Project management edition
OpenAI and other companies are putting AI agents to work. The project team remains responsible for oversight and control.
OpenAI puts AI agents in shared project spaces
AI agents can reschedule appointments and change code. That is why they need clear limits and oversight.
On 29 September, OpenAI presented more than twenty new products. These include ‘dots’. They are AI agents that can work continuously. They can reschedule appointments, book flights or change code on your behalf. OpenAI also announced ChatGPT Space. It is a shared workspace. Teams, ChatGPT and agents work together on projects there. Plugins are also on the way. These are add-ons that work as full apps in ChatGPT. For project managers, the key change is the move from answering to acting. An agent carries out a task. It does not just suggest text. This can save time when coordinating work. But it also changes who can alter your schedule or work environment. The launch comes at a time when people are concerned about oversight. Can people properly control independent systems? So test an agent first with a small, clearly defined task. Agree who will check its actions. An agent receives an instruction and then takes steps on its own to carry it out. OpenAI gives rescheduling appointments, booking flights and changing code as examples. ChatGPT Space brings teams, ChatGPT and agents together in the same workspace. This puts tasks and collaboration in one place. The source also mentions a Decisions API. An API is a link that lets programmes communicate with each other. Businesses can use this link to leave small, repeated decisions to AI. There are different levels of independence. Sometimes AI makes a choice within a process. Sometimes it takes action in a calendar or in code. Decide what information and actions the agent needs for each task. The source reports concerns about controlling independent AI. But it gives no information about permissions, logs or undoing actions. A log is a record of what the agent has done. So do not assume that controls exist if the source does not describe them. Greater independence can save time. But mistakes can also have direct consequences. A wrong appointment is actually rescheduled, rather than merely suggested incorrectly. Decide in advance which tasks an agent can carry out on its own. Also decide which tasks need a person’s approval first. Agree who will step in if something goes wrong. Start small and check that the agent does what you expect.
Background for subscribersThe reports show that AI can do more than create text. Agents can take actions. Other features make work repeatable. Agree in advance which tasks AI can carry out on its own. Also agree who will check the results. And record how to verify unexpected requests through a second channel. Also measure where work is held up. Plan training for people who will use AI.
A voluntary safety agreement leaves oversight to companies
Trump and six AI companies signed a voluntary safety agreement. The companies promise internal checks on their most powerful models. They also promise checks by external auditors. These are inspectors from outside the company. A safety committee will also be added to their boards. The agreement has no penalties, enforcement or deadline. Companies choose their own auditors. For project managers, the main point is the difference from mandatory rules. A supplier’s promise is not an obligation you can rely on. According to the source, the EU AI Act remains legally binding. This is the European law on AI. Ask your supplier which checks apply to the AI feature in your project. The companies promise checks on their most powerful models. External auditors will be allowed to review them. A safety committee will be added to the board. But the agreement sets out no penalties, enforcement or deadline. Each company also chooses its own auditor. So how it is carried out depends on what each company does. There is no regulator to enforce compliance. The source compares this with the European AI Act. According to the report, that remains legally binding. A voluntary agreement can encourage companies to do more to improve safety. But it does not provide a firm guarantee for your project. The source mentions no shared standard for measuring safety. It also gives no public results from checks. So ask about specific measures for the service you want to use. Simply saying that a company supports the agreement tells you little. Do not make a non-binding promise a firm project requirement. Do so only if your supplier has confirmed it.
Background for subscribersThis is a taster.
Subscribers read the whole Project management edition: every story, each with its background in plain language.
Subscribe →Not ready for that? Read the free letter first →
AI and tech, made simple.
This is the Project management edition. Subscribers get the background to every story, in plain language.
Subscribe
inklaretaal · Amsterdam · © 2026