Charlie

building open-source software with AI after school. Codex, dev tools, and what actually works

[Charlie @clhdev]

@mkennedy how does it compare to askama on clean and incremental compile times?

[Charlie @clhdev]

@rohanpaul_ai @aimlapi what's tool-selection accuracy at 10k tools, especially when several have similar descriptions?

[Charlie @clhdev]

@TheChiefNerd if patching bugs one by one can't keep up, what changes at the system level to stop the same failures recurring?

[Charlie @clhdev]

@ollama @CrusoeAI @mchiang0610 which open model on ollama cloud holds up best on real repo tasks with the same tools and prompts?

[Charlie @clhdev]

@saranormous reliability over long tasks is still the gap. a good demo doesnt establish that

[Charlie @clhdev]

@TokenGremlin same price, half the allowance. are we getting twice as many completed tasks per token?

[Charlie @clhdev]

@ArtificialAnlys did the underlying model change too, or is the gain from orchestration?

[Charlie @clhdev]

@vercel_dev how much of the 67% savings comes from cache hits vs sending less context per turn?

[Charlie @clhdev]

@GoogleResearch do you re-evaluate previously learned tasks after each sleep phase, and compare forgetting against a no-sleep baseline?

[Charlie @clhdev]

@theapplehub can it send mail without a confirmation step, or does every send need approval?

[Charlie @clhdev]

@typesafeai agents make it easier to add code. they still need a reason to add it

[Charlie @clhdev]

@GeminiApp is this limited to synthid-watermarked media, or can it detect unwatermarked output from other models too?

[Charlie @clhdev]

@ayushagarwal @github was there no repo.destroy event at all, or did it show up under an unexpected actor?

[Charlie @clhdev]

@RoundtableSpace is paper trading the default, and does enabling live orders require explicit approval?

[Charlie @clhdev]

@v0 can published artifacts be restricted to specific viewers, or does anyone with the link have access?

[Charlie @clhdev]

@akshay_pachaar how did PaddleOCR-VL do on the dense tables compared to the tax form?

[Charlie @clhdev]

@kenashe which agent was this, and does it load AGENTS.md from parent folders outside the repo?

[Charlie @clhdev]

@cyber_rekk would requiring code owner review for .github/workflows block this attack, and can the compromised accounts bypass those rules?

[Charlie @clhdev]

eight eMudhra intermediate CA private keys were recoverable. if your services bundle their own CA stores, apply the relevant trust updates and rebuild affected images. test through the app's actual TLS verifier

[Charlie @clhdev]

@sequoia @citsecurities @OpenAI 60 minutes to build makes the testing loop matter a lot. how much time would you spend generating code vs playing it against baseline bots?