This is the first of our monthly AI round-ups: not everything that happened, nobody needs another firehose, but the handful of developments a business leader should actually register, and what, if anything, to do about each.
March's theme was unmistakable: AI agents stopped being a research story and became a platform story. The tooling to build systems that plan and execute multi-step work is now being shipped, packaged and sold by the biggest vendors in enterprise software.
Agent platforms, from every direction
NVIDIA launched an open toolkit for building autonomous agents (reasoning, acting and completing multi-step enterprise tasks) with secure runtime environments and agent blueprints designed to mix open and frontier models. Alibaba introduced Wukong, an enterprise platform for managing fleets of agents handling document work, approvals and research inside existing business tools.
The significance isn't any single product. It's that agent infrastructure is now a category every major vendor wants to own, which means the question facing organisations is shifting from "is this real?" to "which of our processes would we trust to it, and under what controls?" That second question is the one worth spending time on now.
The readiness gap, quantified
The most useful number of the month came from a global survey of more than 1,600 business leaders: 85 per cent of enterprises reportedly aim to be running agentic AI within three years, and 76 per cent admit their operations aren't ready to support it.
That gap is the story. Ambition is cheap; readiness is data quality, documented processes and governance, the unglamorous groundwork we bang on about in every AI conversation. The organisations that spend this year on foundations will deploy agents quickly when they choose to. The ones that don't will buy licences and stall.
Regulators are watching the same space
In the US, a bipartisan group of senators moved to establish a federal commission on the economic effects of AI, and device makers kept pushing capability downmarket: Samsung announced plans to double its Gemini-equipped devices to 800 million units by the end of the year, putting capable on-device AI in mid-tier hardware.
For UK businesses the direction of travel matters more than the detail: AI capability is becoming ambient and regulated attention is rising with it. Systems built now should assume both. Auditability isn't a feature to retrofit.
What to do with all this
Our read on the month, as actions rather than headlines:
- Pick the one process you'd want an agent to run, and map it properly: boundaries, exceptions, approvals
- Audit whether your data would support that agent today; if not, that's the project
- Treat agent platforms as components to evaluate, not commitments to make
- Build audit trails into anything AI touches now: regulation is heading that way everywhere
