OpenAI’s new “dots” are always-on agents that pursue continuing objectives across applications and communicate through tools such as Slack and Microsoft Teams. Powered by GPT‑6 Astra, they can use Codex and ChatGPT Work while retaining project context over time.
What changed
OpenAI has moved beyond request-by-request agents toward persistent workers that monitor and advance projects continuously.
Why it matters
Long-running agents require durable state, bounded permissions and recovery from partial or duplicated actions.
Practical takeaway
Store each agent goal, checkpoint, tool call, approval, retry key and resulting artifact as a durable workflow record.
Analysis
Demonstration glitches suggest that persistence and cross-application reliability remain harder problems than model reasoning alone.
Published
29 September 2026
