OpenAI Dots: The Always-On Agent That Works While You Sleep

## Table of Contents

1. What are Dots?

2. How Dots is different from ChatGPT

3. Why always-on matters

4. The hybrid architecture: Dots + Jev + Code

5. 5 ready-to-use Dots prompts

6. Limitations

7. Conclusion

## 1. What Are Dots?

OpenAI hosted its annual DevDay 2026 in San Francisco on September 29, announcing Dots as 'always-on AI agents that work 24/7 and act on your behalf rather than waiting for prompts'.

Instead of chat -> response, Dots runs continuously: monitors inbox, code repo, analytics, and takes predefined actions.

## 2. How Dots Is Different From ChatGPT

ChatGPT is turn-based: you prompt, it generates. Dots is state-based: State -> Focused question -> Typed decision -> Action.

It's closer to Jev AI's philosophy: not every task needs generated text, sometimes you just need a decision.

## 3. Why Always-On Matters

If your agent makes one decision, latency is irrelevant. If it makes 100,000 classification, routing, or retry decisions, speed and cost dominate.

TypeSafe reports Jev operating 40x to 200x faster than LLM baselines for such workflows, which is why decision models pair well with always-on agents.

## 4. The Hybrid Architecture: Dots + Jev + Code

User intent -> Dots (generative) proposes action -> Jev (decision) evaluates: should we send? retry? review? -> App code validates permissions and executes.

This separates responsibilities: generation, decision, execution, and human review.

## 5. 5 Ready-To-Use Dots Prompts

Prompt 1: 'Monitor my support inbox. If urgency > 0.8 and category = billing, draft a reply but flag for human review.'

Prompt 2: 'Every hour, check my GitHub PRs. If tests fail, classify failure type and suggest fix.'

Prompt 3: 'Watch my blog analytics. If traffic drops >20% on a trending post, generate 3 new titles.'

Prompt 4: 'From 20 retrieved docs for a RAG query, score relevance and keep top 3.'

Prompt 5: 'Before publishing any AI-generated article, verify it meets editorial rubric [criteria].'

## 6. Limitations

Always-on means always-cost. You need guardrails, idempotency, and verification. Don't treat model probability as guarantee. Use code for truth, not the model.

## 7. Conclusion

Dots moves us from chatbots to coworkers. The winners won't be those with the biggest LLM, but those who architect generation, decision, retrieval, and deterministic logic correctly.