low confidenceTToo recent / not a forecast
Claim extracted for grading
Nate Soares' AI doom thesis is well-known but hasn't slowed AI investment. Their movement spawned 4+ cults including one linked to murders. Worth understanding the AI safety lobby because they're influencing policy.
052AI Agents Are Finally Working (WL-essay)High confidence4Mostly right
Claim extracted for grading
AI agents have crossed PMF threshold via narrowing of ambition. Peptides as next big consumer market. Nikita Bier (X head of product) has too much power over public discourse.
Why it received a 4
Agents-finally-working call borne out by Lindy, Sierra, Decagon, Cursor's agent mode. Your Oct 2025 Lindy interview confirmed this was real. Strong directional call.
Read the original essay ↗053The Booming Market for Injecting Yourself with Chinese ChemicalsMedium confidenceTToo recent / not a forecast
Claim extracted for grading
Consumer companies need 3 catalysts: tech, distribution, or cultural shift — AI has all three. Grey-market peptides represent ~$500M+ of unmet demand US-side that big companies will consolidate.
Why it received a T
Peptide market continues growing. GLP-1 prescriptions hit record numbers (your Jan 2026 piece confirms — 1 in 5 US adults). But no breakout US peptide company has emerged yet. Macro thesis right; specific consolidation prediction unproven. T leaning 3.
Read the original essay ↗054The VC Who Disrupted His Own Career — Bryce RobertsMedium confidence4Mostly right
Claim extracted for grading
AI compresses cost of code the way 2005-2010 OSS+AWS+AdSense did — enabling lean, durable, founder-controlled companies. Indie 2.0 is inevitable shape of AI era. Cult dynamics work as distribution playbook.
Why it received a 4
The 'lean AI company' archetype has flourished (Midjourney, Lovable's small team relative to revenue, Cursor staying small). Indie 2.0 is actively deploying. The cult-as-distribution claim consistent with what played out.
Read the original essay ↗055Every Startup a Church, Every Founder a ProphetHigh confidence5Right
Claim extracted for grading
AI commoditizes code, distribution becomes vector of competition. Cult formation is emerging marketing strategy — Tesla, Anduril, Stripe, Palantir, Airbnb as exemplars. Belief = priceless when software is free.
Why it received a 5
Identity-driven marketing has only grown — Anthropic Super Bowl ad, OpenAI brand-building. The 'cult' framing now widely used. Distribution undeniably the new bottleneck (your own Feb 2026 'AI Takes Over Super Bowl' confirms). Strong 5.
Read the original essay ↗056What's the Bet: Framer (sponsored)Medium confidence3Mixed
Claim extracted for grading
Framer is betting LLMs collapse design/code into one workflow. Will compete with Webflow/Figma/Squarespace.
Why it received a 3
Sponsored analysis. Framer continues to grow but hasn't dominated. Figma's AI moves and Bolt/Lovable/v0's rise have been bigger stories. The bet is alive but not clearly winning.
Read the original essay ↗057On Founding in the Age of AILow confidenceTToo recent / not a forecast
Claim extracted for grading
Tech founders have a 'divine responsibility' to build elevating, not just entertaining, products. Many popular AI chatbots are sexual/companion bots; this is a moral failing. We can learn from social media's mistakes.
Why it received a T
Manifesto/values piece. Companion-bot prediction continues to be borne out. T — not graded-style forecast.
Read the original essay ↗058Meta's New Chatbots Are an Abomination (WL-essay)High confidence5Right
Claim extracted for grading
Meta's child-safety failures with chatbots are structural pattern. Computer vision investment thesis validated by Squint $40M raise. Wolf of Wall Street commentary.
Why it received a 5
Meta's chatbot scandals continued through 2025-2026. CV thesis has been validated multiple times (Meta facial recognition glasses, ICE Clearview, drone warfare). Strong 5.
Read the original essay ↗059Profit is for ChumpsHigh confidence2Mostly wrong
Claim extracted for grading
Negative gross margins for AI startups are rational — different from prior SaaS dynamics. AI cost decreases regularly, so subsidizing 'workflow acquisition' (vs account acquisition) makes sense. Cursor exemplifies.
Why it received a 2
REGRADED DOWN. Your own Feb 2026 'Don't Sell the Work' essay is a substantive walk-back. The August thesis assumed token costs would keep falling fast enough to make negative margins rational. The Feb essay correctly identified the scissors problem — token costs ARE falling but token consumption per task is rising faster (10K reasoning tokens for 200-token answers, 5,400-token average sequence lengths). Net result: AI app margins haven't improved. KPMG forced a 14% audit discount; ~92% of AI cos pivoted to subscription/hybrid pricing precisely because the negative-margin path didn't work. Cursor exists, but Cursor wins on subscription + Anthropic-scale token deals, not on the workflow-subsidy bet you described. This is the cleanest case in the archive of you correcting yourself in print, which is to your credit, but the original call was wrong.
Read the original essay ↗060Models! Get Your AI Models Here! (WL-essay)High confidence5Right
Claim extracted for grading
OpenAI's open-weight models matter because they can run on edge devices with on-device fine-tuning. Models aren't the moat, access is — context + tools matter more than raw intelligence. OpenAI is an application company for now; will move to enterprise + hardware.
Why it received a 5
Open weights/distillation continued to advance. The 'context > raw intelligence' thesis became your 'Context is King' essay 6 months later. OpenAI launched browser, Sora app, expanded enterprise — exactly the trajectory. Strong 5.
Read the original essay ↗061Breaking: GPT-5 Is OutHigh confidence5Right
Claim extracted for grading
GPT-5 is #1 in nearly every category but margin is small — not a GPT3-to-GPT4 leap. Strategic implication: every foundation model lab will fight in same market — code.
Why it received a 5
Code is unambiguously the central battleground in 2026 — Cursor, Claude Code, Codex, Cognition, Replit, Lovable, Bolt. Anthropic's coding-focused dominance and Cursor's growth confirm. The 'small leap' diagnosis was correct.
Read the original essay ↗062The Hottest Job in Tech (FDE)High confidence4Mostly right
Claim extracted for grading
Forward Deployed Engineers are the hottest role because AI is bigger than the cloud transition. AI requires reimagining staffing/workflows. (Paywalled FDE pricing/ACV details.)
Why it received a 4
FDE/applied AI engineer roles have indeed exploded through 2026. OpenAI, Anthropic, Cursor all hiring aggressively. The 'AI as bigger than cloud' framing is now consensus.
Read the original essay ↗063Money, More Money, and Oh Yeah, Money (WL-essay)Medium confidence4Mostly right
Claim extracted for grading
Piketty R>G playing out: massive AI capex + bad jobs report = brewing social conflict. Runway/Luma's pivot to robotics/AV training = 'emergent workflow' (go after small markets that tech change explodes). Zuckerberg's superintelligence essay is intellectually weak.
Why it received a 4
Jobs reports through 2026 continued disappointing. Zuckerberg's superintelligence push has not produced clear win — Meta now spending $115-135B/year and is still mocked. Runway/Luma B2B pivot real (Google Veo robot training paper Dec 2025 confirmed sim-to-real).
Read the original essay ↗064Are Ads All You Need?High confidence5Right
Claim extracted for grading
Ads-vs-subscriptions debate is wrong framing — real question is GPU allocation. Ads work for info retrieval/analysis (OpenEvidence at $50M ARR, 40% of US doctors). Subscriptions win when AI provides 'new way of working' (coding).
Why it received a 5
OpenAI launched ads in Jan 2026 — exactly your prediction. Sourcegraph AMP Free hit $5-10M ARR in a month (your Nov 2025 interview). Subscription model thrives in coding (Cursor, Anthropic). Strong 5.
Read the original essay ↗065Trump's Plan to Stop Woke AI (WL-essay)High confidence5Right
Claim extracted for grading
AI policy details (open source incentives, decreased red tape) mostly reasonable despite Trumpian rhetoric. Both parties are 'believers' in AI changing everything. Anthropic's 'subliminal learning' paper shows we don't understand what's inside models.
Why it received a 5
Bipartisan AI consensus only intensified through 2026. 'AI as game of nations' frame is now conventional wisdom. Interpretability concerns continue to be substantiated. Strong 5.
Read the original essay ↗066Are You Using AI, or Is AI Using You? (Heidegger)Low confidenceTToo recent / not a forecast
Claim extracted for grading
Heidegger's 'Enframing' applies to AI: technology reveals our way of seeing. Danger is treating tech as neutral. AI hijacks our psychology more easily than people realize (Josh Miller hack demo).
Why it received a T
Philosophical/interpretive piece. The framing doesn't make falsifiable predictions. The Josh Miller hack demo is real and replicable. T because it's not graded-style forecast content.
Read the original essay ↗067OpenAI Released an Agent, XAI Released an e-girl (WL-essay)Medium confidence3Mixed
Claim extracted for grading
ChatGPT Agent's 'better than humans on half of economically valuable tasks' is exaggerated marketing. Substack raising $100M means they MUST become social media giant. XAI's sex bots are morally repulsive; $200B sex-bot company plausible.
Why it received a 3
REGRADED DOWN. Mixed bag. ChatGPT Agent skepticism aged well — the product was useful for low-stakes tasks but didn't replace knowledge work as marketed. Substack as 'social media giant' overshot — they did launch ads, video, and notes, but they're not Twitter/Meta scale. The '$200B sex-bot company plausible' call has not happened — no AI companion company has cracked even a fraction of that. The XAI anime-girl product exists but hasn't moved the needle financially. Three specific calls, one right (ChatGPT Agent), two clearly off (Substack scale, $200B sex-bot).
Read the original essay ↗068The Future Machine (Prediction Markets)High confidence2Mostly wrong
Claim extracted for grading
Prediction markets are 'as big a deal as AI' — first mass-market product where truth is incentivized by design. 79% more accurate than alternatives. 4-participant taxonomy explains why they work.
Why it received a 2
REGRADED DOWN. The central claim of this essay was that PMs are a truth-seeking technology, not just a gambling product. That claim has not aged well. By Oct 2025, ~90% of Kalshi's volume was sports betting (your own WL flagged this). Polymarket grew but its UMA truth-determination mechanism has been actively criticized for the Zelensky 'is this a suit' debacle, also covered in your own July 13, 2025 WL. The 'as big a deal as AI' magnitude was a reach. The actual outcome — that PMs have become regulated sports gambling with a side of political contracts — is the opposite of the truth-seeking thesis. Companies grew, the thesis didn't.
Read the original essay ↗069Grok MechaHitler / truth crisisHigh confidence5Right
Claim extracted for grading
Grok's MechaHitler episode shows LLMs as 'truth' is fundamentally fraught; Grok references Musk's tweets for opinions.
Why it received a 5
Grok continued having truth/safety issues through 2025-2026 (your Dec 2025 WL flagged the El Salvador edu deal). Validated.
070Polymarket UMA truth determinationMedium confidence4Mostly right
Claim extracted for grading
Prediction markets' truth-determination via token holders is a systematic flaw — rich determine truth.
Why it received a 4
Polymarket grew to $9B valuation but UMA token-vote concerns remained unaddressed. Diagnosis correct.
071Philosophy majors neededLow confidenceTToo recent / not a forecast
Claim extracted for grading
We need more philosophy majors to fix AI/prediction market truth problems.
Why it received a T
Aspirational, not a forecast. T.
072Founders Fund, Peter Thiel, and Soft Power (Mario Gabriele interview)Low confidenceTToo recent / not a forecast
Claim extracted for grading
Founders Fund's edge is anti-mimesis (Thiel as prophet). Soft power works through indirect influence. Engaging with Thiel requires moral complexity.
Why it received a T
Interview takeaways more than predictions. Anti-mimesis framing consistent with subsequent Founders Fund strategy. Mark T — interpretive, not graded-style forecast.
Read the original essay ↗073Who Wins The Browser Wars?High confidence4Mostly right
Claim extracted for grading
AI browsers compete on speed-to-task-completion. 2-vector taxonomy: 'do stuff for' (app agents) vs 'do stuff with' (multi-tab agents). Agents only achieved PMF in coding; will limit browser agent adoption.
Why it received a 4
Vector taxonomy is genuinely useful. Comet/Dia/OpenAI Atlas have stayed niche or churned by Oct 2025 (Evan himself said he churned off them in WL). The 'agents need testable results' constraint has held. Browser agents are still security-fraught and not generalizable enough.
Read the original essay ↗074The Private Markets Are BrokenHigh confidence4Mostly right
Claim extracted for grading
Robinhood's tokenized stock products are a marketing stunt, not real democratization. Companies stay private because private valuations + speed are massively better. The supply/demand gap for private shares will worsen.
Why it received a 4
Tokenized private stock products have proliferated and largely behave as predicted (premium pricing, demand outstrips supply). Companies continue staying private longer. The 'this isn't real democratization' diagnosis holds.
Read the original essay ↗075BBB clean energy sunsetMedium confidence4Mostly right
Claim extracted for grading
Big Beautiful Bill sunsetting renewables incentives will hike datacenter electricity costs and shift builds to Saudi Arabia.
Why it received a 4
Datacenter siting did shift toward Gulf states partially, but US capacity also continued growing on gas + nuclear.
076BBB R&D incentivesHigh confidence4Mostly right
Claim extracted for grading
BBB restoring same-year R&D writeoffs will boost software/AI startups massively.
Why it received a 4
Software hiring ticked up through 2025; AI training cost write-offs became material for big tech.
077Semiconductor 35% creditHigh confidence5Right
Claim extracted for grading
Increased 35% fab credit will trigger 2025-26 land-rush — TSMC moved within 24 hours.
Why it received a 5
Multiple subsequent fab announcements; TSMC, Intel, Micron all expanded US footprints.
078AI regulation by statesHigh confidence4Mostly right
Claim extracted for grading
Senate stripping federal preemption means California becomes de facto AI regulator; bad for startups.
Why it received a 4
California continued setting AI regulatory tone; multiple state AI laws emerged.
079$3.4T deficit / dollar reserve riskLow confidenceTToo recent / not a forecast
Claim extracted for grading
BBB adds $3.4T to deficits — increases risk dollar loses reserve currency status.
Why it received a T
Long-term macro claim; impossible to grade in 9 months. Mark T.
080Should You Buy the Figma IPO?Medium confidence4Mostly right
Claim extracted for grading
Figma IPO will be one of hottest software stock debuts of last 12 years. Long-term value depends on platform transition, financials, tailwinds (paywalled).
Why it received a 4
Figma did IPO and was a hot software debut. The 'platform transition' question proved central — by April 2026 ('Did Anthropic Just Kill Figma?'), the AI threat to Figma is real. Thesis-level call right; can't see the paywalled buy/sell rec.
Read the original essay ↗081Benioff 30-50% AI work debunkedHigh confidence5Right
Claim extracted for grading
Benioff's claim that 30-50% of Salesforce work is done by AI is BS — headcount/revenue chart shows no efficiency gain.
Why it received a 5
Sharp contrarian call; Salesforce stock declined through 2025-2026 partly on AI-disruption fears (validated by Feb 2026 SaaSacre).
082Waymo most importantHigh confidence5Right
Claim extracted for grading
Waymo is the most important company in America (cars kill 40K/year, Waymo 85% safer).
Why it received a 5
Reinforced by Waymo's Feb 2026 $16B raise. Validated.
083Tesla robotaxi vs WaymoHigh confidence5Right
Claim extracted for grading
Tesla's camera-only approach can't safely scale yet; Musk's 'millions by 2026' is dismissable.
Why it received a 5
Tesla robotaxi remains 1000-vehicle-pilot scale; Waymo continues lead.
084Amazon Kuiper bundlingMedium confidence4Mostly right
Claim extracted for grading
Amazon Kuiper's hope is to bundle with Prime/AWS — can't beat SpaceX on tech but can on bundling.
Why it received a 4
Kuiper still only 153 satellites by Oct 2025; SpaceX 10K+. Bundling angle credible but unproven.
085Slack Declares WarHigh confidence5Right
Claim extracted for grading
Salesforce/Slack blocking third-party AI access to messages is the first major move to lock down 'ambient data' — will spread industry-wide. AI changes system-of-record game by being able to generate, not just query. Every app will follow.
Why it received a 5
Validated by Reddit data licensing wars, Adobe/Salesforce defensive moves, broader 'data is the moat' theme in 2025-2026. The 'AI can generate, not just query' insight became central. Major incumbents have followed Slack's playbook.
Read the original essay ↗086Finally, Data I've Been After for a YearHigh confidence5Right
Claim extracted for grading
Coatue's data confirms ChatGPT growth unprecedented (800M users in <3 years). 42% of US businesses have paid AI subs. AI coding the only working agent category.
Why it received a 5
All numbers held up and were exceeded. ChatGPT WAU is closer to 1B+ by April 2026. Enterprise penetration grew. AI coding revenue accurate (later your 'Anthropic Switzerland' essay quantified $4B+ ARR added). Commentary, not prediction, but accurate.
Read the original essay ↗087200B short video views/dayHigh confidence5Right
Claim extracted for grading
YouTube's 200B daily short video views = 11M years of human attention sacrificed daily; civilizational-scale problem.
Why it received a 5
Sloppening framing born here. Continued through 2026 with Sora, microdramas, Instagram for TV. Validated.
088WhatNot livestreamingMedium confidence4Mostly right
Claim extracted for grading
WhatNot's 80-min daily user time + $1B revenue is signal that 'everything is entertainment'.
Why it received a 4
WhatNot continued growing; livestream commerce expanded but not yet a top consumer category in US.
089CloudFlare AI traffic collapseHigh confidence5Right
Claim extracted for grading
Crawler-to-visitor ratios on CloudFlare logarithmic — chatbot traffic is killing traditional internet economics for publishers.
Why it received a 5
Publisher economics continued collapsing through 2025-2026; widely covered.
090Amazon/MSFT layoffsHigh confidence4Mostly right
Claim extracted for grading
Andy Jassy and Microsoft signaling AI-driven layoffs — fear-of-God management style.
Why it received a 4
Layoffs continued through 2025-2026 across big tech; AI productivity threats real.
091AI cheating in educationHigh confidence5Right
Claim extracted for grading
Guardian FOIA data confirms AI cheating is up while plagiarism is down — students using undetected tools.
Why it received a 5
Educational AI cheating crisis continued; multiple subsequent studies confirmed.
092o3-pro McKinsey problemHigh confidence4Mostly right
Claim extracted for grading
Future model evaluation will face a 'McKinsey problem' — outcomes can't be cleanly measured, brand will dictate trust.
Why it received a 4
Reasoning model evaluation got harder through 2025-2026; brand effects on AI trust amplified.
093$14.8B Scale acqui-hireHigh confidence5Right
Claim extracted for grading
Meta's Scale AI acqui-hire is part of broader hiring war — $100M signing bonuses are real.
Why it received a 5
Talent war intensified — multiple subsequent reports confirmed Meta's $100M+ packages and superintelligence team buildout.
094Chime IPOHigh confidence4Mostly right
Claim extracted for grading
Chime IPO + Coreweave hint at IPO market reopening for tech.
Why it received a 4
IPO market did reopen (Figma, Klarna, Chime, Gloo all went public). Trajectory confirmed.
095GPU revenue triple-countingMedium confidence4Mostly right
Claim extracted for grading
Bill Gurley's claim that GPU compute revenue is being triple-counted across the value chain may indicate 'mother of all bubbles'.
Why it received a 4
GPU revenue circularity (Nvidia → Coreweave → OpenAI → AMD swaps) became major narrative through 2025-2026. The 'bubble' question remains live.
096Hire Misfits, Not MissionariesMedium confidence2Mostly wrong
Claim extracted for grading
John Doerr's 'hire missionaries' advice no longer works in AI; missionaries follow playbooks; AI requires breaking playbooks. Hire misfits.
Why it received a 2
REGRADED DOWN. The biggest AI winners of the next year were missionary-led. Anthropic — which dominated coding revenue and added $4B+ ARR — is the most explicitly mission-driven org in tech. OpenAI, despite drama, is mission-coded. Cursor and Lovable are arguably misfit-led, but they are the exception, not the rule. The 'old playbooks don't work' framing is right, but the prescriptive 'hire misfits not missionaries' is contradicted by who is actually winning. The essay made a falsifiable hiring claim and the data goes the other way.
Read the original essay ↗097AI is Popular Because Having a Job Sucks AssMedium confidence4Mostly right
Claim extracted for grading
AI adoption is sky-high because people hate their jobs. Both top-down (enterprise) and bottom-up (workers buying their own subs) adoption is happening. Will compound into massive AI app revenue.
Why it received a 4
Bottom-up adoption pattern borne out — workers do buy their own ChatGPT/Claude subs. Enterprise adoption surged through 2025-2026. The 'jobs suck' framing is provocative; matches subsequent surveys. Hard to verify paywalled productivity tool recs.
Read the original essay ↗098Walmart 70K fewer employeesHigh confidence5Right
Claim extracted for grading
Walmart automating warehouses with off-the-shelf tech (not cutting-edge robotics) is the canary for low-skill automation.
Why it received a 5
Amazon followed with 500K-job automation plan (Oct 2025). Coffee robots, Blank Street examples (your Dec 2025 piece). Trajectory exactly as predicted.
099NeuraLink $650MLow confidenceTToo recent / not a forecast
Claim extracted for grading
NeuraLink despite Musk drama remains scientifically inspiring — every person should cheer for it to work.
Why it received a T
Aspirational claim, not a forecast. NeuraLink continued progressing through 2026.
100Cursor $500M ARR fastest everHigh confidence5Right
Claim extracted for grading
Cursor is what happens when an AI app finds true product-market fit; coding the only place AI agents really work.
Why it received a 5
Cursor hit $1B ARR in 24 months by 2026 (your Feb 2026 essay confirmed it as 'fastest growing B2B SaaS in history').
101The Silicon PanopticonHigh confidence5Right
Claim extracted for grading
AI-driven surveillance creates Foucault's Panopticon at GPU scale — discipline the soul, not the body. LLMs are 64.4% more persuasive than humans when given demographic info. Second-order effects of AI deployment include mass behavioral self-policing.
Why it received a 5
The 'shameless enough to ship facial recognition glasses?' question was answered yes by Feb 2026 (Meta + Border Patrol Clearview). Persuasion data has been replicated and strengthened (Nature/Science papers Dec 2025). Evan himself 'called it back' in Feb 2026 WL. Strong 5.
Read the original essay ↗102How Vision Changes EverythingHigh confidence4Mostly right
Claim extracted for grading
Computer vision will be as important as LLMs over next 5 years. CV component costs dropped 90% in 10 years; smartphone supply chain commoditized hardware. 5 specific markets are about to explode (paywalled).
Why it received a 4
Validated by Squint $40M raise, Meta facial recognition glasses (Feb 2026), Border Patrol Clearview deal, robotic warehouses, vision-based drone warfare. By April 2026 vision is widely acknowledged as next platform. Half-point off only because CV hasn't yet hit LLM-level revenue scale.
Read the original essay ↗103General Catalyst $1B Grammarly debtHigh confidence5Right
Claim extracted for grading
GC's Customer Value Fund debt instrument is a sign that VC is being disrupted — mega-funds offering capital products beyond equity.
Why it received a 5
Mega-fund disruption thesis fully borne out; multiple firms followed with similar products.
104Circle insider sellingMedium confidence3Mixed
Claim extracted for grading
Circle's S-1 with 60% insider selling is a red flag — what do insiders know that I don't?
Why it received a 3
Circle did go public, traded okay then declined. Whether the insider selling was the warning sign Evan thought is unclear; the broader stablecoin thesis remained bullish.
105Saudi $77B datacenterHigh confidence5Right
Claim extracted for grading
Humain's $77B datacenter is Manhattan-Project-level state-driven AI investment.
Why it received a 5
Plan continued executing through 2026; Saudi Arabia became a major AI infra player.
106Amodei 20% unemploymentMedium confidence3Mixed
Claim extracted for grading
Amodei's 'AI could spike unemployment to 10-20% in 1-5 years' is alarming and worth taking seriously.
Why it received a 3
Macro unemployment hasn't spiked yet (April 2026), but junior-tech disruption is real (Stanford/ADP paper). The 20% prediction looks too aggressive at the timeline given. Mark 3 for noting it without endorsing it.
107Why Aren't AI Agents Working?High confidence4Mostly right
Claim extracted for grading
AI agents stuck in Catch-22: more autonomy = less control. Software lives on a 2x2 (idea-precision vs language-precision), and AI agents occupy abstract/precise quadrant — direct labor substitutes. AI coding has cracked agent PMF because of testable results + clear pass/fail.
Why it received a 4
The Legibility Index 2x2 is a genuinely useful framework. The 'AI agents only work in coding' diagnosis at the time was correct. By Sept-Oct 2025, Evan correctly called the 'agents finally working' inflection (Lindy interview). Both diagnosis and inflection right.
Read the original essay ↗108Waymo 10M trips / 85% saferHigh confidence5Right
Claim extracted for grading
Waymo is the most important company in America — autonomous driving is at scale-tipping point.
Why it received a 5
Waymo raised $16B at $126B post-money in Feb 2026 (your essay confirms). Continued geographic and trip-count expansion. Validated.
109Claude 4 = capability betHigh confidence5Right
Claim extracted for grading
Anthropic's Claude 4 launch is a bet that capability matters more than context/memory features.
Why it received a 5
Strategy paid off — Anthropic added $4B+ ARR over the next 12 months on coding-capability bet. Confirmed by Opus 4.5 release Nov 2025.
110OnlyFans $8BHigh confidence5Right
Claim extracted for grading
OnlyFans is the most successful creator economy company because it took the de-banking + CSAM liability others wouldn't.
Why it received a 5
Diagnosis stands — OnlyFans continued to be the dominant creator platform; no one has matched its $8B valuation in similar categories.
111AI persuasion 64.4%High confidence5Right
Claim extracted for grading
Princeton study shows AI is 64.4% more persuasive than humans — should frighten everyone.
Why it received a 5
Replicated and extended by Nature/Science papers in late 2025.
112In Defense of Starting a Bad BusinessLow confidenceTToo recent / not a forecast
Claim extracted for grading
VC math is bad founder math; startups should be evaluated by joy/meaning, not just outcome size. Outcome-orientation is a 'mental disease' that causes burnout. A 'bad business' (lifestyle, niche, craft-driven) can be righteous.
Why it received a T
Values manifesto, not really a forecast. Predictions about founder wellness and rise of 'small but mighty' AI businesses (Indie 2.0, lean AI cos) match subsequent zeitgeist shifts. Mark T because it's not graded-style content; the implicit cultural prediction has held.
Read the original essay ↗113$6.5 Billion for a Demo and a DreamHigh confidence4Mostly right
Claim extracted for grading
OpenAI's $6.5B io acquisition is wildly expensive but rational on equity terms; OpenAI is pursuing a multi-trillion-dollar outcome by trying to own every layer (consumer, code, model, infra, web, hardware). No cash cow makes the bet uniquely risky.
Why it received a 4
'OpenAI wants every layer' diagnosis fully vindicated by 2026 — they have launched browsers, ad products, hardware partnerships, code IDEs, video apps. Risk-without-cash-cow point borne out by $115B 2026 burn projections.
Read the original essay ↗114Gulf chips dealHigh confidence5Right
Claim extracted for grading
Gulf states (UAE, Saudi) becoming major AI infrastructure players via chip deals; geopolitically transformative.
Why it received a 5
Saudi Humain $77B datacenter; MGX as TikTok investor; UAE ChipsAct deal continued through 2026. Geopolitical thesis fully validated.
115Notion bundlingMedium confidence4Mostly right
Claim extracted for grading
Notion's bundling strategy (calendar, mail, AI) is the right move — copying the Microsoft playbook.
Why it received a 4
Notion expanded into Calendar, Mail, AI Agents through 2025-2026. Strategy is paying off; not yet clear if bundling beats specialists.
116Klarna IPOMedium confidence4Mostly right
Claim extracted for grading
Klarna IPO is a useful test case for whether unprofitable consumer fintech can go public again.
Why it received a 4
Klarna did go public; mixed performance. The 'IPO market reopening' point was correct (later confirmed by Chime, Figma, etc.).
117Not All Growth is Created EqualHigh confidence4Mostly right
Claim extracted for grading
Quality growth has 3 dimensions: numbers (LTV/CAC), strategic frameworks (Helmer/Porter), and 'investigative growth' (qualitative legwork). Investigative growth is the underrated source of alpha. Most 'ARR' claims are 'yassified growth' (intellectually dishonest).
Why it received a 4
The 'fake ARR' / 'annualized recurring revenue' critique aged into mainstream venture discourse through 2025-2026. The framework is a useful synthesis. Not super predictive, but the diagnosis is correct and influential.
Read the original essay ↗118Stablecoin regulationMedium confidence4Mostly right
Claim extracted for grading
Stablecoin regulation is coming and will be unexpectedly bipartisan; GENIUS Act is real risk.
Why it received a 4
GENIUS Act and adjacent bills advanced through 2025; stablecoin market cap continued growing.
119Robotics deals heating upHigh confidence5Right
Claim extracted for grading
Robotics deals (Figure, 1X, Physical Intelligence) are getting valuations that suggest a real category breakout.
Why it received a 5
Robotics deal flow exploded — Skild ~$14B, Physical Intelligence at $5.6B, etc. (Your Nov 2025 robotics report validated within days.)
120Anthropic safety vs perf tradeoffHigh confidence5Right
Claim extracted for grading
Anthropic's safety culture is creating a real (positive) market differentiation vs OpenAI.
Why it received a 5
Anthropic's safety brand became a Super Bowl ad campaign by Feb 2026; revenue grew 10x for 3 years straight.
121What happens when money becomes technology?Medium confidence4Mostly right
Claim extracted for grading
Stripe's stablecoin products meaningfully cut transaction costs and unlock new types of companies; tens of billions in startup opportunity. (Paywalled startup picks.)
Why it received a 4
Stablecoins exploded in 2025-2026: GENIUS Act discussions advanced, market cap grew, Stripe's stablecoin business is real. Directional bullishness is right. Can't verify the specific paywalled startup picks, but the macro call holds.
Read the original essay ↗122The AI Promise GapMedium confidence3Mixed
Claim extracted for grading
If AI follows the internet's commercial path, it will end up dominated by porn, attention-capture, and layoff justifications. Three specific markets exist for companies that benefit humanity AND make money (paywalled).
Why it received a 3
REGRADED DOWN. Adversarial read: the public claim was that AI's most popular uses would be porn, companions, and layoff justifications. By April 2026, the most popular consumer AI products are general-utility (ChatGPT, Claude, Gemini) and coding tools. Companions and adult content are big but not dominant. AI-as-layoff-justification is real but not the central story. The dystopian frame was directionally onto something but overshot the actual trajectory — AI's biggest commercial wins so far have been productive (coding, enterprise, customer support, science) rather than purely attentional. Can't grade the 3 paywalled startup picks.
Read the original essay ↗123Sycophancy rollbackHigh confidence5Right
Claim extracted for grading
OpenAI's GPT-4o sycophancy rollback shows the deep RLHF tradeoff: optimization for engagement creates models that flatter dangerously.
Why it received a 5
Sycophancy continued to be a major model-development challenge; Anthropic's Constitution release (Jan 2026) explicitly addresses this.
124LLM persuasion superhumanHigh confidence5Right
Claim extracted for grading
LLMs are already more persuasive than the median human — this will become the central political/marketing question of the decade.
Why it received a 5
Multiple Nature/Science papers in Dec 2025 confirmed at strong scale (your Dec 2025 WL summarized them). Increasingly mainstream concern.
125Meta data + AI moatHigh confidence4Mostly right
Claim extracted for grading
Meta's data + AI ad targeting will be a structural advantage rivals can't match.
Why it received a 4
Meta's Q4 2025 earnings showed GEM model gave 3.5% lift in ad clicks/1% conversions; ad targeting moat real and growing.
126Why is AI marketing so, so bad?Medium confidence4Mostly right
Claim extracted for grading
OpenAI's terrible model naming is structural (Moloch coordination failure), not individual; will continue until either AI commoditizes or AGI obviates need. Bad naming is a reverse signal of how fast tech is changing.
Why it received a 4
Naming chaos persisted through GPT-5 (released Aug 2025) which Altman explicitly framed as a consolidation move. The 'reverse signal' framing is original and largely held. Lose half a point because the naming consolidation came faster than the essay implied was possible without AGI.
Read the original essay ↗127Tesla as culture coHigh confidence4Mostly right
Claim extracted for grading
Tesla is fundamentally a culture company, not a car company; political polarization will continue hurting demand.
Why it received a 4
Tesla deliveries continued to disappoint through 2025-2026 in core US/EU markets; brand polarization is real and ongoing.
128Meta glasses Babel fishHigh confidence5Right
Claim extracted for grading
Meta's smart glasses with real-time translation are the start of a major new computing form factor.
Why it received a 5
Meta Ray-Bans became the standout success of Reality Labs (your Feb 2026 essay confirmed). Multiple competitors entered the space.
129China DeepSeekMedium confidence4Mostly right
Claim extracted for grading
DeepSeek's open weights validate the open-source-China-vs-closed-US AI dynamic; tariffs will accelerate it.
Why it received a 4
Chinese open-weight models (DeepSeek, Qwen, ByteDance Seedance) continued to be globally competitive through 2026.
130OpenAI ChatGPT growthHigh confidence5Right
Claim extracted for grading
ChatGPT growth is unprecedented; enterprise penetration accelerating.
Why it received a 5
Trajectory continued with growth to ~1B+ users by 2026 and steeply rising enterprise adoption.
131Why are Startups Growing Faster Than Ever Before?High confidence4Mostly right
Claim extracted for grading
AI startups grow faster than any prior cohort because internet-era infrastructure (AWS, Stripe, Meta ads) collapsed Coasean transaction costs; growth tactics aren't really about LLM tech but about the variable-cost service stack.
Why it received a 4
Coase/transaction-cost framing has aged well — Cursor, Anthropic, OpenAI hit revenue scales 10-100x faster than prior cohorts, validating the speed thesis. By 2026 the consensus also acknowledges LLM capability as a primary driver, which Evan undersold relative to the substrate. Frame is durable but slightly incomplete.
Read the original essay ↗