The guys in the token selling business will be delighted.
Intent-Based Commits
31–40 of 54 posts
Re: Intent-Based Commits
#32I noticed in the README that each commit message includes the agent and model, which is a nice start toward reproducibility. I’m wondering how deep you plan to go on environment pinning beyond that. Is the system prompt / agent configuration versioned? Do you record tool versions or surrounding runtime context? My mental model is that reproducible intent requires capturing the full "execution envelope", not just the…
LLMs are non-deterministic so I don't see how it's reproducible.
That’s fair - strict determinism isn’t possible in the traditional sense. I was thinking more along the lines of bounded reproducibility.
If the model, parameters, system prompt, and toolchain are pinned, you might not get identical output, but you can constrain the space of possible diffs.
It reminds me a bit of how StrongDM talks about reproducibility in their “Digital Twin” concept - not bit-for-bit replay, but reproducing the same observable behavior.
Re: Intent-Based Commits
#33[flagged]
Re: Intent-Based Commits
#34It would be good to see a real example. There’s a sketch of one in the README.md but I’d be interested to see how it works in real life with something complicated. > Add users with authentication > No, not like that > Closer, but I don’t want avatars > I need email validation too > Use something off the shelf? Someone in this place was saying this the other day: a lot of what might seem like public commits to main ar…
> It’s unclear what the “squash” process is for “make me a foo” + “no not like that”.
Commit your specs, not your prompts. When a change is correct, any information of value contained in the prompt should be reflected in the spec.
Re: Intent-Based Commits
#35[flagged]
Thanks for the ad
Re: Intent-Based Commits
#36> Every commit is reproducible. The prompt is preserved with > some extra attributes about which model and agent was used. > You can re-run any commit against a fresh checkout to see > what Claude generates from the same instruction. I don't see how this is true. LLMs can generate different outputs even with the same model and inputs.
as do most humans
And? The claim is about AI.
Re: Intent-Based Commits
#37It would be good to see a real example. There’s a sketch of one in the README.md but I’d be interested to see how it works in real life with something complicated. > Add users with authentication > No, not like that > Closer, but I don’t want avatars > I need email validation too > Use something off the shelf? Someone in this place was saying this the other day: a lot of what might seem like public commits to main ar…
> It’s unclear what the “squash” process is for “make me a foo” + “no not like that”. Commit your specs, not your prompts. When a change is correct, any information of value contained in the prompt should be reflected in the spec.
The problem I have is that when you squash the code it does the same thing.
PR.patch = a.patch | b.patch
exec(PR.patch) = exec(a.patch | b.patch)
PR.md = a.md | b.md
ai(PR.md) != ai(a.md) | ai(b.md)Re: Intent-Based Commits
#38So, using git exactly as usual?
edit: and also, not bothering to represent manual edits to the code...
Re: Intent-Based Commits
#39> Every commit is reproducible. The prompt is preserved with > some extra attributes about which model and agent was used. > You can re-run any commit against a fresh checkout to see > what Claude generates from the same instruction. I don't see how this is true. LLMs can generate different outputs even with the same model and inputs.
How different? So do compilers.
If you have a compiler the same source code and the same options, it should generate the same output everything provided you aren't using some compiler pragmas or something similar that embeds timestamps or random numbers or similar. If you give an LLM the same input, it can generate different outputs (controlled by the temperature setting).
Re: Intent-Based Commits
#40> Every commit is reproducible. The prompt is preserved with > some extra attributes about which model and agent was used. > You can re-run any commit against a fresh checkout to see > what Claude generates from the same instruction. I don't see how this is true. LLMs can generate different outputs even with the same model and inputs.
the models themselves don't even last a few years. Do they even last a few weeks? Who knows how much they subtly change behind the scenes.