@jarredsumner@kentcdodds@webdevcody Yep exactly. And rust works well for this when combined with robust tests. But you still need to go back and review. Even if you can't read every line of code, have the agent explain the algorithm/design/structure and spot check often to make sure it's not hallucinating.
@rsmck@wesbos I've got some 500v 40a power cords, the other end is a field of solar panels though.
60a at 5v sounds crazy though.
The only thing I have in that ratio is 50v 600a (a large LiFePO4 bank)
The resulting code is amazing and upon initial inspection looks perfect. Even running it appears to just work on the first try.
But you said it needs to *look* correct, not that it had to *be* correct.
Be careful out there and wish wisely.
Remember everyone, LLMs are amazing math models that can make extremely correct *looking* code.
By pure coincidence, these are sometimes right and the more guardrails and feedback you give it, the more likely it's actually correct.
But finding error is *very* hard, by design.
Imagine if you had one of those genies that would grant wishes, but always in the most subtly wrong way while technically doing exactly what you asked.
Then imagine you asked it to write code that looked correct and big free.
The genie smirks and says "as you wish"
@rikarends Also the agent frameworks are all optimized for mega models with massive contexts. I want to see more optimized to perform best on smaller local models on consumer hardware.
When Opus 4.5 came out, it was a one-way door to a new way of engineering. Agents now do most of our coding.
Knowing the inherent flaws and over-confidence of LLMs, we sent a clear message to our teams. Vibing and mission-critical infrastructure don’t go together.
We’re sharing some of our early internal guidance in how we’re “agenting responsibly”, prioritizing security, durability, and availability at all times.
https://t.co/b36GiE76Ue
@fat@pierrecomputer Can we get a we socket based protocol to fetch git objects in batches?
Something like one side sends a list of hashes in a message and the other side sends the corresponding objects as messages.
Think lazy loaded VFS for ultra low latency clones.
@cramforce@royalicing But great minds think alike. My first version did exactly what you suggest and walk forwards and back to find the newline delimiters. It worked, but I didn't like the complexity of the code. Pre-scanning makes the hot-path simpler and cheaper.
@cramforce@royalicing We do a quick scan for newlines when we first fetch the file into the worker and cache the full bytes along with an array of line offsets in the memory cache.
Then when doing the binary search we use the cached line offsets to greatly simplify the logic.
@dCodes03@vercel_dev@vercel@rauchg Bloom filters are simpler, the encoding is plenty small as-is, and we need to optimize for CPU cost and simplicity when doing the lookups.
Bulk redirects on Vercel now scale to 1 million per project with near-constant lookup time.
We shipped the simplest thing first, profiled it in production, and iterated from there ↓
https://t.co/h7woUGbVFj
@thdxr I'm finding that it lets us experiment with new directions quickly, but we need to be OK throwing away lots of code as well once we decide it wasn't a good idea.