Sandbox::https://x.com/TomaszPosluszny

Visitar URL original

Tom
@TomaszPosluszny
Sep 26
Spec Kit, Kiro and superpowers put a plan between spec and code. I built one feature with and without it. With plan: 25m48s, $5.26 Without: 17m33s, $4.83 Both PRs: one P1 bug The direct run caught a gap in the spec. The planned run shipped it. I read that as drift. Sooo...
Tom
@TomaszPosluszny
Sep 22
I wrote a spec for new feature (cost-effective TTS streaming in a frontend app). Afterwards I refined it with agent using a brainstorming skill (aka interview-me). Then I looked into it. It seemed fine for the most part, but I found one error and had a feeling it would be better
Tom
@TomaszPosluszny
Sep 21
This is the AI model type we were long waiting for: text IN -> decision OUT. Adding guardrails and traceability to Jev should be easier and more reliable than to LLMs that make decisions.
Maggie Appleton
@Mappletons
Sep 20
Can we agree to call Jev a "Decision Model"? I feel like they buried this in the explainations and docs. Language models output language. Decision models output decisions.
Tom
@TomaszPosluszny
Sep 16
I had a bit of AI hangover, so I slowed down a bit and started reviewing in-depth what it generates. There is still bloat in newer models. Is spec-driven development and reviewing only tests where it settles?
Tom
@TomaszPosluszny
Sep 8
I built a website blocker. Everyone who needs it has already blocked the place where I'd tell them about it. If it works, you won't see this. If you're seeing this, it's working on someone else. It blocks whatever you decide is the problem. Probably this.
1

Tom

Tom’s posts