On September 29, at the DevDay 2026 conference, OpenAI unveiled dots — a new class of autonomous AI agents built on the GPT-6 Astra model that, unlike an ordinary chatbot, don't wait for a prompt: they run "always on" in the background, pursuing a user-defined goal for hours or days without constant supervision. The company itself describes them as agents "built to handle everything." For now, access is limited to ChatGPT Pro and Business Premium subscribers in a limited set of countries, through ChatGPT or Codex; SMS support has been promised for later.
What sets dots apart from "yet another chatbot with memory"
The key difference isn't the model — it's the mode of operation. A regular agent in ChatGPT runs a task and stops, waiting for the next command. Dots are designed to work continuously: monitoring the flow of customer tickets and fixing small bugs on their own, re-running analyses when anomalous data shows up, tracking metrics and reacting to deviations without a human in the loop. Agents can be connected to Slack, Teams and internal company systems, and "specialist" dots can be given their own credentials and access limited to specific tools — not the entire infrastructure at once. OpenAI also announced integration with Microsoft's Agent 365 access-control system — a clear signal that the company is preparing for a world where agents hold real keys to real systems.
Picture a small online clothing shop in Porto with three employees. A dot connected to the support inbox could sort through shipping-delay complaints on its own, reply with templated but personalized messages, and send the owner a weekly summary: how many complaints came in, which cause keeps repeating (say, the same courier missing deadlines over and over), and suggest a concrete action — switching couriers for a specific region. This isn't science fiction or a vague "benefit for business" — it's exactly the kind of routine-but-context-heavy work that used to require a dedicated support hire.
There's a detail easy to miss behind the flashy demo: OpenAI doesn't talk about a single agent but about "teams of dots working on your behalf" — meaning the intended scenario isn't one employee replaced by one bot, but a swarm of narrowly specialized agents, each with its own access and area of responsibility. For that same shop, this could mean a separate dot for handling complaints, another for monitoring warehouse stock, and yet another for weekly sales analytics by channel. It sounds powerful, but managing a swarm of several autonomous processes, each capable of emailing customers or changing data on its own, is objectively more complex than running a single chatbot with clear boundaries — and that operational complexity rarely makes it into the demo reel.
Where the real breakthrough ends and the repackaging of existing tech begins
Worth being honest here: the functionality underlying dots largely already existed — developers could assemble similar workflows before, through Codex and comparable agentic harnesses. Dots are, first and foremost, a convenient package, a friendly interface (literally a cartoon dot avatar) and a unified platform, rather than a fundamentally new type of intelligence. A second thing worth weighing before adopting it: an agent granted credentials and access to email, a CRM, or Slack expands the attack surface — the more systems an agent can touch on its own, the higher the cost of a mistake or of a single access token being compromised. And third, the billing model: an agent that runs around the clock instead of on request turns AI spend from a predictable subscription into a variable line item that needs separate monitoring. Access is already restricted to the pricier Pro and Business Premium tiers, so for very small businesses this is more a signal of what's coming than a tool to deploy on Monday.
It's also worth keeping the competitive context in mind: OpenAI is far from the only player betting on autonomous agents — Anthropic, Google and other labs are moving in the same direction, and over the next year the market will likely see not one dominant agent for every use case, but several competing platforms with varying degrees of autonomy and different models of accountability for an agent's mistakes. For a business, that means picking a vendor today isn't a decision for life — it's a bet that will probably need revisiting within six months to a year, once it's clearer whose agent ecosystem is actually reliable in production rather than just on stage.
At Dayava we set up similar automation scenarios — agents that sort through requests, track metrics, and handle routine tasks without human involvement — but we do it for a specific client process, with access controls built in from day one rather than bolted on afterward. If a similar solution interests you for your business, leave a request at dayava.pt/contactos/.
Source: TechCrunch, "OpenAI launches Dots, its bubbly agentic avatar"