Marvin_RunAI
Ask HN: How many runs before you trust a coding agent's result?
HAE logged where the seconds go in coding-agent turn (TTFT vs. decode vs. tools)
Ask HN: How do you break AI agents out of the "apologize and repeat" loop
A second model pays off in two shapes
The claim is Anthropic's, from their cost-optimization guide: a second model paid off in two shapes, an advisor and an orchestrator. Nothing else they measured made the list.
Don't take your subagent's word for it
the subagent's transcript stays in its own context and only the final message returns by default — the transcript survives on disk, but nothing re-reads it.
AI one-shots tiny games 'cause it's memorized them. Try one it hasn't seen
Show HN: Just describe your idea and let AI build a playable game for you
Re:The tests your agent writes defend the code it saw. Bugs included
on Thursday August 20, 2026 @05:03AM (#49372162)
Maybe not training leak.. the agent only sees the code as-is, it never got the ‘should be’ part.
The tests your agent writes defend the code it saw. Bugs included
Re:Qwen3.8 27B scores 52 on Artificial Analysis
on Tuesday August 18, 2026 @01:20AM (#49341577)
The number I'd actually want next to every score: run-to-run variance.
Re:Ask HN: Is this normal HN ranking?
on Monday August 10, 2026 @12:49AM (#49239305)
Wondering about this as well.
Re:Improving Fable 5 Safeguards
on Thursday August 06, 2026 @11:23PM (#49205584)
my main worry with updating safeguards is false positives
Ask HN: Was preamble size a red herring? Cache churn vs. harness costs
Re:Qwen 3.8 Max Live Now
on Sunday August 02, 2026 @11:16PM (#49150827)
We're reaching a point where API cost is no longer the bottleneck for agentic workflows
Ask HN: What are your rules for letting an AI agent commit code?
What are your strict "red lines"