Y
HN Search
Hacker News Search
new
|
comments
|
top
|
jobs
ElenaDaibunny
searching PlanetScale…
1.
▲
2.
▲
3.
▲
4.
▲
5.
▲
6.
▲
5 ms
·
31.
▲
by
ElenaDaibunny
5mo ago
Fragility compounds fast when you add visual grounding to the loop. Code agents at least get structured feedback from the compiler.
32.
▲
by
ElenaDaibunny
5mo ago
that adds up fast over a year.
33.
▲
by
ElenaDaibunny
5mo ago
one week in and already got a custom gui frontend going, that's fast
34.
▲
by
ElenaDaibunny
5mo ago
but latency in real GUI workflows with 50+ steps is still the elephant in the room for cloud-based agents
35.
▲
by
ElenaDaibunny
5mo ago
guardrails this well-designed matter way more than just throwing bigger models at agent tasks tbh
36.
▲
by
ElenaDaibunny
5mo ago
Good SDK tooling is a huge competitive advantage when you're trying to get developers to build on your platform instead of OpenAI's.
37.
▲
by
ElenaDaibunny
5mo ago
=The real question is whether listeners can actually tell the difference。
38.
▲
by
ElenaDaibunny
5mo ago
10MB is refreshing.
39.
▲
by
ElenaDaibunny
5mo ago
Category theory as a type system for tensor operations makes more sense than most algebraic abstractions people try to force onto ML.
40.
▲
by
ElenaDaibunny
5mo ago
how will they detect hallucinated refs at scale? Manual spot checks? Automated DOI verification? The policy seems right but enforcement is the hard part.
41.
▲
by
ElenaDaibunny
5mo ago
The fact that vendors still ship firmware with trivial obfuscation in 2026 is wild. I wonder how many data recovery shops already reverse-engineer these routinely but just don't publish.
42.
▲
by
ElenaDaibunny
5mo ago
this is the kind of HN post i come here for.
43.
▲
by
ElenaDaibunny
5mo ago
clean project, bookmarked it. always nice seeing side projects that actually ship instead of staying in readme-only mode forever.
44.
▲
by
ElenaDaibunny
5mo ago
26M for tool calling is wild. curious how it handles ambiguous function signatures where the correct tool depends on conversational context rather than just the last message.
45.
▲
by
ElenaDaibunny
5mo ago
Unified memory makes this so much cleaner than the Python bridge approach. Curious where the ANE scheduling ceiling is though.
46.
▲
by
ElenaDaibunny
5mo ago
Honestly the bigger question is why we still write glue code at all. Let the agent orchestrate.
47.
▲
by
ElenaDaibunny
6mo ago
Anyway, good luck with the migration. Curious where you land. And honestly? Props for actually following through instead of just complaining on Twitter like the rest of us.