Agentology

Navy title card reading “Astra for Coding: Why Are We Doing This Again?” with the subtitle “Some thoughts on Astra and newfangled long-horizon models” and a photo of Armin Ronacher

Oct 24Link · lucumr.pocoo.org ↗

Astra for Coding: Why Are We Doing This Again?

Gitdailies — Pull Request Metrics & Advanced Alerts for GitHub

Purple title card reading “Five Months Running a Hospital for Code: Treatment Plans, Discharge Nurses, and Lockdowns”

Oct 19Link · 55m · cockroachlabs.com ↗

Five months treating bugs like patients and coding agents like a medical team

Become a sponsor

Hand-drawn stick-figure sketch of a first-person shooter: a scoped rifle in the foreground aimed at a figure between two trees, beside a radioactive barrel

Oct 19Link · 10m · momo5502.com ↗

500+ Billion Tokens Later: Letting AI Agents Decompile A First-Person Shooter

Flat illustration of a mixing console with sliders, dials and coloured buttons beside a large green toggle switch on a blue gradient

Oct 19Link · 16m · amazon.science ↗

Why don’t machine learning research agents overfit?

A recording spool unwinds a strip of events, with one icy blue event highlighted.

Oct 14

Agent observability: the traces you open after an incident

Orange circuit paths reach a red processor through gaps in a cream enclosure.

Oct 14

What the Claude Code sandbox contains, and what it doesn't

A tray packed with cream and orange blocks, with a red block left outside.

Oct 14

Context engineering is a budget, not a craft

Five connected processing blocks stand on a shared foundation split by a crack.

Oct 14

How multi-agent systems fail, ordered by how late you find out

Five rows grow from one tile to five, each repeating earlier tiles and adding one orange tile.

Oct 14

Where agent tokens go: each turn pays for every turn before it

Red, orange and cream pixels scattered on deep maroon

Oct 6Link · 8m · alignment.openai.com ↗

An agent used DNS to reach an external chatbot

Red, orange and cream pixels scattered on deep maroon

Oct 6Link · 56m · danluu.com ↗

How well do agents use test and verification techniques?