Every Company Needs an AI Agent Strategy Now

Something shifted in Q1 2025, and if you weren't paying attention, you might have missed the inflection point. OpenClaw—an open-source AI agent that can actually control your computer—went from technical experiment to genuine productivity tool in a matter of weeks. And now? Nvidia CEO Jensen Huang just told the entire software industry they need an OpenClaw strategy.

Let that sink in. The CEO of the world's most valuable company just declared that AI agents controlling computers aren't a novelty—they're infrastructure.

The OpenClaw Moment

Kevin Simbach from Deli Labs captured the shift perfectly: "Before OpenClaw, agents were mostly technical experiments that produced nothing more than timeline sllo. After OpenClaw and with the advent of Opus 45 and 46, agents became accessible, just a telegram message away, always on, actually doing helpful things."

OpenClaw proved two things simultaneously, according to Simbach: People don't want AI chat—they want to get work done. And giving an LLM broad access to your machine is "both insanely useful and mildly terrifying."

That tension? That's basically the entire story of Q2.

The Clawification of Everything

The moment OpenClaw showed what was possible, the race was on. And I mean on. We're seeing variations like Nanobot, ZeroClaw, Picoclaw trying to simplify the complexity. Then there's Open Fang, Hermes, and Ironclaw focusing on security through self-hosting. Everyone's trying to solve a different piece of the puzzle.

But the really interesting moves are coming from established players who are basically saying: "Cool proof of concept, now watch us make it enterprise-grade."

Notion introduced custom agents with deep integration into where companies already store their information. Perplexity launched Computer—a complete reimagining of their platform as what CEO Arvind Srinivas calls a "problem solution design system." His argument? The full potential of AI agents requires the complete canvas of what your computer offers, bridging local files to cloud systems.

Perplexity's also released Computer for Enterprise (operating within Slack with connections to 400+ applications) and Personal Computer (an always-on local version). They're not tiptoeing into this space—they're sprinting.

The Desktop Agent Wave

Yesterday alone brought three major announcements that all converged on the same idea: the agent lives on your machine.

Manis (acquired by Meta in December) launched their desktop app with a feature called "my computer." Use cases include organizing thousands of photos, renaming invoices, even building desktop apps in Swift with no manual coding. They're explicitly acknowledging what became obvious with OpenClaw: "Your most important work happens on your own computer. Your project files, development environments, and essential applications all reside locally, not in the cloud."

Adaptive introduced their own version, pitching it to—wait for it—hardware store owners. Seriously. Their example: drag in a spreadsheet with 47 new products, tell the agent to add them to Square, and it handles the rest. Then it remembers how Square works, how your catalog is organized, how you prefer things done. Next time you ask for a daily sales report, it just... knows.

There's something fascinating about bleeding-edge AI companies targeting the hardware store use case. It suggests they're seeing something beyond the developer crowd.

Nvidia's Enterprise Play

But the biggest move came from Nvidia. Jensen Huang introduced NemoClaw at their GTC event—not a standalone agent, but a security and privacy layer built on top of OpenClaw. It gives agents an isolated sandbox to work in while formalizing access control through policy-based security and guardrails.

OpenClaw creator Peter Steinberger called it "a huge step towards secure agents you can trust."

The response has been notably positive, even from skeptics. Kevin Simbach, who's "been pretty vocal about OpenClaw not being enterprise ready," wrote: "The concept of an agentic workforce is a killer and enterprises are going to want it, so this may be what really kicks it off."

One data point worth noting: The AI Daily Brief runs two programs—Claw Camp (free, self-directed OpenClaw setup) and Enterprise Claw (managed six-week sprint for companies). When they gave enterprise participants the choice between learning OpenClaw or generic agent building with other tools, it split 50/50. Even without enterprise-grade features, demand exists.

OpenAI's Refocus

Meanwhile, OpenAI is having its own moment of clarity. The Wall Street Journal reports that CEO of applications Fiji Simo told staff: "We cannot miss this moment because we are distracted by side quests. We really have to nail productivity in general and particularly productivity on the business front."

This is a big shift from Sam Altman's approach of betting on multiple internal startups simultaneously—Sora, Atlas browser, the Jony Ive device. Now they're converging on the same territory as Anthropic: agentic coding that expands into broader enterprise knowledge work.

Their latest move? Native sub-agent integration in Codex. The pitch: "You can accelerate your workflow by spinning up specialized agents to keep your main context window clean, tackle different parts of a task in parallel, steer individual agents as work unfolds."

One developer described the upgrade: Instead of creating 100 different custom agent roles, you can just prompt your agent to spawn whatever model or reasoning level you want with natural language. One agent for code review, another for test coverage, another for edge cases—all orchestrated by a main agent that delegates.

The numbers suggest it's working. OpenAI President Greg Brockman reported that GPT-5.4 hit 5 trillion tokens per day within a week of launch—"handling more volume than our entire API 1 year ago" and reaching an annualized run rate of $1 billion in net new revenue.

The Complexity Question

Here's what I keep thinking about: conventional wisdom says you simplify for mass adoption. Make it easier, dumb it down, reduce friction. But OpenClaw—the thing that actually broke through—is incredibly complex. It's giving an AI broad access to your entire computer. That's not simple. It's not dumbed down. And it's working.

So what's the right complexity level for Q2? Is there even a single answer? Or are we looking at a spectrum where different users need different complexity bands—some want the full power, some want the curated experience, some want something in between?

What I know for sure: Q1 was about proving agents are viable. Q2 is about making them enterprise-ready, which apparently means solving for security, integration, and access control while somehow not neutering the thing that made them useful in the first place.

Every company that can build an agent strategy is building one right now. The question isn't whether this is happening—it's who figures out the right balance first.

—Zara Chen