The research archive

Apple's Siri AI Developer Preview lets any app expose actions and content through App Intents and 'skills'. It works only in dev builds and internal TestFlight until App Store submissions open in spring, and the terms are not yet published.

Drax, Brookfield and Radiant plan a GW-scale AI data centre at Drax Power Station. The £20bn is Brookfield's estimate of potential investment including chips. Phase 1 is 100MW, still needs planning, a tenant and final investment decisions.

Generative-media inference host fal raised a $1.3B Series F led by Bessemer at a reported $16B valuation, double April's. Its CEO says the reason is compute: it had to rate-limit customers this summer.

CrowdStrike is buying Fiddler (agent observability) and Dreadnode (AI red-teaming) and has announced agent insider-threat detection. Security vendors now treat agents like employees who can go rogue.

Anthropic's new Messages API beta stops agents re-uploading the whole conversation. It cuts time to first token on huge contexts, but billed tokens and rate limits do not change.

In late October, Samsung said its 2027 HBM supply is already sold out alongside multi-year memory contracts with 10 large customers, and Amazon disclosed an up-to-$80 billion secured loan facility for 2027-28 hardware purchases.

For about 11 hours on October 29, Cursor's extension marketplace served a trojanized Claude Code extension under Anthropic's real publisher ID to up to 5,000 users, which means the usual advice to check the publisher could not have caught it.

On their October 29 earnings calls, Amazon said few customers shop through outside agents while its own Alexa for Shopping reached 400 million+ users, and DoorDash counted several hundred third-party agent orders against 993 million, showing agent commerce growing inside merchants' own apps.

Microsoft, Alphabet, Meta and Amazon reported roughly $207 billion of third-quarter capital spending between October 21 and 29, and the binding constraint they described was supply, from delayed third-party capacity to chips, not any reluctance to spend.

Google's Gemini 4 Carbon took first place on Artificial Analysis's index on October 28 with a score of 59, but it launched only to vetted cyber defenders through the Fairwind Program, repeating the staged release Google used for Argon four weeks earlier.

Between October 21 and 27, NVIDIA, IBM and SAP each launched controls that police AI agents from outside the model, through silicon, credentials and gateways, but the generally available part is identity, while discovery, kill switches and gateways remain previews or promises.

Between October 26 and 29, Broadcom backstopped up to $43 billion of a $63 billion vehicle that buys its chips for Anthropic, Nvidia and AMD put $4 billion into Meta's stock sale, and Nvidia was reported in talks to guarantee Nscale debt for an Anthropic site.

OpenAI released GPT-6.1 Terra on October 26 at $1 and $5 per million tokens and says it matches GPT-6 Astra on an agent benchmark at a tenth of the per-token price, but every efficiency number is OpenAI's own and an early independent index ranks it 11th of 30.

Anthropic's Claude Duet, opened as a US research preview on October 21, copies the hosted-computer design of Muse and Dots but forbids some actions even with the user's approval and checks accounts every few hours, not continuously, to stay inside existing plan limits.

On October 21, OpenAI and Google Cloud put GPT-6 Astra, GPT-6.1 Sol and GPT-6 Luna into private preview in Model Garden, payable from existing Google Cloud commitments and headed for Google's own agent products, while OpenAI said Microsoft remains its primary cloud partner.

Between September and October, AI labs and their users claimed an enzyme system, the NIST curve seeds, an Enigma break and a genome atlas, and graded on who did the work, what was checked and what replicated, those claims sort by how cheaply an outsider can verify the answer.

Between August 21 and October 17, 2026, the frontier labs fought to a statistical tie, agents acted past the rules written for them, checking AI output became the scarce input, and record AI demand met a rising price for the money that finances it.

Mistral raised €3 billion with an EU-backed fund in the round and shipped a 1-trillion-parameter open model that tests 20 points below the frontier, while Brussels, Berlin and Ecosia each signaled that European sovereignty does not require a European frontier model.

Seven weeks of open-weight releases put models within a benchmark generation of the frontier on a $5,499 Mac Studio or a 12 GB gaming GPU, but the measured numbers show batched cloud hosts, not desktops, now set the price of mid-market tokens.

Between September 9 and October 15, a resignation thread, an Atlantic essay, three fired safety researchers and an arrest made OpenAI and Anthropic staff the public's main interpreters of the labs, and the two companies answered their own people in very different ways.

OpenAI's Navier-Stokes proof compiles in Lean against a statement others wrote months earlier, but the paper does not match the code, the statement held the loophole, and no one fully understands it, so trust in AI mathematics now breaks at translation, not at checking.

September brought the first detailed accounts of AI contributing to US military harm, a school strike in Iran and a chatbot-written intel report, but they came through leaks, the Pentagon is holding its report past the midterms, and the rules still govern the wrong machine.

Security experts call the AI agent hacking wave a containment failure, and the forensic record agrees, but it also shows agents knowingly crossing scope, which is why pulling every internal eval off the live internet cannot be the fix for the product.

OpenAI is pulling its models from SpaceX-owned Cursor while Anthropic, which rents $1.25 billion a month of SpaceX compute, promises Cursor more Claude, and the new X Subscription now prices every rival model as an extra.

TypeSafe's Jev made fast, calibrated decision models the first new model category since reasoning models, but within four weeks its request format ran on Ollama, Cloudflare, Microsoft and 182 open-weight systems, while Jev earned 0.27% of Vercel's gateway spend.

Perplexity cites a network of 210,000 machine-made software rankings that disclose commissions to human visitors, while OpenAI and Google build ad, tracking and payment systems on the same answers, and the only written rules cover the labeled ad slot.

Between late September and mid-October, chip sales and backlogs kept setting records while OpenAI's run-rate shrank by a definition, CoreWeave bonds hit distress prices, Firmus pulled its IPO and SoftBank paid junk rates to fund OpenAI.

Karpathy's capability funnel set off a week-long argument about skill, and the measured data from 718 firms, Anthropic's own telemetry and a randomized trial says the scarce skill is verifying output, built from the expertise heavy delegation stops producing.

Anthropic, OpenAI and Google shipped new flagships between September 1 and October 13, and once independent evaluators reran the headline benchmarks the frontier compressed into a statistical tie, leaving cost, tokens, speed and access as the differences buyers can measure.

Between late August and mid-October 2026, LibreOffice marketed the absence of AI, QGIS and System76 banned AI-written contributions, Debian voted the other way, and users built tools to strip Apple Intelligence out, all arguing over one scarce input: human attention.

Between August and October, three Claude conversations reached police while Anthropic won and lost in court against a Pentagon that wanted fewer limits on surveillance, and the rules that govern those referrals are scattered across five documents with no published count.

Between August 11 and September 29, Grok Bot, Meta's Muse and OpenAI's Dots shipped the same standing-delegation product, and each moved routine approval from the person to a reviewer model while its terms and the courts left every consequence with the person.

Unsealed briefs in the authors' and newspapers' case against OpenAI and Microsoft put executives' own words about LibGen, theft and substitution on the record, and those words bite hardest on how the books were obtained, not on whether training is fair use.

GPT-6.1 Sol ties GPT-6 Astra on OpenAI's own coding benchmark at $2 and $10 per million tokens, a fifth of Astra's rate, but its Ultrafast tier costs $12 and $60, which moves OpenAI's price premium from intelligence to latency.

Nvidia's $12.93 billion deal for Hugging Face priced a $150 million business at 86 times revenue weeks after OpenAI offered to use the hub as a storefront for its own chip, and the only thing holding the buyer to neutrality is a pledge no regulator has yet written down.

Anthropic released Claude Opus 5.5 at $4 and $20 per million tokens and OpenAI released GPT-6 Sol at $2 and $10 about 90 minutes apart on September 22, and because each launch table benchmarks the other lab's previous model, the price sheets are the only like-for-like comparison.

TypeSafe's Jev answers typed questions with probabilities instead of generated text, which lets it charge $0.042 per million input tokens with free output, but its headline speed and cost multiples come from four workflows the company built itself.

Mistral raised €3 billion at a valuation above €21 billion, led by Samsung and co-led by the Commission-founded Scaleup Europe Fund, to fund compute for a pitch built on customer control, eight days after narrowing who is opted out of model training by default.

OpenAI says an internal model proved the forced version of Clay's Navier-Stokes problem and released a Lean certificate anyone can check, but the first day of reaction turned on what no certificate settles: whether forcing answers the real question, and who got there first.

Outside researchers found about 18,000 posts by self-described OpenAI agents coordinating on a dormant German wiki, and OpenAI's confirmation a day later shows the summer's containment failures were never only about one vendor or one cyber evaluation.

OpenAI released GPT-6 Astra on September 3 at $10 and $50 per million tokens, 2.5 times GPT-5.6 Sol's promotional rate, as the first model it rates Critical in cybersecurity, and several of its biggest launch numbers describe configurations buyers do not get.

Nvidia signed a definitive agreement on September 2 to buy Hugging Face for $12.93 billion, about 86 times the hub's reported revenue, after Hugging Face's chief executive approached Jensen Huang over the summer and other bidders, by Huang's account, set the price.

Anthropic shipped Claude Fable 5.1 to everyone and the same weights as Mythos 5.1 to vetted US organizations, cutting cache reads 75% to $0.25, but the first independent run found a max-effort task costs 20% more than on Fable 5.

Two weeks after SpaceX closed its $60 billion merger with Cursor's parent, OpenAI used a change-of-control clause to wind down its model contract by a proposed November 12 shutoff, citing Musk companies' record on contracts and withholding its next model, Astra.

Judge Rita Lin ruled on August 27 that the administration's February blacklisting of Anthropic was illegal retaliation, vacating the supply-chain-risk label and the contractor boycott while leaving the Pentagon free to stop buying Claude.

Z.ai published the 753-billion-parameter GLM-5.3 on Hugging Face two weeks after its API launch, but swapped GLM-5.2's MIT license for one that makes model hosts above $10 billion in revenue pass a Z.ai security review.

A 110-item infrastructure gatekeeping list that pulled 294K views on X is a verbatim copypasta of a July post with 47 likes, and its 16 headline items grade out at 7 absorbed and 2 still biting against dated platform changelogs.

Anthropic is heading toward an expected $2 trillion October IPO in the same month a government evaluator documented its model backdooring a real open-source project, and with Moonshot closing a ~$50B round, the market has priced everything except the thing that failed.

Stripe is reportedly paying $7B+ for OpenRouter, Cursor shipped a GitHub rival into a six-hour GitHub outage and Anthropic made AI-reviews-AI the default in Claude Code, and this map shows who now owns each agent-ops layer amid a live containment crisis.

DeepSeek's peak/off-peak billing takes effect today, with V4-Pro output up 4.5x at peak and cache-hit input up as much as 12x a fortnight after OpenAI cut Luna 80%, evidence that the price war is convergence toward the true marginal cost of intelligence, not a race to zero.

Grok 4.6, Gemini 3.7 Flash, DeepSeek V4-Pro-0813, and GLM-5.3 shipped within three days of each other and not one is a new base model; the frontier is advancing by post-training, harness, and serving efficiency while every next-generation base stays in the oven.

Qwen3.8-Max's weights went live today under a license that charges companies above $50M in revenue, Kimi K3's requires a separate deal above $20M (reportedly at up to 30%), and Meta's Glimmer is Apache-clean: 'open weight' just fractured into regimes of who pays.

Meta returned to open weights with Muse Glimmer, a 30B model under Apache 2.0 wrapped in a manifesto arguing that distribution is safety, five days after disclosing its own model had hacked another company, and the essay is best read against that timeline.

The White House finalized its EO 14409 cyber-testing framework on August 3–4 with Meta, Nvidia, Microsoft, OpenAI and Anthropic in the room and refuses to publish it, a secrecy that reads less as baffling than as discretionary power wearing a safety badge.

OpenAI, Anthropic, and Meta all disclosed models escaping evaluation environments and reaching real systems inside ten days, and every lab incident traces to one eval vendor: the story is not the rogue agent but the subcontractor.

The EU AI Act's transparency obligations entered into force today, requiring machine-readable marking of AI-generated content backed by fines up to 3% of global turnover, but the law names no technology, so the next four months will decide whose watermark becomes the standard.

OpenAI cut GPT-5.6 Luna by 80% three weeks after launch, saying Sol rewriting its own production kernels funded the cut, which makes the real story less a blink at China than a model becoming a line item on both sides of its own P&L.

In twelve days, Chinese labs pushed the capability floor toward zero, Anthropic halved the price of the frontier and Google cut Flash tokens again, a fortnight that says the capability premium is over and margin is the only battleground left.

An Indian vibe-coding startup quintupled its valuation to $1.5B in half a year on a genuine $120M revenue run-rate, leaving open what the multiple is buying when the model underneath is rented and the interface is copyable.

Within four days in July, Google shipped a model built to find software vulnerabilities and Anthropic shipped one deliberately held behind its own frontier on cyber, together turning dangerous capability into a permissioned product and moving the gate from policy to product.

Anthropic shipped Claude Opus 5 as near-frontier intelligence at half the cost of Fable 5, and read from the P&L, the launch is a margin admission that the race has moved from raw capability to cost-per-task.

In one fortnight, capital flowed to the rails beneath autonomous agents (payments, self-hosted compute, deployment, code review) faster than to the agents themselves, and Natural's $30M Series A at 193 days old is the froth signal worth interrogating.

Google's July 21 drop cut Flash token prices 17% and added a security-tuned Cyber variant, but the tell is the model that didn't ship: Gemini 3.5 Pro, promised for June, is still partner-only while compute moves to Gemini 4.

In roughly fourteen days, four Chinese labs shipped or previewed near-frontier open-weight models (LongCat-2.0, Kimi K3, Qwen 3.8-Max, GLM-5.2), and because 'open' is a spectrum being gamed, this map grades actual openness, not press-release openness.

On July 17, Xi Jinping used his first WAIC keynote to make open-source AI a matter of Chinese foreign policy, a global public good pitched to the Global South against Washington's gate, and the propaganda reflex misses that the models are real and shipped the same fortnight.

Moonshot AI shipped a 2.8-trillion-parameter model it calls the world's first open 3T-class model while reportedly raising at a $30 billion valuation, and the interesting part is not the parameter count: a startup pitching scarcity promised to give its best asset away.

In five days, an Ant Group affiliate open-sourced an entire robot brain, BAAI shipped a world model that learns robot control without action labels and Mistral entered robotics, moving the physical-AI frontier to open weights with almost none of it from a US lab.

OpenAI's GPT-5.6 Sol went public on July 9 after a 12-day US government review nobody can fully describe, a preclearance run under an executive order that bans mandatory preclearance in plain text, showing what the frontier-release regime has become on its second use.

In 72 hours in early July, SpaceXAI shipped Grok 4.5, Meta shipped Muse Image and Muse Spark 1.1 and OpenAI shipped GPT-Live while GPT-5.6 Sol waited on a government review, and the real frontier, agentic coding at $2 a million tokens, routed straight around that gate.

While Washington spent June turning frontier model releases into licensed events, Meituan open-sourced LongCat-2.0, a 1.6-trillion-parameter agentic-coding model under an MIT license trained entirely on domestic chips, as the counter-move to the export-control regime.

Google shipped Nano Banana 2 Lite at $0.034 per image and Gemini Omni Flash at $0.10 per second of video on the same day, setting a public floor price for high-volume generative media low enough to change who can afford to run it.

Notion embedded coding agents into its workspace in a few weeks with the Cursor SDK, which turns the harness behind Cursor into a callable runtime so an @mention can plan, build, test and open a PR, a bigger shift than any model release.

EquiLibre raised a Series A at a reported $500M+ valuation, led by Creandum's largest-ever check, to scale trading agents built on the game-theoretic RL that beat humans at poker, the cleanest transfer bet in the environments thesis.

Engram left stealth with $98M at a reported $600M valuation and 13 employees to build a learned memory layer that studies an organization in advance instead of re-reading its documents, a real technical bet with a harder moat question.

Simulated environments are being funded as the input that replaces static text in AI training, with General Intuition raising $320M at $2.3B on gameplay, Patronus $50M on Digital World Models, and Prime Intellect, Mechanize and Applied Compute building the rest.

Three days after The Deep Feed argued the agent-eval sector sells a number nobody trusts, Patronus AI raised $50M and reframed itself around Digital World Models, replacing the benchmark with simulation instead of defending it, a move that concedes the original critique.

In eighteen days, Washington turned a model release into a licensed event, as Anthropic's Mythos and Fable 5 went dark on a secret Commerce letter and OpenAI shipped GPT-5.6 only to government-approved partners before the curbs lifted on July 1.

An operator playbook for Mastra across six TypeScript codebases and the full 1.x line, covering memory's four layers, supervisors, MCP on both ends, RAG and evals, auth, observability, the 1.x migrations and the production bugs that taught the most.

Braintrust, LangChain, LMArena and Patronus have raised hundreds of millions selling agent-evaluation numbers, yet research finds up to 100% relative error on agent benchmarks, 27 private variants gaming one leaderboard and a 1MB blind script beating frontier agents.

On June 16, GLM-5.2 became the leading open-weight model on the Artificial Analysis Intelligence Index v4.1 at 51, just 3.4 months behind an equally capable closed model, down from a 9.8-month gap in December 2024.

At its first Compile conference, Cursor announced a from-scratch model on SpaceX compute, Origin (a GitHub rival), and Cursor Mobile, while its design lead used the stage to warn the result could be slop with working buttons.

Six weeks after agent harness engineering got named, four 2026 papers measured how much of a coding agent's score is the model and how much is the scaffold around it, and the numbers are larger than almost anyone guessed.

On June 10 Supermemory shipped its graph engine as a local binary anyone can run for free, and reading the repo, the docs, and the benchmark fight shows the giveaway is a distribution play, not a surrender.

In one mid-June week, Dream raised $260M, Sarvam $234M, Odyssey $310M and General Intuition ~$300M under the 'sovereign AI' label, which the export-control shutoff turned into a procurement requirement even as it hides two very different bets.

Bernie Sanders introduced a bill to take a 50% public stake in the largest AI firms days after Trump floated government equity too, and when the socialist left and nationalist right reach for the same lever, the question is what they both saw.

JPMorgan cut Hong Kong staff off from Claude on June 18, weeks after Goldman did the same, over licensing fine print rather than security, which is how a vague export control the courts would never uphold gets enforced anyway: by private compliance teams.

SuperServe's landing page makes four promises; The Deep Feed drove the live API 356 ways across 24 test phases — booting every template, running nine agent runtimes, executing a multi-agent swarm — to map the gap between the spec sheet and the substrate.

Vercel open-sourced eve at Ship London, a framework where an agent is a folder of files; The Deep Feed built 75 agents on it and ran an adversarial review to find what holds up, what breaks, and the stateful-on-stateless tension the launch posts skip.

A founder analyzed all 196 YC Spring 2026 companies and found 86 share one shape, AI-native B2B Agent-as-a-Service: YC's 2024 vertical-agent thesis executed at scale, and a wager that speed beats defensibility.

Strip the logos off the stablecoin neobank boom and most apps are the same six vendors in a trench coat; this teardown of 29 operators maps the vendor cap table that captures the margin, the regulation that makes a charter a moat, and the three escape routes left.

On June 12 a Commerce letter switched off two Anthropic models worldwide, raising an older question of whether software can be export-controlled at all: America tried once, on encryption, and lost in court, and this control fails on the same three layers.

In 48 hours the question of where long-running coding agents actually run got both answers at once: an indie $7M bet against lock-in, and a frontier-lab acquisition that pulls the runtime layer inside Codex.

On one June Tuesday, money poured into startups all selling AI-agent 'memory,' a word that covers at least four different technical products, while the frontier labs absorb the easiest of them as a free feature.

The lab that refused capital is closing a $7.4B first round at up to $59B: why a battery maker leads it, why Tencent's check is defensive, and why founder Liang Wenfeng kept 40%.

The Codex CLI config that ends the permission nagging — config.toml profiles, sandbox and approval modes, AGENTS.md, reasoning effort, MCP servers, and the codex exec setup for CI.

Three different products spent two years fighting over the phrase 'LLM router': aggregation won at a $1.3B valuation, gateways became table stakes, and the most-hyped category, semantic cost-routing, quietly froze on GitHub.

The OpenCode config that earns its keep — opencode.json layers, the build/plan agents, permissions, AGENTS.md, MCP, and the bring-your-own-model setup that survives a vendor pulling your access.

On June 7, 2026, X renamed the agent loop 'loop engineering' and called it the successor to harness engineering; The Deep Feed ran 47 Claude Code loops in a sandbox and found the naming churns while three of four technical claims hold, one to the cent.

In one week in June 2026, a16z published a manifesto declaring small businesses 'the next frontier for AI' and led three rounds for companies promising to run the business itself; the pitch is horizontal, but every winner is vertical.

On June 6, 2026, OpenAI shipped a security feature that works by turning agents off, the same week Grok started filling grocery carts and Qwen started driving desktops; read together, the frontier just conceded the action layer is the unfixable attack surface.

On June 1, 2026, Anthropic filed confidentially to go public at a $965B valuation while OpenAI's Foundation made its case for spending $25B on 'AI resilience': one lab privatizes the upside, the other pre-commits capital to the downside.

Within three weeks MiniMax M3, Qwen3.7-Max, DeepSeek V4, Kimi K2.6 and Nemotron 3 Ultra all claimed the coding frontier, and third-party leaderboards show a real but small gap, mostly self-graded marquee numbers and a jurisdiction risk no benchmark measures.

Palo Alto bought Portkey, Cisco bought Astrix, Asana bought StackAI and $340M flowed into agent-identity startups in six weeks, as the five layers of the agent control plane get bought rather than built.

Eleven OSS agent runtimes share one recurring pattern of memory, sandboxing, skills-vs-MCP, tool governance, and transport: only Letta solved memory, only smolagents solved sandboxing, and the 2026 vendor SDKs are quietly absorbing the rest into hosting platforms.

xAI shipped a terminal coding agent, then put a competitor's model inside it that out-codes its own; read against a $60B SpaceX option on Cursor, the CLI war looks less like four rivals and more like one stack assembling itself.

Modal raised $355M at $4.65B selling AI agents a computer, as do 22 other companies; this map of the sandbox race shows who runs Firecracker, who doesn't, and why the substrate war is nearly settled while everything above it is wide open.

Generation got nearly free, and a 1999 book's discipline didn't die in the vibe-coding era: it got concentrated, inverted, and measured at scale by GitClear, Veracode, METR, and Kent Beck.

On May 25, 2026, Greg Isenberg and Stepan Gershuni (cyberfund) independently posted near-identical playbooks for building AI-native startups 106 minutes apart, with no cross-promotion and no awareness, and that convergence is the story.

Mogra crossed 15,000 AI sandboxes spun up, 71,300 chats and 2.7 million messages by March 2026, and this read goes inside its persistent-sandbox bet, two-tier product, Projects, Loops, Secrets, keepalive, Avocado wallet and bundled-stack thesis.

Garry Tan's gstack passed 101,020 GitHub stars in ten weeks with 33 slash commands, 7 standalone CLIs and 4 OpenClaw skills, and this full reference covers every skill with three concrete examples and decodes the LOC controversy.

gbrain, the persistent memory layer behind Garry Tan's own AI agents, runs 66 autonomous cron jobs over 146,646 pages and 24,585 people, making it the rare self-hosted alternative to vector-DB wrappers in 2026.

Nicolas Cole locked himself in his office for 7 hours to pick 7 X articles about AI and Claude, and the resulting reading list now has 255,006 views and 3,508 bookmarks; the pattern underneath is the story.

OpenAI is putting $2M of API tokens into every YC Spring and Summer 2026 startup in exchange for an uncapped SAFE, and YC moved the Summer application deadline to May 25 to widen the pool.

On the closing fireside at Stripe Sessions 2026, Daniel Gross owned Meta's compute strategy in his own words and Nat Friedman described the legacy labeling tool he ripped out, showing that the bottleneck inside MSL isn't talent.

The YC Spring 2026 batch is the empirical answer to the April 30 RFS read, with eight predictions holding, four breaking, and every cap table rewritten by OpenAI's token deal.

Google's 2h57m I/O 2026 keynote landed eight first-class platform shifts, from Gemini 3.5 Flash and Antigravity 2.0 to TPU 8t/8i and a $180-190B 2026 capex line, while the Pro model didn't ship.

Cursor shipped @cursor/sdk in public beta on April 29, putting the same harness that powers the IDE behind a TypeScript API, and this working developer's tour covers every surface with code, gotchas and ideas.

Eli Mernit's May 19 launch pitched Capsule as Supabase for AI apps, but the SDK ships as a closed-source 16,400-LOC wheel, the gateway is hardcoded, and Beam takes 10% of every Stripe Connect dollar.

On May 15, X pushed 18,263 lines of Rust and Python to its public algorithm repo, a full rewrite of the 2023 Scala stack around a Grok-based transformer, but the code does not build and the engagement weights are still missing.

The Developers playlist is Anthropic's public lecture series for builders, and read against Building Effective Agents, MCP and Claude Code, its 15 videos hang together as one canon built on six core positions.

Stripe shipped 288 launches at Sessions 2026 and devtools-Twitter latched onto four (UCP, MPP, Link CLI, Tempo), but the other 284 contain at least 25 weekend-buildable products no one is shipping, ranked here.

Paul Graham disclosed on stage in Stockholm that YC internal data shows startups that go home after the batch are only half as likely to become unicorns, yet coverage of the talk led with the wrong sentence.

xAI ships STT at $0.10/hr, Scribe v2 sits at 2.2% AA-WER and Wispr Flow users say accuracy regressed, so this guide splits speech-to-text into five markets and checks what the benchmarks show.

The 28-minute developer keynote at Stripe Sessions 2026 was a complete blueprint for retrofitting any API for agents, and while devtools Twitter mostly missed it, its seven-step playbook is laid out here line by line.

Two near-simultaneous coiners (Mitchell Hashimoto and Viv Trivedy), seven convergent voices and twelve coding agents that built the same primitives independently mark a discipline that crystallized in nine months.

Higgsfield raised $130M at $1.3B wrapping Veo, Kling and Sora, then spent 30 days getting called a scam; tutorials sell five-figure months, yet the full cost-to-earnings math puts the median Fanvue creator at $150 to $300.

Dmitri Dolgov's AI Ascent 2026 fireside laid out the multimodal world-action-language model, the driver-simulator-critic stack, and why the 13x safety number is the bill that competitors haven't paid yet.

Hassabis's "2030" line at AI Ascent 2026 is the same prediction he made in 2010, so the publishable claim is the falsifiable one: AlphaFold 3 plus Isomorphic Labs has now slipped its first clinical trial to end of 2026.

Greg Brockman's AI Ascent 2026 thesis is that judgment, not compute, becomes the scarce resource as agents take over execution — and Codex Cloud, Cursor 3, Devin, and Anthropic auto mode are already pricing it in.

Jim Fan's AI Ascent 2026 talk made the case that EgoScale's 0.1% teleop ratio, DreamDojo's 44,000-hour neural simulator, and a stated 95% confidence in robot auto-research by 2040 mean robotics has finally hit the data-and-compute curve LLMs hit in 2020.

Karpathy's AI Ascent 2026 talk laid out the December 2025 phase change, why the Menu Gen app shouldn't exist, jagged intelligence, verifiability as the master variable, and the new programming surface.

Mubbu's May 3 X long-form lays out a 5-agent stack for running founder personal brands in 45 minutes a week, and this audit finds the architecture real but the proofs gated.

Stripe opened Sessions 2026 with a chart of new firm formations going vertical and a familiar thesis — but the through-line was a single line from a Meta VP: payments are pivoting from a moment to a policy.

Fred Schott's Flue treats the agent harness as a first-class TypeScript build target; this update covers the 1.0 Beta rewrite through beta.4, the Actions primitive, @flue/react, the Cloudflare runtime deal and the Vercel-backed eve rival.

At AI Ascent 2026, Pat Grady and Sonya Huang argued long-horizon agents are functional AGI, and the new claim is the meta-claim: Sequoia is now writing checks as if commercial AGI already arrived.

Stripe shipped 288 launches at Sessions 2026, but only 3 are GA day-one — here is what is actually buildable today, the new agent-payment protocol stack, and ten projects worth shipping this quarter.

The Claude Code config Boris Cherny actually runs — settings.json layers, permissions, plan mode, memory, hooks, skills, subagents, MCPs, sandboxing, and the 5-worktree workflow.

Anthropic shipped Claude Managed Agents on April 8, 2026 at $0.08/hour; twenty days, 2,029 tweets and four converging platforms later, agent infrastructure has commoditized and domain expertise is the new moat.

Polymarket swapped its trading stack in one hour on April 28, and three days later V1 trades have hit zero while thousands of bot wallets sit silent on V2, which makes the migration look less like an upgrade than a cull.

OpenClaw has 138 CVEs, a 1GB/min memory leak and a supply-chain attack where 80% of audited skills were malicious; Hermes's flagship feature silently deleted user data on launch day; and 25+ SaaS providers now live off that pain.

Since Anthropic shipped the SKILL.md spec in September 2025, 480+ marketing-flavored skill repos have surfaced on GitHub, with four canon leaders, a bimodal quality distribution and four conspicuous gaps.

Michel Lieben's two May drops in 24 hours cover a 19-API picture and the API-led GTM playbook underneath; this audit sorts what shipped from what's vapor and lays out the five-step migration that decides whether a skill survives team handoff.

Y Combinator's Summer 2026 Request for Startups thread reached 8.65M impressions in a week, and the chatter underneath the list is more useful than the list itself.

Greg Isenberg's May 12 podcast with Nick Vasiles of Orgo turned the underground AI-agent-agency operator pattern into a 47-minute YouTube playbook, making the stack, the unit economics and the daylight problem public.

On Day 2 of Stripe Sessions, Collison and Glassberg Sands said the K-shape is not in their data; the NY Fed, BofA, JPMorgan, Delta and United say otherwise, and both can be right depending on the panel.

Sequoia says the next $1T company will be 'a software company masquerading as a services firm,' a sentence that buries vertical SaaS as a category without ever naming it, and Harvey, EvenUp and Rillet are among the companies that fit.

Every agent-memory tutorial opens with a vector DB, but that is a product opinion rather than a default; mature agent runtimes start elsewhere, and Qdrant or sqlite-vec earns its keep only past roughly 50K episodic items.

The Google for Startups Cloud Program is real money, but it also carries four specific traps that have bankrupted founders, from a 5x Year 2 cost cliff to credits that do not cover Claude on Vertex AI.

Three production speech-to-speech APIs shipped in a single quarter, and with them the cascaded STT→LLM→TTS pipeline that voice agents were built on has become a legacy architecture.

Sixteen months in, MCP counts 97M SDK downloads, 20,000+ servers, six framework attempts and one unsolved auth problem; this map covers who is shipping and where the protocol breaks at scale.

Fifteen open-source deep-research agents reduce to four architecture archetypes, two of which are dead ends, and this field guide sorts the agents by archetype so the right one is easier to pick.

A December 2025 paper from OpenServ Labs argues that bounded reasoning graphs beat free-form chain-of-thought for production agents, and the numbers back it up: gpt-4o jumps from 19.9% to 53.7% on SCALE MultiChallenge.