I do all my development as a longish running pipeline that mimics how I used to write software.
There is a high-level plan, it gets decomposed into steps that get performed sequentially. If during a step something occurs that challenges the original assumptions, the remaining steps can be reconfigured or the implementation stopped.
Well, I guess it's more rigorous than me, because it does a lot of quality checking along the way and if course, in the end.
This is the most boring approach to this work as you can imagine. But, after running this process hundreds of times and tuning it, today I can do about 10-20 PRs a day that are often quite good. I manually review and manually test and as models and my pipeline have improved the % of PRs that require zero human intervention after the initial planning is getting higher. Maybe 50%. For the other half, it's good, but there are often glaring bugs, especially for UI/UX stuff.
We are all people. This ally-of-the-west framing is propaganda. Who has harmed me more: this US or China? Who do I have more in common with: a tech worker in China or a US government official?
(I'm based in US - I use the best tech for the task).
I think git it the perfect system to bring order AI coding. Generate code however you like, but ultimately, it goes through a sensible, proven, and auditable pipeline. As someone also building AI tools, git it something I find myself "building with" - not against.
I used to use GLM before I knew about coding subscriptions and it was okay. I've tried every version since 4.6 and this one is doing a great job a spec-implementation runner. If I had to guess... somewhere between Sonnet and Opus in terms of quality. Z.ai's issue has been service reliability. So far so good on day one.
I don't think it matters how code is produced -- it matters what it achieves. Is there evidence that there is something wrong with recent Bun releases?
I'm not a fan of LLM's injecting themselves into PR/commit content. If you use multiple models, basically whichever one is operating git gets all the credit. But, even if you wrote all the code yourself, and just submitted the PR with Claude Code (or whatever) it would attempt to take credit for the changes.
I currently have rules in all of my skill files forbidding models from advertising themselves or taking credit.
The people who trust bad information from LLMs are the same people who trusted bad information from search results and new articles, it just takes them less time to get bad information.
There is a high-level plan, it gets decomposed into steps that get performed sequentially. If during a step something occurs that challenges the original assumptions, the remaining steps can be reconfigured or the implementation stopped.
Well, I guess it's more rigorous than me, because it does a lot of quality checking along the way and if course, in the end.
This is the most boring approach to this work as you can imagine. But, after running this process hundreds of times and tuning it, today I can do about 10-20 PRs a day that are often quite good. I manually review and manually test and as models and my pipeline have improved the % of PRs that require zero human intervention after the initial planning is getting higher. Maybe 50%. For the other half, it's good, but there are often glaring bugs, especially for UI/UX stuff.