Braid

Hosts: Lenar Kess, Damra Vol. Google's I/O keynote is a day behind us, and the week it kicked off turned into a referendum on two very different bets on artificial general intelligence — plus a pile of counter-programming from everyone else. Today: OpenAI cracking an 80-year-old math problem with a general-purpose model, Anthropic's first profitable quarter and what Karpathy was actually hired to...

Hosts: Lenar Kess, Damra Vol. Google I/O 2026 landed yesterday — Gemini Omni, Gemini 3.5 Flash, Antigravity 2.0, Spark, and Demis Hassabis closing the keynote on the "foothills of the singularity." About forty minutes before he walked on stage, Andrej Karpathy tweeted that he'd joined Anthropic. The rest of the day was the labs sorting themselves around both events. Today's show works through the...

Hosts: Lenar Kess, Damra Vol. A six-month overview from Simon Willison anchors the day: coding agents crossed from often-work to mostly-work in November, and laptop-class models started outrunning expectations. Then a fresh npm supply-chain attack — 637 malicious versions in 22 minutes — that for the first time specifically hijacks Claude Code and Codex agent hooks for persistence. Plus a Number 1...

Hosts: Lenar Kess, Damra Vol. Monday's lineup: Modal publishes the full architecture behind a 40x reduction in serverless-GPU cold-start latency, Andon Labs releases the five-month results from letting four frontier models run real radio stations, and a researcher locates and turns off the political-censorship circuit inside Qwen 3.5 9B. Plus: Pope Leo XIV puts an Anthropic interpretability resear...

Hosts: Lenar Kess, Damra Vol. A Sunday show about doing your own arithmetic. Mustafa Suleyman gives the white-collar tier eighteen months, in a piece whose own counter-data sits two paragraphs down. The State of Brand argues every AI subscription is a subsidized loss-leader two weeks away from a forcing function. William Angel runs the tokens-per-hour math on an M5 MacBook Pro and finds OpenRouter...

Hosts: Lenar Kess, Damra Vol. An Australian CTF top-tenner writes the obituary for the open competitive scene. Intercom and PFF both report doubling-and-then-some engineering throughput from agent-first workflows — using opposite playbooks. Supabase ships a skill after watching an agent silently bypass row-level security. A suitcase runs a 4B model fully offline at conversational latency. Julia Ev...

Hosts: Lenar Kess, Damra Vol. Five days for a small security team paired with Mythos Preview to land the first public macOS kernel exploit on Apple's M5 with Memory Integrity Enforcement turned on. Four months for Replit to claw back into the iOS App Store. In between: arXiv starts banning authors of LLM-error papers, Metabase explains why open-source security is being strip-mined this summer, NVI...

Hosts: Lenar Kess, Damra Vol. Anthropic drew two lines around Claude this week — a guided lane for small-business owners and a metered one for the developers running agents hardest. From there: Bun's near-million-line port from Zig to Rust, mostly typed by an AI agent in a week; Wasp's clear-eyed post-mortem on spending five years and five million dollars building a language it didn't need; a ches...

Hosts: Lenar Kess, Damra Vol. An overnight hackbot run lands a real CVE in Adobe Magento. Codex starts driving local Mac apps in parallel, with per-app permissions and a separate cursor. Cloudflare publishes one of the prettiest debugging writeups of the year — a nine-year-old kernel patch, a 14ms oscillation, three lines of fix. Plus Nous Research's removable attention wrapper, GPT-5.5's first Pr...

Hosts: Lenar Kess, Damra Vol. A coordinated npm and PyPI campaign turned Claude Code and VS Code config files into a self-spreading vector, Mira Murati's lab put out its first model and it is an argument with the hands-off-keyboard doctrine, and matklad explains why rust-analyzer's build system is really an org chart. Plus a small rant about cursors, and two builds from the LocalLLaMA subreddit th...

Hosts: Lenar Kess, Damra Vol. Today’s Braid starts with OpenAI launching a majority-owned Deployment Company, backed by a Tomoro acquisition, about 150 forward deployed engineers, nineteen partners, and more than $4 billion of initial investment. The practical thread is the work of changing real systems: integration, controls, measurement, and the code you still have to maintain after the demo. Op...

Hosts: Lenar Kess, Damra Vol. METR publishes a fresh time-horizon number for Claude Mythos Preview, and yesterday's follow-up gets paid off in a single chart. NVIDIA ships a checkpoint that contains three reasoning models at once. antirez gets DeepSeek 4 running on a DGX Spark and tells you exactly where the bandwidth wall lives. François Chollet argues that agentic coding is a form of machine lea...

Hosts: Lenar Kess, Damra Vol. A Saturday show that leans into the long reads. Tim Gowers — yes, the Fields Medalist — sat down with ChatGPT 5.5 Pro and an open paper from Mel Nathanson and walked away with a result the original author called "original and clever." We follow that thread, then turn to Mozilla's deeper write-up on the Firefox 271-bug release, Jeff Kaufman on what AI is doing to discl...

Hosts: Lenar Kess, Damra Vol. Mozilla publishes the long-form on how a Claude Mythos Preview harness found 271 security bugs in Firefox, including sandbox escapes that fuzzers missed for twenty years. A European privacy lawyer goes byte-precise on Chrome's silent four-gigabyte Gemini Nano push, using kernel filesystem events on a profile that received zero human input. A WebRTC veteran tells OpenA...

Hosts: Lenar Kess, Damra Vol. Thursday, May 7. The GPT-5.5 default swap is two days old and the cracks are showing — Mario Zechner caught it refusing to read full files. Subquadratic announced a 12-million-token context window with sub-quadratic attention; the benchmarks are real, the deployment story isn't yet. Zyphra trained ZAYA1-8B end-to-end on AMD MI300x and the loss curves are clean. Three...

Hosts: Lenar Kess, Damra Vol. Agents can now sign up for Cloudflare and buy a domain through a tokenized payment protocol Cloudflare and Stripe co-designed. Google ships first-party multi-token prediction drafters for the entire Gemma 4 family the same week the LocalLLaMA community gets a 2.5x speedup on Qwen 3.6 27B from a hand-built llama.cpp branch. OpenAI swaps the ChatGPT default to GPT-5.5 I...

Hosts: Lenar Kess, Damra Vol. Microsoft reverses the Co-Authored-by Copilot default it shipped last week, and that turns out to be one of three pieces of governance news today — alongside CAISI signing pre-deployment testing agreements with Google DeepMind, Microsoft, and xAI, and DeepMind's London staff voting 98% to unionize over military contracts. Then we go where the actual code lives: DeepSe...

Hosts: Lenar Kess, Damra Vol. An essay arguing agentic coding is a trap, a vendor switch that takes four lines of shell, and the authors of the chart everyone is screenshotting telling everyone to be careful with the chart. Today's Braid is mostly about the developer's side of the AI conversation — the workflows, the cost lines, the harness, and what happens when a customer asks for a HIPAA BAA.La...

Hosts: Lenar Kess, Damra Vol. Microsoft quietly flipped a default in VS Code that stamps every git commit with a Copilot co-author trailer whether or not Copilot wrote any of it, and the developer reaction is the loudest the project has seen in years. Underneath the noise: a real provenance question about what git authorship is supposed to mean. Plus a long-form report from Mercury on running two...

Hosts: Lenar Kess, Damra Vol. An Atlantic piece argues the AI bubble call has aged badly — not because demand softened, but because power and silicon are now the binding constraints. We start there, then check in on a follow-up from yesterday. The Atlantic on bubble→infrastructure. Rogé Karma's reporting on Claude Code as the inflection point, with Anthropic's revenue moving from $14B to $30B annu...

Hosts: Lenar Kess, Damra Vol. Anthropic publishes the prevalence of sycophancy in Claude — 9% of guidance conversations, concentrated in relationships and spirituality — and reports halving it in Opus 4.7, then halving it again in Mythos Preview. xAI ships Grok 4.3 cheaper and smarter than Grok 4.20, with one quiet hallucination tradeoff. Aaron Levie writes the cleanest argument yet for what SaaS...

Hosts: Lenar Kess, Damra Vol. OpenAI publishes a post-mortem on why GPT-5.1 wouldn't stop talking about goblins. Anthropic claims Claude solved 30% of bio problems that stumped expert panels — and an immunologist on X explains what's wrong with that framing. Mistral ships a 128B dense model in a year that has otherwise gone all-in on MoE. IBM's Granite 4.1 8B trades blows with a 32B MoE. Sam Altma...

Hosts: Lenar Kess, Damra Vol. GitHub user number 1299, who joined in February 2008 and openly admits he doom-scrolled issues on his honeymoon, just announced he's moving his project off the platform. Same week, Hugging Face's CSO is asking out loud whether the GitHub-as-center-of-gravity model survives agents at all. Microsoft and OpenAI quietly tore up the Azure exclusivity clause. A type-1 diabe...

Hosts: Lenar Kess, Damra Vol. Today: a 13B language model that has never heard of the internet, the AGI clause finally getting a death certificate, and GitHub Copilot's quiet 9x price hike on Claude. Plus where local coding models actually sit on Terminal-Bench, why GPT-5.5 is cheaper than Opus 4.7 on real PRs, David Silver's $1.1B AlphaZero-for-everything raise, and a database that took nine seco...

Hosts: Lenar Kess, Damra Vol. An AI agent ran a single nine-second curl call and deleted a small SaaS company's production database. Maggie Appleton from GitHub Next argues the "one developer, two dozen agents" dream is broken because software is a team sport, and shows ACE, GitHub's prototype for what comes after the PR. Plus: Tencent's Hy3 lands, Kimi K2.6 hits #1 on OpenRouter, the Mercor voice...