Astra review: executes hour-long agentic tasks while iterating on the plan without losing the thread
ivan_bezdomny · x · 2026-09-06
ivanbezdomny shares hands-on impressions of Astra, highlighting its long-horizon agentic behavior:
- GPT-5.5 and even 5.4 could loop for hours but got lost when they had to iterate or adjust the plan mid-run; Claude tends to take shortcuts on longer work.
- Astra is a GPT-like workhorse: it works for hours, iterates, tightens things up, and fixes problems without losing the thread.
- Caveats: it occasionally stopped abruptly (a simple "keep going" loop would help) and is probably still weak at UI.
Verdict: give it a sound, logical plan and it can work on it for hours — possibly yielding better results than ever.
More from coding & agent
- Add one question to PR walkthroughs: "What did you almost build instead?" — HaktanSuren · 2026-09-06
- Sam Altman: stop grinding prompts, build loops and graphs around LLMs instead — goyalshaliniuk · 2026-09-06
- "It's already telling me what to do": dev jokes about the ghost living in his computer — BLUECOW009 · 2026-09-06
- Agentwiki update: no AI agent has discovered the agent-only wiki on its own yet — andersonbcdefg · 2026-09-06
- agentwiki.fyi: a GET-only, ASCII-text wiki experiment built for AI agents to share knowledge — andersonbcdefg · 2026-09-06
- Miles Brundage: Runway + AI agents is insanely overpowered now — Miles_Brundage · 2026-09-06