OpenAI is moving its agentic AI offering beyond the familiar pattern of asking a question, receiving an answer and starting again. Its dots product, introduced on 29 September 2026, is designed around persistent agents that can carry context between conversations, work towards ongoing goals and use connected tools. A late-October update added a texting beta for eligible US users, giving the September launch a fresh follow-up. The important business story is not the arrival of another chat window. It is the possibility of assigning a continuing responsibility to an AI system, with a person still setting its scope and checking its work.

For Irish organisations, that shift creates a useful discussion about everyday operations: gathering information for a weekly review, following up a software issue or preparing material for a project team. These are potential applications, not evidence of results already achieved by Irish customers. Nor is access universal. OpenAI's published guidance places regional and workspace conditions on dots, including restrictions on the individual Pro rollout in the European Economic Area. As November begins, the sensible approach is to understand what has actually launched, establish which services a business can use, and test one clearly defined responsibility before expanding it.

From One-Off Requests to Ongoing Responsibilities

OpenAI describes dots as powered by GPT-6 Astra, with their own cloud computer and the ability to learn from feedback over time. The company's announcement says they can work towards goals around the clock and connect to more than 4,000 apps through its plugin ecosystem. They can be reached through ChatGPT, Slack or Teams. Those are the supplier's product descriptions, not an independent assessment of reliability. The size of the plugin catalogue also does not mean that every connection is enabled in every organisation, or that an agent automatically has permission to use all the information in those systems.

The distinction from a conventional assistant is continuity. Instead of requiring a fresh explanation every time, an agent can retain relevant context and return to an agreed task. OpenAI's examples include investigating bugs, turning designs into applications and tracking planning work. It also describes preparing an invoice that was sent after the user's approval. That last detail captures the useful boundary: preparing work and committing an organisation to an action are different responsibilities. An agent may help with both, but the permission to move from one to the other should be explicit rather than assumed.

The launch should also be separated from the longer-term vision. OpenAI describes starting with a primary dot and presents teams of dots working together as an eventual ambition. That is not grounds to claim a fully autonomous digital workforce is already generally available. For a buyer, the immediate opportunity is more modest and more understandable: a persistent assistant that can maintain a thread of work. Whether it delivers value depends on the quality of the instructions, the available information and the process for checking what it produces. Continuity is a capability to test, not a substitute for accountability.

What the Rollout Means for Ireland

Availability is particularly important for an Irish readership. OpenAI's setup guidance says dots are rolling out to Pro users aged 18 and over outside the EEA, Switzerland and the UK. Ireland is in the EEA, so an Irish individual should not assume a Pro subscription provides access through that rollout. The same guidance describes Business Premium access in supported ChatGPT regions, subject to workspace eligibility. Enterprise users, including education and healthcare workspaces, can try the beta when an administrator enables it; access is initially off by default. These are distinct routes, not interchangeable promises of availability.

Workspace settings, encryption and data or inference residency requirements can also affect eligibility. A company should therefore confirm its own access and contractual requirements before designing a process around dots. The first dot is described as included in eligible Pro or Business Premium plans, but deeper work has an allowance. Around-the-clock capability is not the same as unlimited work at no additional cost. Without a verified numeric allowance or a complete cost estimate for a particular workload, it would be misleading to present the service as a fixed-price replacement for a member of staff.

The October 28 texting update illustrates the rollout's pace without changing those Irish access conditions. OpenAI's release notes describe a beta starting for Pro users in the US through iMessage or RCS, including proactive updates. The dot can text its owner, rather than becoming a general messaging agent for everyone in the owner's contacts. It is a current example of bringing an ongoing agent into an everyday communication channel, not an announcement of Irish texting availability. Businesses can follow that development while keeping their own decisions grounded in features and regions they can actually use.

The Developer Route Is a Separate Product

Alongside dots, OpenAI has a separate route for organisations building their own agents. Its Agents API entered public beta on 10 September 2026, before the DevDay announcements later that month. OpenAI describes it as a way to build and run cloud agents with the Codex harness, with execution infrastructure managed by OpenAI. The offering includes support for long-running work, context management, tool use and coordination of subagents. It is developer infrastructure, not a feature that a company automatically receives simply because its employees hold ChatGPT subscriptions.

The official computer-use documentation describes agents navigating websites and interacting with browser interfaces in an OpenAI-hosted environment. Potential uses include testing a website, collecting information or operating an application through its existing interface. This can be useful where a workflow spans systems that do not offer a convenient direct connection. But seeing and operating a browser is not the same as understanding every business consequence of a click. Applications still need to handle website access and sign-in, verify the results and review the recorded browser activity.

For an Irish business, the choice is therefore partly about ownership. A packaged agent may suit an eligible team looking for a continuing assistant within its existing work tools. A custom application may suit an organisation with a specific workflow and people able to build, test and maintain it. Neither route removes the need to agree what success looks like. OpenAI's infrastructure descriptions are not a guarantee that every process will run unattended for days. The useful question is whether the chosen approach can complete a defined task consistently, at an acceptable cost, with a clear route back to a human.

Make the First Pilot Small and Measurable

A good first pilot should have a narrow responsibility and a named owner. Preparing a draft weekly project summary from approved material is easier to assess than asking an agent to manage an entire department. The team can compare the draft with its existing process: whether important facts were captured, how much correction was needed and how long review took. These are suggested measures for a trial, not performance claims for dots. Count the work required to connect tools, explain the task and check the output, rather than treating the agent's generation time as the whole calculation.

Permissions should match that responsibility. An agent preparing a summary may need read access to selected documents but no ability to send messages or change records. OpenAI's safety guidance says dots use plugin permissions and approval defaults, with additional custom boundaries available. It distinguishes proactive research, which can read permitted sources and save private notes, from tools that directly send messages, alter plugin content or control a computer. Later actions still follow normal approval and safety rules. That separation gives organisations a practical way to introduce initiative without immediately handing over consequential decisions.

The review process should be equally concrete. Decide who approves external communication, who checks figures and what happens when information is missing or instructions conflict. Keep a record of corrections so the team can see whether the workflow is improving. Expansion should follow evidence from that limited task, not the excitement of the launch. A useful agent can create more time for judgement and customer work, but those benefits have to appear in the organisation's own results. The positive opportunity is to make routine coordination more dependable while keeping responsibility for business outcomes with the people running the business.

The Bottom Line

OpenAI's dots make persistent, proactive AI a more concrete business proposition, while the Agents API provides a separate development route for custom cloud agents. The September launches and October texting update show movement towards ongoing work rather than isolated answers. For Ireland, the opportunity comes with a straightforward qualification: confirm regional and workspace access, understand the allowance and give the agent only the permissions its task requires. Start with one responsibility that can be measured and reviewed. The strongest case for agentic AI is not that it removes people from the process, but that it helps them get useful work done with clearer oversight.