Peter Schuller

Surprised so many model/harness combos have trouble waiting for a scripted event. claude code seems decent at it by default. Codex sucks - to the point that I have a whole skill just with a pre-written script and harsh instructions for the agent to not burn cached input tokens polling all the time (github.com/scode/dotfiles/tr…). Grok bot also just randomly failed to succeed in having a script wake itself up.

Apparently I need to adjust my skills or claude code will refuse to write sub agent reports to files even when explicitly told to.

If you've been hearing about the Starlink/Ukraine thing over the past week, you might be under the impression that what happened is that they slowed down internet speeds. Well, that's wrong. I only realized because I accidentally stumbled upon reporting that wasn't misleading, after only hearing misleading reporting for days. I had a bunch of agents work on a TLDR of what actually happened and how it's been reported: snippets.scode.org/s/starlin… (I am making a point of quality of reporting, this is not an opinion on anything else.)

Well, this is highly unscientific but better than "I used it for a while", and definitely not benchmaxxed: snippets.scode.org/s/rust-co… Quality variance is basically noise at this sample size but still consistent with same ballpark performance - and way faster than both Luna and Haiku.

codex and claude code are so yesterday, let me introduce you to Palantír

I just tried it, this is potentially exciting! I have only done basic stuff (repo navigation, listing PRs), but 2.5 fell immediately flat when I tried it. This worked fine, and it's *so so so* fast.

Haha. I was daily driving a claude/codex load balancer I am building, and a bug caused a session to be unrecoverable. The other session which was debugging the problem instead hit a bug in codex-lb so now *that* session is unrecoverable. (Luckily just asking an agent to read the other agent's transcript more or less works pretty well as a fallback resume mechanism.)

Perfect example of agents still lacking judgement (in this case Opus 5.5). Me: Hey lets turn a bunch of this stacked-PR handling skill into a deterministic script so we have less tool calls. I don't want to wait 30 seconds for the agent to churn tool calls for basic operations all the time. Agent: Here we are, unit tests, e2e tests, and evals. Me: Cool, merge. Me in another session: Wtf is taking so long landing this stack? We disabled all github CI, this should be fast! Agent: It's a fixed always-on 2 minute wait for every PR while waiting for a GitHub status check. Me: *facepalm*

Looks like I started exceeding the limit on 2026-09-12. I am fixing it. I am super curious how much this ends up explaining why Codex was so much less reliable than Claude, which I now wonder - maybe it was mostly in this repo (where I do various long unattended runs a *lot*). It's muddled because I'm also having some *probably* unrelated issue where the model gets confused about how to spawn sub agents, though we'll see. My confirmation biased brain remains more and more firm on "silent failure sucks".

TIL that Codex *silently* (as far as I can tell) truncates project instructions injected at session start/post-compaction at project_doc_max_bytes bytes. And the default is 32kb.

lol Google. I was actually considering upgrading to Ultra to try Argon, but your broken end to end user experience prevents me from giving you money.

Much better way to get your Dot to run agents on your devbox - make an MCP server: snippets.scode.org/s/devbox-… Prototype implementation if you want to try/crib it: github.com/scode/devbox-mcp

Been daily driving my agent orchestrator (farhelm.io/) for several weeks and I now feel its ready for others to use. If you try it and have feedback, or even don't and have feedback on presentation, feel free to reply to me or use the in-app feedback flow.

*Loving* the instant steering in codex they recently shipped.

Truly gone full circle. The AI asking me for permission to solve CAPTCHAs to get stuff done. Truly just proof of work at this point with the side effect of annoying humans.

Claude just on its own started talking to another claude on the same box (I asked it to fix a test, but another agent had already fixed it; the first agent noticed the PR and told the other agent "hey can you merge"). It's cool, but not sure how I feel about it :)

Prediction: Reasonably soon we'll have at least one commonly used programming language optimized specifically for agents (eg. token efficiency) instead of a human reader. Probably with "expansion" to human readable form when needed.

After using claude / opus 5.5 for a while except for code review, I tried using gpt 6.1 sol high for a bit. So many issues with broken tool calls, premature ending, generating broken shell scripts, failure to load a skill after it was named. I'm switching back and I will end up wasting usage - it's that bad. :( Until this improves I'm only using it for code review.

I'm now at the point where I'm probably gonna make a system that tails agent sessions just to detect and send me a phone push notification when something gets safeguard blocked and auto-generate a hand-off so I can find the agent and /clear + "continue from this handoff". Siiigh.