Users slam Dots: permission blocks, dead sub-agents, and "unlimited Astra" quietly burning local Codex limits
brandon_galang · x · 2026-10-03
Developer brandongalang quote-posted a detailed rundown of his frustrating experience with Dots (built on Astra / the Codex harness):
- Permission classifiers keep blocking explicitly requested actions no matter how instructions are tweaked
- The Dot computer is "effectively useless": it almost always delegates to local Codex, burning the user's own limits — so "unlimited Astra" isn't real
- Tasks Codex Remote happily does get refused by Dot; some sub-agent threads simply never return updates
- Pages sound useful but Dots often can't save or write to them — and neither can the user
- Agency feels wildly inconsistent across accounts: some Dots are hyper-proactive, his "just kinda there"
He says he's had magic moments, which is why the disappointment stings — this is the result of "6 months of plans moved up thanks to Astra." A snapshot of how immature permissioning and task handoff still are in agentic products.
More from coding & agent
- ChapterPal dev: frontier vision models consistently fail to spot obvious webpage conversion artifacts — burkov · 2026-10-03
- Dev finds Argon enough for nearly all coding tasks, misses it after switching to Opus 5.5 — m2saxon · 2026-10-03
- Air-gap file transfer via animated QR codes flashing at 10-30 frames per second — Thionne_WTZ · 2026-10-03
- Rust no longer the de facto best language for agents, dev argues — zack_overflow · 2026-10-03
- Knowledge search beats data connectors for agents: up to 88% faster, 95% cheaper — williamLberman · 2026-10-03
- INT21: 2 engineers direct AI to build 20 inference engines in 2 weeks, up to 2.4× faster than SGLang — bingxu_ · 2026-10-03