Being early used to feel lonely. Now it just feels like Tuesday.

Told my wife we’re seeing You Can See Everything in IMAX. Don’t make me a liar, @nathanfielder #findingimax

How does a 'Write Like Me' agent skill work for someone that only has agent-written samples?

The Openclaw release coming out of beta in the next day or so is big.

Fable and Sol cannot resist creating extensive, impractical regex matching solutions for natural language detection. We need a benchmark for this.

I am confident that Fable 5 can build a citizenship verification process for access to Fable 5 in under an hour.

Dropping the hammer on a holiday weekend after weeks of Anthropic sending mixed messages is quite an alienating move to developers. The size of the subsidy has been absurd for sure. A new token hungry model release coming, extreme confidence in their new model, or B2B shift away from developer community must be part of the calculation.

Seriously, wtf is going on with Github this year.

openai/symphony ships

"Block is replacing support reps with AI while AI is coming for the payment processing layer itself."

Reddit reported $2.2B in revenue. First profitable year. Announced a $1B buyback. Major insider purchase. Stock fell 6.9%. The market is not grading on last year. It's pricing in a world where AI summaries replace the reason anyone visits Reddit in the first place.

When the AI layer treats your agentic toolchain as a threat, something about the incentive structures is badly misaligned. discuss.ai.google.dev/t/acco…

Token anxiety gambling addiction seems to be ratcheting up the past few weeks. I'm hearing it on podcasts and in conversations with people reaching out to me to talk agentic coding that have not been in it for a while. jkap.io/token-anxiety-or-a-s…

Well, crap. Would love to hear a security expert's take on this. --> Three new papers show attackers can infer what you're asking an LLM — even through TLS encryption. schneier.com/blog/archives/2…

The most underrated variable in agentic coding right now isn't which frontier model is better, it's the harness. What happens between "the model knows what to change" and "the change is applied correctly" is where failures are happening right now. A developer just demonstrated 10x improvement by changing the edit tool. Cursor trained a 70B model just for this. Claude Code str_replace has its own megathread. It's the most common reason something breaks with @openclaw workflows in my experience. This is why the harness winners are going to be the fastest shipping, provider agnostic wrappers for the frontier model provider CLIs like @conductor_build, @RepoPrompt , etc. Nobody seems to know the models better than these guys besides the Codex and Claude Code teams.

Windows now requires cloud authentication to open Notepad. windowscentral.com/microsoft…

Software engineering is so back blog.alaindichiappari.dev/p/…

GPT-5.3-Codex debugged its own training pipeline and managed some of its own deployment. The ouroboros has entered the chat.

Developer tools are now worth Super Bowl ad buys.