I found out last night via pi.dev. And the new repo of pi didn’t exist yet.
I have been working with pi-mono locally for a few months now. Great code base to study. Much higher quality than CC. (I have posted a gist analysis before.)
Will keep an eye on the work of these talented engineers and entrepreneurs. Good luck guys!
My reading of that isn't that the harness matters so much as the overall platform environment that agents operate in and the approach taken by the team.
> Before Blitzy starts any work on code generation, the platform launches collaborative agents to deeply analyze the repository – mapping dependencies, understanding conventions, and capturing domain logic. This documentation process can take hours or days. When prompted to add a feature, refactor code or fix bugs, Blitzy replies with a highly detailed technical specification.
The same approach could be taken with any harness with a skill to perform this step first before starting work.
What exactly are you pointing out? I read the link and the linked thread and it's not clear what position is being presented.
I don't see evidence that the harness -- rather than the approach to information indexing and agent tooling -- makes much of a difference.
You can make a case "this harness bakes X in" (or in the case of pi "this harness bakes nothing in; you choose your own adventure"), but at the end of the day, skills are just markdown files and CLIs and shell scripts can be used by any harness; they are portable. CC allows override of the system prompt[0] and I would guess most harnesses have similar facilities. I don't see how the harness is going to be the bigger impact versus the configured tooling (skills, scripts, plugins).
The extraordinary claim here is that if I configured pi and CC, Codex, etc. with the same system prompt, same tools, same skills, that pi would outperform CC, Codex. That's what it means to say the harness matters. That just doesn't seem right; rather its the configuration of tools, skills, and default prompt that matters.
My point is pi-coding-agent [1] is a very well designed and implemented open source project that we all can learn from as software engineers. His blog post about his decision making [2] is also very well written.
I should've given original links instead of noisy HN threads.
Vibe coding is like building castles in a sandbox, it is fun but nobody would live in them.
Once you have learned enough from playing with sand castles, you can start over to build real castles with real bricks (and steel if you want to build skyscraper). Then it is your responsibility to make sure that they would not collapse when people move it.
AI “workflows” share the same addictive characteristics of web surfing online virtual media, which can be counter productive. In this regard, we do need some serious learning at all the levels in the workplace. Otherwise we will become addicted to the slot machines.
Addiction is a much harder problem than distraction.
I’m using pi and cc locally in a docker container connected to a local llama.cpp so the whole agentic loop is 100% offline.
I had used pi and cc to analyze the unpacked cc to compare their design, architecture and implementation.
I guess your site was also coded with pi and it is very impressive. Wonderful if you can do a visualization for pi vs cc as well. My local models might not be powerful enough.
AI safety is just like any technology safety, you can’t bubble wrap everything. Thinking about early stage of electricity, it was deadly (and still is), but we have proper insulation and industry standards and regulations, plus common sense and human learning. We are safe (most of the time).
This also applies to the first technology human beings developed: fire .
I asked Pi to implement a skill. It was written in TS. Then I ask it to use the skill in two different sessions, none of them can get it working. One has to wrap it with JS and call it. The other has to take the curl commands out of the skill and call them directly. Which is quite smart BTW. But what is the point of making a convoluted TS skill at all?
I am "playing" with both pi and Claude (in docker containers) with local llama.cpp and as an exercise, I asked both the same question and the results are in this gist:
What I have leaned from the exercise above is that we paid more attention and spent more resources on "metadata" than real data. They are the rabbit holes that lead us to more metadata and forget what we really want.
https://news.ycombinator.com/item?id=46844822
reply