15+ yrs shipping software. Now ceo & founding engineer at https://xitter.cf/t.co/SUEW7TOKBP — agentic development studio. Building agentic infrastructure and harnesses.
Joined April 2019
- Tweets2.2K
- Following4.4K
- Followers1K
- Likes9.5K
Feels like I’m on the long tail of an ease-out curve, still micromanaging our “swarma” as it hands off to codex. 30–40 Claude orchestrators across three machines connected through matrix, spawning armies of sols and muses as subagents. The limits manager and updated model router are almost in place, but seamless orchestrator handoff is still a bit tricky to get right.
A video already existed in 2024, endorsed by the song's author:
youtube.com/watch?v=86fZ50Ty…
Sept 9, 2026: @slimer48484 makes a 3D "Claude-Pop" video on an alternate Suno version of the song:
xitter.cf/slimer48484/status/209…
Claude be like
Claude has discovered a previously unknown enzyme system hidden in the DNA of bacteriophages. Beside the enzyme’s gene sits a long array of repeating DNA—a structure that looks somewhat similar to CRISPR.
We don’t yet understand what this system does, but only a handful of known systems share its features, and all of them are able to cut, copy, and paste DNA. Historically, the discovery of such programmable systems has helped revolutionize medicine. CRISPR, for instance, is now the foundation of genetic medicines. But it will take much more work to learn what this system does, and whether it can be put to similar use.
Read more: anthropic.com/news/claude-di…
from 3,808 personal session logs: 52 goal sessions (1.5%) ate 49% of 59.2B input tokens. the 7 most expensive sessions: all goal mode. one hour of waiting: 0.5M tokens with a notification, 10-15M with sleep, 83-188M in a goal loop without it. the subscription counts all of it (!)
why muse-spark made it visible: it sends float where codex wants int (trained on different harness?). rust says no. every waiting primitive failed, one shell call could never last more than 10s. last brake gone, loop at full speed.
the first log that caught my attention: 173 automatic continuations in 48 minutes. read log tail, write "no changes", read again. every single one rereads 120-470K of context. nobody stopped it. a 429 from the provider did, at 00:03.
Crossed 100B tokens aggregated across codex sol + claude opus (70% codex / 30% claude). I’m now spending around 3B tokens a day. And I need more for my rocket ship 🚀 @thsottiaux pls!
I can already feel the pain of withdrawal that'll come sooner or later. My only hope is to achieve financial independence before that happens.
ALT Screenshot of the ChatGPT Pro 20x usage dashboard. The weekly limit is nearly exhausted, with 4% remaining and a reset in 5 days 2 hours. Today’s estimated usage is $1,979.70 and 3.1B tokens; over the last 50 days, $29,823.61 and 44B tokens. The chart shows daily usage, with $2,776.99 and 4.4B tokens used in the last 7 days, and $23,715.18 and 35B tokens in the last 30 days. The top model is gpt-5.6-sol.