TIL Opus 5.5 confuses "boolean blindness" with "boolean trap" 🤦‍♂️
Strange I haven't noticed this before, wonder if this is a regression in 5.5 ?
So you have to include a brief definition of boolean blindness where you mention it.
Relevant because Claude (tell it to avoid garden path sentences to generate output that's more readable)
everything i know about managing agents i learned from the amazing programmers and computer scientists that came before
constraints in your codebase are freeing for both humans and agents! this has always been true, but if you were a small dev team you likely never had to think that deeply because you could trust your fellow programmers to write code and review well
but large companies have had to solve this problem since even before agents. because before you had agent slop, you had human slop. lots of engineers wrote bad code out of necessity (eg backend engineer forced to write frontend or reverse). the solution to this was constraints: lint rules, smarter compilers and diagnostics, high quality tests, investments into observability, and so on
the arrival of agents just means that big company problems are now everyone's problems. the good news is that none of this is really that novel: it's just good engineering
youtube.com/watch?v=GqmsQeSz…
Amazing how github copilot reviews went from being shit to being amazing to now being mostly an annoyance, all in a few months.
They now apparently run skills randomly with no chance of turning that off ffs
Mauricio Scheffer retweeted
I'm ready to be wrong on this, but the idea that assembly is, or ever will be, a good language for agents feels ridiculous. It forces the agent to attend to a huge number of extra details. Same problem as with humans! Why spend limited attention and context on that?
"So the honest answer is: I implemented the shared-prompt target you asked for, but I did not run the optimization method I told you I would use"
Well fuck you very much for wasting my tokens gpt-6-sol
I still don't trust agents to push back as much as they should.
Some of the things I often say:
"what do you think?"
"I could be wrong, be objective"
because as it turns out I am very often wrong.
It's 2026, AI is everywhere in web dev, how is it possible that I still run into this ancient UX mistake with greyed out buttons ffs
nngroup.com/videos/why-disab…
What I'm settling on for now is a first pass of agent reviews, which include several project-specific properties, and then a human review, which is actually mainly trying to infer missing properties to be encoded as automated checks.
Unfortunately that extraction of properties cannot be automated yet, as far as I have tried at least
I bet most people that say AGENTS.md is an antipattern just vibe it and let the clankers manage it and have no idea what's in it.
Though I do agree that a bad AGENTS.md is worse than having none at all.
You really really really have to own every bit that's in there.
Devs: I hate Claudese / claudeisms! The clankers write too much / wrong code documentation!
Also devs: ackshually AGENTS.md / CLAUDE.md is an antipattern, delete it
🙄
Great, now I can’t snack while I work because chewing interferes with shouting at the clanker.
My monitor’s got bits of peanut stuck to it now.
If your name is on it, whatever you said *you* are saying it, regardless of the tool you used to craft the message.
Same thing goes for code BTW. If that slop has your name on it it's *your* slop, I don't care what tool you used to create it.
Was diagnosing some radio issues in a cheap tablet last night and seeing good old AT commands in the logs brought a smile to my face :)











