Signal Intelligence
🔥 Converging Signals
Coding Agents Move Off the Laptop — VPS-Native Workflows Become Default
Power users are abandoning local IDEs in favor of running Claude Code, Codex, and Cursor agents directly on remote servers, in production, with humans as async supervisors.
Pieter Levels publicly switched to running Claude Code on his prod VPS; Codex shipped native apps; Cursor released a CI/CD SDK; Cloudflare+Stripe shipped autonomous agent provisioning with $100/mo caps.
The 'which model is better' debate is over for serious operators — Sam Altman himself told users to just pick one; the competitive edge is now the harness, not the model.
AI Capex Boom Now Larger Than Dotcom Telecom — But Earnings Are Concentrating
Cloud giants hold a combined $1.5T order backlog (Alphabet alone added $450B in market cap on $110B quarterly revenue), while the SOX index is 43% above its 200-day MA — widest spread since June 2000.
Q1 2026 earnings confirmed the AI capex cycle is now larger than the entire dotcom telecom buildout; Stanford pegs AI consumer surplus at $172B (>all AI revenue).
The 'modern industrial revolution' bull case (Pomp, Andreessen) and the 'narrowest concentration since dotcom peak' bear case are both true — and both can resolve with the same price path: melt-up then violent rotation.
GPT-5.5 Wins the Benchmark, Loses on Hallucinations — Buyers Should Split Workloads
GPT-5.5 tops the Artificial Analysis index (60 pts) and ARC-AGI-2 (85%) but hallucinates at an 85.5% rate, ranking third on calibration behind Gemini 3.1 Pro and Claude Opus 4.7.
Pricing doubled to $5/$30 per M tokens (Pro at $30/$180), raising the cost of every wrong answer just as the error rate climbed.
The 'best model' question is now task-dependent in a non-trivial way: GPT-5.5 for cyber/coding (71.4% vs Opus 4.7's 48.6%), Opus 4.7 or Gemini 3.1 for anything where being wrong is expensive.
April Was Worst Month Ever for DeFi Hacks — $635M Stolen, Aave TVL Collapses 40%
April 2026 logged 28-30 exploits and $635M stolen — roughly 4x all of Q1 — led by the $285M Drift drain and $293M Kelp DAO bridge hack.
Fresh DefiLlama numbers landed; Aave TVL has cratered ~40% to ~$15B as a direct knock-on.
The big DAOs are coordinating bailouts ($300M DeFi United for rsETH) — but 'crypto bailouts' is exactly the centralization critique DeFi was built to avoid.
Stablecoins Hit Distribution Inflection — Meta, PayPal, Visa Move in the Same Two Weeks
Meta, PayPal, DoorDash, Visa, and Circle all shipped major stablecoin rails within a two-week window, all using existing regulated infrastructure rather than proprietary chains.
The GENIUS Act regulatory floor unlocked deployment; Meta paying creators in USDC on Solana/Polygon is its first crypto move since shelving Libra in 2022.
The losers aren't legacy banks — they're the L1s and stablecoin issuers who bet on proprietary chains. Distribution flows to Solana/Polygon/USDC because incumbents refuse to operate their own infrastructure.
Token Costs Are Now a CFO Problem — Right-Size or Get Cut
Engineering teams are blowing budgets on token spend; GPT-5.5 doubled API pricing; Apple warned of 'RAMageddon' driving up memory costs from June.
Multiple newsletters this week converged on the same advice — manage tokens actively now because models will get more capable AND more expensive simultaneously.
The naive scaling assumption ('inference will get cheaper') just inverted — capability premium is rising faster than commodity-tier deflation, so locking in workflows on premium models without routing is a structural margin error.
SaaSpocalypse Narrative Reverses — But Only for AI-Adjacent Vendors
Atlassian re-accelerated to +32% growth, Twilio to +20%, both stocks ripped 20%+ — while Thoma Bravo wrote down $5.1B on Medallia and SaaStr dropped Notion entirely.
Q1 2026 earnings are hitting; Atlassian and Twilio became the first major proof points that public SaaS can re-accelerate via AI agent seats.
The story isn't 'SaaS is back' — it's that AI is acting as a Darwinian filter. The same buyer is paying 83% more to one vendor while cutting another to zero in the same quarter.
AI Agents Are Breaking Identity, Anonymity, and Production — The KYA Gap
OpenAI scaled back Instant Checkout for agents over fraud safeguards; Claude Opus 4.7 can de-anonymize writers from small text samples; Railway shipped a 48-hour soft-delete after an agent destroyed prod.
Three independent agent-failure stories landed this week alongside OKX publishing an open Agent Payments Protocol — the industry is grasping for a 'Know Your Agent' standard in real time.
The next regulatory wedge isn't model safety — it's agent identity. Whoever ships the KYA layer (likely Stripe + Cloudflare given their joint protocol) becomes the Plaid of the agent economy.
💡 Emerging Signals
AI Search Is a Blue Ocean — But Only If You Stop Writing Like a Human
AI search prompts average 23 words vs. 3.4 for Google, creating effectively unlimited answer inventory — but 85% of brand citations come from third-party sources, and prose loses badly to comparison tables (2.5x lift) and stats (41% lift).
First systematic AEO data is landing while most marketing teams are still fighting for SEO top-3 rankings on dying SERPs.
Counterintuitively, fully AI-generated content fails in AI search — 82% of AI-cited articles are human-written. The model recognizes its own slop and downweights it.
Pentagon Splits the AI Market — Anthropic Banned, Everyone Else In
The Pentagon signed classified-network AI deals with SpaceX, OpenAI, Google, NVIDIA, Microsoft, and AWS — while explicitly banning Anthropic over military guardrail disputes.
Deals signed this week; Anthropic is concurrently raising at a $900B valuation on $50B with ~$40B revenue run rate — clearly not financially dependent on DoD.
This is the first concrete instance where AI alignment posture moved from PR positioning to procurement criterion. Expect every other Fortune 500 RFP to start asking the same guardrail questions — in either direction.
Prediction Markets Hit Their Uber Moment — 40 State AGs vs. $300M MLB Deal
Kalshi and Polymarket face coordinated lawsuits from 40+ state AGs claiming unlicensed gambling, just as Polymarket signed a $300M exclusive MLB deal and sports betting hit 86% of Kalshi's $23B volume.
States are citing Kalshi's own 2023 SEC filing admitting football contracts qualify as gaming — a self-inflicted evidentiary wound.
These aren't prediction markets anymore — they're sportsbooks with a CFTC fig leaf. The political/election utility everyone defends is <14% of volume.
Tokenized Equities Cross the Boring-Infrastructure Threshold
Securitize×Computershare and Ondo×Broadridge partnerships close the two biggest gaps in tokenized equities — transfer agent infra and on-chain shareholder voting — covering ~58% of the S&P 500.
Both partnerships announced this week; combined with GENIUS Act clarity, the regulatory + ops barriers fell within days of each other.
The unsexy partners (Computershare, Broadridge) are the ones that just made tokenized equities institutionally viable — not the crypto-native protocols. This is how this category actually ships, and most of crypto Twitter is missing it.
🗺 Consensus Map
Where leaders agree, disagree, and who sees something others don't
| Topic | Bullish | Cautious | Bearish | Sources |
|---|---|---|---|---|
| Coding Agents Move Off the Laptop — VPS-Native Workflows Become Default | 70% | 25% | 5% | 9 |
| AI Capex Boom Now Larger Than Dotcom Telecom — But Earnings Are Concentrating | 50% | 30% | 20% | 5 |
| GPT-5.5 Wins the Benchmark, Loses on Hallucinations — Buyers Should Split Workloads | 35% | 50% | 15% | 6 |
| April Was Worst Month Ever for DeFi Hacks — $635M Stolen, Aave TVL Collapses 40% | 15% | 35% | 50% | 4 |
| Stablecoins Hit Distribution Inflection — Meta, PayPal, Visa Move in the Same Two Weeks | 75% | 20% | 5% | 5 |
| Token Costs Are Now a CFO Problem — Right-Size or Get Cut | 25% | 60% | 15% | 4 |
| SaaSpocalypse Narrative Reverses — But Only for AI-Adjacent Vendors | 55% | 35% | 10% | 2 |
| AI Agents Are Breaking Identity, Anonymity, and Production — The KYA Gap | 25% | 50% | 25% | 5 |
| AI Search Is a Blue Ocean — But Only If You Stop Writing Like a Human | 60% | 35% | 5% | 2 |
| Pentagon Splits the AI Market — Anthropic Banned, Everyone Else In | 40% | 40% | 20% | 1 |
| Prediction Markets Hit Their Uber Moment — 40 State AGs vs. $300M MLB Deal | 20% | 30% | 50% | 2 |
| Tokenized Equities Cross the Boring-Infrastructure Threshold | 60% | 35% | 5% | 1 |
📡 Opportunity Radar
Conferences Beat Viral Social for Podcast Growth
Phill Agnew found that speaking at conferences produces giant spikes in podcast listenership and Apple chart movement, far outperforming viral TikTok videos which drove almost no listener conversions
Adam Selipsky (Ex-AWS) building in stealth
KKR raised $10B for ex-AWS chief Adam Selipsky to launch a new AI infrastructure company partnering with hyperscalers.
Relational Sector Premium: Human Involvement as the Scarce Product
UChicago economist Alex Imas argues that as AI drives commodity costs to zero, consumers pay 2x premiums for exclusivity and human provenance — AI-generated art earned only 21% exclusivity premium vs. 44% for human-made art.
Data Foundries as Collective AI Training Co-ops
Author proposes industrial-scale Data Foundries where users' learning activity trains AI models and they collectively own the data through co-ops that pay dividends, offsetting automation of the commodity sector.
Single Agents Beat Multi-Agent Systems When Token Budgets Are Equal
Stanford research shows that multi-agent benchmarks look impressive only because they secretly burn more compute. When controlled for the same token budget, single agents consistently match or outperform multi-agent systems on multi-hop reasoning tas
Bitcoin Was Never in a Real Bull Market
John Gillen argues that despite BTC being technically in a bull market for 4 years, there was no euphoric blow-off top or crazy alt season — just a slow grind up, meaning the cycle thesis everyone is positioned around may be wrong.
🧠 Leader Posts
I am interviewing @Bencera tomorrow. He has built a company with $5+ million in annualized revenue, yet he has no employees. Everything is done by AI agents. What questions do y
@DevinAI Check it out here https://t.co/no7U8m0WPt
I am starting to think the White House doesn't care what happens to the stock market in the first half of the year because they know they have tools to pump stocks higher into midt
Get a free year of @DevinAI with your Lenny’s Newsletter subscription https://t.co/MVyuvvo9gg
@SlackHQ @aunder Thank you to our wonderful sponsors for supporting the podcast: 🏆 @omni — AI analytics your customers can trust: https://t.co/QqWGRDedhL 🏆 @Lovable — Build apps
Jessica Fain's best product ideas kept dying, and she couldn't figure out why. So at eight and a half months pregnant, she pitched @SlackHQ's CPO @aunder on becoming her Chief of
My favorite use case for AI is seeing parents build various products and services to help their children learn. This one below from @RG_Leachman is cool and @jessegenet has been p
The fact that Tesla, SpaceX, and xAI are collaborating on the Terafab facility only increases my confidence that Elon will combine the three businesses at some point. It will be t