Agents are the new PCs

At GTC 2026, Jensen Huang framed OpenClaw as the operating system for agentic computers. That feels like the right mental model for what agents actually are becoming.

Agents = PCs

Over the past 12 months, "agent" has been stretched to mean almost anything: a prompt, a wrapper, a workflow, even a markdown file. It's like calling a Facebook page a SaaS app.

We need to sort this out...

My definition of an agent is simple: software that can operate software on your behalf.

Agents can do this because they hold a few core components (and this is where the PC analogy begins) - memory, tools (i.e skills), they can manage resources and take action.

There's actually a lot of AI SaaS apps today that don't meet this definition - like most use-cases of ChatGPT or even the legal platform Harvey (which I'm sure will change).

Think OpenClaw or Claude Code. These don't replace your computer so much as sit on top of it: operating your terminal, your file system, your apps, your browser, your tools. They're less like standalone apps and more like a new execution layer for the software you already use.

The GUI let humans use computers.

Agents let computers use computers.

This is probably why Nvidia just launched their own enterprise version of OpenClaw called NemoClaw.

What does this mean?

When I worked at Microsoft, they used to proudly trumpet the founding mission of "a PC on every desk". Well we might be looking at a future that involves 'an Agent on every PC, on every desk' - admittedly it's more of a mouthful.

But the idea makes sense to me.

Individuals will have a personal agent which is unique to them ✅

However, will businesses also have an agent which is bespoke to them? Or will there be a modern day Steve Jobs who births the ultimate horizontal or 'god' agent to handle all company work/processes/tools without the need for customisation?

Today, it feels far more likely that every company will need to build its own agent. That's what we're doing at Flare. That makes sense because companies differ wildly in their data, tooling, controls, standards, customers, and ways of working.

In the same way every company has its own operating procedures, every company will end up with its own operating intelligence.

Your company agent becomes a new frontier of competition. In the same way you can't just buy a company's workflows off the shelf, you probably won't be able to buy an agent that works perfectly out of the box for your exact mix of customers, people, tools, data, and operating constraints.

Don't get me wrong: the holy grail would be a universal agent that could work inside any company. Maybe that becomes possible. But from where I sit today, it still looks unlikely. I see this first-hand with AI accounting "agent" companies.

I see this first hand with the AI accounting 'agent' companies. None of them seem like a perfect fit to our accounting firm and workflows. It seems like we're being sold the old fashioned SaaS thinking of having to change our business to fit the software.

The whole promise of AI was that the software could fit the business, wasn't it?

My bet is that companies that build proprietary agents around their own workflows, tools, and standards will outperform those that rely entirely on generic ones.

We're still early.

Play the analogy forward and we still haven't built the internet or cloud layer for agents. Today's agents feel like early PCs: mostly local, slightly awkward, undeniably powerful, and full of trapped potential.

We're starting to see more cloud-based agents emerge. At Flare, that's where we've focused too, because most of our accountants' tools (Xero, Google Drive etc.) already live in the cloud. But even there, it still feels early. We still have to invent the agent-native layers for identity, permissions, payments, discovery, search, and agent-to-agent communication.

The PC era put a computer on every desk.

The agent era may put a company-specific agent inside every business.

The biggest winners won't just adopt agents - they'll build their own.