AI in March 2026: agents take the mouse and keyboard
March brought AI agents a step closer to working the way people do: operating ordinary software through the screen, keyboard and mouse. Microsoft folded agents into its core business bundle, Europe moved to give businesses more time on high-risk AI rules, and the automation market showed where enterprise demand is heading. Here is what you need to know.
GPT-5.4 can operate your software directly
On 5 March, OpenAI released GPT-5.4, which it calls its most capable and efficient model for professional work. Its headline feature is native computer use: the model can read screenshots and issue keyboard and mouse commands to work across websites and applications. OpenAI reported a 75.0% success rate on the OSWorld-Verified benchmark, above the 72.4% human baseline it cites, and a big jump on spreadsheet modelling tasks compared with GPT-5.2. The model also supports up to one million tokens of context and is available in ChatGPT and through the API.
What it means for your business: Many organisations still rely on legacy systems with no modern API, which has traditionally been the territory of robotic process automation (RPA). Models that can use a screen like a person open up new ways to automate those systems, and to handle the exceptions that rigid bots struggle with. Benchmarks are not the same as reliability on your own processes, though, so pilot carefully, keep a human in the loop for anything consequential, and log what the agent does.
Microsoft puts agents into its core business bundle
On 9 March, Microsoft announced Copilot Cowork, a feature for handling multi-step tasks across Microsoft 365, built on technology developed with Anthropic and entering research preview in March through Microsoft's Frontier programme. It also announced Microsoft 365 E7, launching on 1 May at $99 per user per month, which bundles E5, Microsoft 365 Copilot, the Entra Suite and Agent 365, a new tool for IT and security teams to manage agents. Agent 365 will also be sold separately at $15 per user per month.
What it means for your business: For organisations that run on Microsoft 365, agents are becoming part of the standard toolkit rather than a separate project. That brings licensing decisions: compare the cost of a bundle with the value of the specific tasks you expect to automate, and do not pay for licences across the whole workforce before you know who will use them. The arrival of Agent 365 also signals that managing and securing agents is now an IT discipline in its own right.
The EU moves to delay high-risk AI rules
The EU institutions took a major step towards postponing parts of the AI Act. The Council of the EU adopted its negotiating position on the Digital Omnibus on AI on 13 March, and the European Parliament followed on 26 March. Both back moving the compliance date for stand-alone high-risk AI systems from August 2026 to 2 December 2027, and to 2 August 2028 for high-risk AI built into products. The institutions then began negotiating a final text, which still needs to complete the legislative process before it becomes law.
What it means for your business: A delay looks increasingly likely, but it is not yet final, and the August 2026 date technically still stands until the changes are adopted. Use the extra time wisely rather than pausing. Mapping where you use AI, documenting decisions and building basic oversight are good practice whatever the final deadline, and they make customer and procurement questions much easier to answer.
UiPath results show automation moving to scale
On 11 March, UiPath reported results for its 2026 financial year, with annual recurring revenue of $1.853 billion, up 11% year on year, and full-year revenue of $1.611 billion. Chief executive Daniel Dines said that as enterprise AI adoption moves from experimentation to scaled deployment, customers need a platform that can execute complex processes "with reliability, governance, and scale". The company is positioning its platform around combining traditional rule-based automation with AI agents and orchestration.
What it means for your business: The most effective automation programmes are not choosing between RPA and AI. They use predictable, rule-based automation where processes are stable, and bring in AI agents for judgement-heavy steps such as reading documents or handling exceptions, with orchestration tying it all together. If you already have bots in place, they are a foundation to build on, not something to replace.
Practical takeaways
Identify legacy systems without APIs where computer-use agents could help, and pilot one with human review.
Audit Microsoft 365 licences and usage before deciding whether a bundle such as E7 is worthwhile.
Assign responsibility for monitoring and securing agents, just as you would for user accounts.
Keep your EU AI Act preparation moving; treat any delay as time to get governance right.
Combine existing RPA with AI agents for the steps that need judgement, rather than starting again.
If you would like a hand working out where agents, RPA and analytics fit together in your organisation, we would love to hear from you. Book a call with Manta Global AI and let us explore what is possible together.
Comments