HackerLangs
TopNewTrendsCommentsPastAskShowJobs

polynomial

no profile record

Submissions

Treating Agent Reasoning as a Span

forestmars.substack.com
1 points·by polynomial·23 dagen geleden·2 comments

Why OpenClaw's observability model can't keep pace with its own scope

ctolunchnyc.substack.com
4 points·by polynomial·5 maanden geleden·1 comments

comments

polynomial
·10 dagen geleden·discuss
If a conscious model is born here, it should automatically be a (US) citizen.
polynomial
·12 dagen geleden·discuss
The most concerning thing here is the temperature problem. If your harness isn't providing deterministic output at a temperature setting of 0.0, it is broken.
polynomial
·22 dagen geleden·discuss
You don't know that No Feigning Surprise is actually from an xkcd comic, before it was a wizardzines post? U+1F632
polynomial
·22 dagen geleden·discuss
Honestly, if you compare Tim's pre-GPT writing style, to his current stylometrics, you'll see a remarkably delta. It's as if he suddenly decided to change his established writing style and voice right when LLMs become widely available.

What a coincidence!
polynomial
·23 dagen geleden·discuss
Reasoning chains as first-class OTel spans, which is what makes autonomous agent production deployments auditable (rather than indefensible.)
polynomial
·2 maanden geleden·discuss
Less MTTF and more MTTR
polynomial
·2 maanden geleden·discuss
This is horrifying on so many levels even if you don't already know how the story ends.
polynomial
·2 maanden geleden·discuss
Calling it "Act 2" certainly isn't helping with investors.
polynomial
·2 maanden geleden·discuss
A 67 year old woman living out of her car? JFC.
polynomial
·2 maanden geleden·discuss
Very obviously LLM written to anyone who's spent any time using them.
polynomial
·2 maanden geleden·discuss
Not everybody isn't worried: https://ctolunchnyc.substack.com/p/cto-lunch-nyc-spring-2026...
polynomial
·3 maanden geleden·discuss
"The gap is larger than most people realize" is a a dead giveaway.
polynomial
·4 maanden geleden·discuss
Fully agree.

If you don't change your approach but just use CLI "intead of" MCP, you'll end up with a new spin on the same problems. The guardrails MCP provides (identity, entitlement, multi-principal trust boundaries) still need to exist somewhere.

https://forestmars.substack.com/p/twilight-of-the-mcp-idols
polynomial
·4 maanden geleden·discuss
This is the right framing. The chain policy problem is what happens when you ask the registry to be the entitlement layer.

Here's a longer piece on why the trust boundary has to live at the runtime level, not the interface level, and what that means for MCP's actual job: https://forestmars.substack.com/p/twilight-of-the-mcp-idols
polynomial
·4 maanden geleden·discuss
Understandable as always a proxy for predictable.
polynomial
·4 maanden geleden·discuss
This is the under acknowledged "secret" of this reconfiguration.

It's like the Bill Joy point about mediocre technology taken to the next level.
polynomial
·4 maanden geleden·discuss
Robots making fun of us complaining about them.
polynomial
·4 maanden geleden·discuss
Well at least you know who to fire
polynomial
·4 maanden geleden·discuss
"Petri dish rewrites React in Rust"
polynomial
·4 maanden geleden·discuss
Tragically this reference is all but lost generationally.