OpenAI DevDay 2026 and "Dots": How Always-On AI Agents Will Change Business Automation

In brief: OpenAI made more than 20 announcements at DevDay in San Francisco on 29 September 2026. The most talked about is Dots: always-on AI agents inside ChatGPT. According to the event coverage Dots run 24/7, learn from feedback, have their own computer and browser and can connect to over 4,000 apps; they are included in the Pro plan. Other headlines: GPT-6.1 Sol (a claim of near-Astra performance at one-fifth the price), an Ultrafast speed tier in Codex (up to 300 tokens/second, up to 8x faster than standard Codex), the $500/month Pro 500 plan and @ChatGPT in Slack/Teams. The takeaway for businesses: agents are moving from “a feature you try” to “a continuously running, employee-like process,” which brings authority, oversight and cost management along with the opportunity. The information is based on two independent news reports (9to5Mac, Analytics Insight); I could not reach OpenAI's own recap. Details may change.

Last updated: 23 October 2026

What are Dots?

As reported from OpenAI, Dots are always-on agents that take on recurring responsibilities across connected applications:

• They run 24/7 in ChatGPT and learn preferences and feedback.
• They have their own computer and browser and can use thousands of connected apps (4,000+).
• They are powered by Astra, are included in the Pro plan and do not draw down usage limits (event coverage).
• In the collaborative workspace ChatGPT Space people work with pages, presentations and living documents; Business and Enterprise users can summon @ChatGPT in Slack and Microsoft Teams.

Other announcements (as reported)

Some advanced tools are only on Pro, Business, Enterprise or selected developer plans. See OpenAI's official page for general availability and pricing details.

What does it change for businesses?

1. Automation is moving from “flow” to “responsibility”

Classic automation (e.g. “when the form arrives, send an email”) runs on fixed rules. Agents like Dots take over a recurring responsibility (preparing the weekly report, sorting incoming requests, checking invoices) and act across connected apps. That means more flexible but less predictable.

2. The speed and cost balance is shifting

Options such as Ultrafast and the cheaper Sol aim to make agent-based workloads economical. But agents that run 24/7 create continuous cost; usage and spending limits must be defined. For model choice, see our model selection framework.

3. The authority problem grows

An agent having access to 4,000 apps means a risk of data leaks or unintended actions if authority is given wrongly. Per-agent identity, least privilege, time-limited access and human approval are needed: see our 8-controls guide.

4. Integrations become critical

An agent's value comes from being able to connect to your company's systems (CRM, ERP, email, calendar). Where ready-made connectors fall short, custom integrations are needed. See our system integration guide.

5. Vendor dependence

If you build everything inside one provider's agent environment, you depend on that provider's price and policy changes. Common standards reduce this risk: Agent Skills and the Agentic AI Foundation.

Where to start with agent automation

1. Choose a low-risk, highly repetitive process: such as a weekly report summary or sorting incoming email.
2. Start read-only: Let the agent only read data and produce suggestions first; grant write authority later.
3. Require human approval: Anything going to customers, involving money or deleting should need approval.
4. Log everything: Track what the agent did and why.
5. Set cost limits: Define daily/monthly usage and spending caps.
6. Measure: Time saved, error rate, rework rate. See ROI on AI investments.
7. Write a policy: What data may be given to an agent? See our AI use policy guide.

What not to exaggerate

• An “always-on” agent is not a flawless agent. Outputs need verification.
• Price and performance claims are the company's own statements; test them on your own jobs.
• Details of some API announcements (Decisions API, computer use in the Agents API) have not yet been shared.

Agent and automation projects with Aktaş Digital

We plan projects that transfer your business processes to AI agents safely, build custom integrations with your systems and design the authority/oversight layer. To get a scope and cost breakdown, use our quote form. Related services: AI solutions.

Frequently Asked Questions

Are Dots for everyone?

According to the event coverage, Dots are included in the Pro plan; some advanced tools are only on Pro, Business, Enterprise or selected developer plans. See OpenAI's official page for current access.

What is the difference between Dots and classic automation (Zapier-style)?

Classic automation runs on fixed rules; given a goal, agents can plan steps themselves and move across apps. Flexibility rises and predictability falls, which is why oversight matters.

Is an agent running 24/7 safe?

Safety depends on the authority you give and on monitoring and approval mechanisms. I wouldn't recommend one without least privilege, logging and human approval.

Should I use GPT-6.1 Sol right away?

Test it on your own examples first. The company's “one-fifth the price” claim is its own statement.

What does it mean for SMEs?

It can save time on low-risk, repetitive processes; but it needs setup, integration and oversight effort. Starting with a small pilot is safest.

Conclusion

DevDay 2026 shows AI agents moving from “helper” to a layer of “continuously working labour.” The opportunity is large, but authority, oversight and cost discipline are essential. The first thing you can do today: pick a repetitive process in your team that could be delegated to an agent and list in writing which data it would need to access.