The Week AI Became an Economic Actor

Something shifted this week, and it was not another model benchmark or product launch. It was something more fundamental. Across finance, enterprise software, research labs, and open source communities, AI crossed a line we have been theorising about for years but had not truly witnessed at scale.

Let me explain what I mean, because the pattern only becomes visible when you line up the week’s headlines side by side.

---

The Researcher

OpenAI announced that its “North Star” for the next few years is building a fully autonomous AI researcher. Not a better chatbot. Not a fancier copilot. A system that can take on research problems independently, run experiments, and write up results over the course of days, with minimal human oversight.

Chief scientist Jakub Pachocki described the near-term goal as “an autonomous AI research intern” by September 2026, with a full-blown automated researcher by 2028. They already see their Codex agent as an early version of this.

The Customer

Meanwhile, Visa revealed that its Intelligent Commerce platform is now facilitating real, end-to-end purchases initiated entirely by AI agents. Not mock transactions. Not sandboxed experiments. Actual money moving through the economy because a piece of software decided to buy something.

Over 100 partners are building in the ecosystem, and Visa predicts millions of consumers will delegate purchasing to AI agents by the 2026 holiday season. McKinsey estimates the global agentic commerce market could hit $3 to 5 trillion by 2030.

Your next customer might not be a person.

The Replacement

And then there is the labour side of the equation. Block, Jack Dorsey’s payments company, cut 4,000 employees, nearly 40% of its workforce, explicitly citing AI’s growing ability to perform a wider range of tasks. This was not buried in a restructuring announcement. Dorsey said most companies will make similar cuts within the next year.

Oracle is evaluating laying off 20,000 to 30,000 people to redirect $8 to 10 billion toward AI infrastructure, driven partly by a $156 billion OpenAI deal requiring 3 million GPUs.

Sam Altman himself acknowledged the shift in a remarkably candid Fortune interview, admitting that “the traditional balance between labour and capital is shifting drastically” and that, frankly, “nobody knows what to do about it.”

---

The Convergence

What makes this week different from the steady drumbeat of AI hype is the convergence. We are not talking about one company making a bold prediction. We are watching the infrastructure being built, the money being redirected, and the headcount being cut, all at once.

• Visa is building the payment rails for an agent economy.
• OpenAI is building the agents that will operate in it.
• Block and Oracle are reshaping their organisations around the assumption that this is not five years away.

It is happening now.

Management, Not Prompting

Ethan Mollick, the Wharton professor who has become one of the most thoughtful voices on AI’s practical impact, offered a critical nuance in his latest writing. He notes that AI cannot easily replace jobs as wholes, because jobs are bundles of many tasks. But the new generation of AI agents, from Claude Code to OpenAI’s Codex, represent something different from chatbots: they are systems you manage rather than tools you use.

That reframing matters enormously, because it suggests the disruption will not look like a light switch being flipped. It will look like a slow restructuring of what every role actually involves, task by task, until the job itself is unrecognisable.

The Open Source Collapse

Add to this the open source story quietly unfolding in the background. March 2026 may be remembered as the month the proprietary moat finally collapsed in meaningful ways.

Over a single week, at least 12 major models launched across the US, China, and Europe. NVIDIA’s Nemotron 3 Super hit 60.47% on SWE-Bench Verified as an open-weight model. GLM-5, released under MIT license, entered the top 5 globally, displacing GPT-5.2 on human preference benchmarks. Models with 49 billion parameters are now achieving results that required 600 billion+ just twelve months ago.

---

So Where Does This Leave Us?

I think we are at the moment where the abstract becomes concrete. For two years, the conversation has been “AI will change everything.” This week, the conversation became “AI is changing this specific budget line, this specific headcount, this specific transaction flow, this specific research pipeline.”

The shift from theoretical to operational is the shift that actually matters, and it is happening faster than most organisations have planned for.

The question I keep turning over: if AI agents are becoming economic participants, who is building the governance, the accountability frameworks, and the organisational structures to manage them?

We have spent years debating AI safety in the context of superintelligence. But the more immediate challenge might be much more mundane and much more urgent:

• Who is responsible when an AI agent makes a bad purchase?
• Who owns the output of an AI research intern?
• How do you performance-manage a fleet of agents?

These are not philosophical questions anymore. They are operational ones, and the companies that figure them out first will have an enormous advantage.

I would love to hear how your company is thinking about this shift. Are you restructuring roles around AI capabilities? Are you preparing for agent-driven commerce? Or does this still feel like something that is happening to other industries?