AI Agents Are Learning to Use Your Computer

It's happening faster than even I expected.

In the last few weeks, we've seen four huge announcements about computer-using type AI agents. Anthropic, OpenAI, and Google are all racing to build AI that can actually use computers like humans do.

In today's email, I'm breaking down:

• What happened, and what these announcements mean beyond the hype

• Where we are in the evolution of AI agents (hint: think GPT-3 era)

• What it means for developers and entrepreneurs

Hope you enjoy!

What’s happened in the last few weeks?

The way I like to think about the currently announced computer-use type systems is like the GPT-3 days of OpenAI's big ChatGPT explosion. We're seeing the first glimpses of something revolutionary, but we're not quite there yet.

Let's look at what's been announced (and the rumors) just in the last few weeks:

Anthropic's Computer Use API (Available now)

OpenAI's Operator (Rumored)

ChatGPT's macOS App Integration (Live in Beta)

Google's 'Project Jarvis' (Rumored)

Why is This a Big Deal?

Until now, AI has been like a really smart person trapped in a chat window.

They could give you great advice, but they couldn't actually do anything for you. They were like a brilliant consultant stuck behind bulletproof glass – lots of ideas, but no way to implement them.

That's about to change. Dramatically.

No longer will it be necessary for an API to exist for the specific functionality you need to access. Instead, agents will be able to get to the same features/functions that a human can get to.

• Before: Need to wait for companies to build specific APIs

• Future: If a human can do it on a computer, an AI potentially can too

Real-world example: Instead of waiting for your favorite design tool to release an API, agents will be able to just... use the software.

Need to automate a workflow across multiple tools that don't talk to each other? An agent could handle that too, just like a human would.

What it Means for Builders

The future isn't just about general-purpose AI agents that can do everything. While those will exist, there's also massive opportunity in specialized agents that are exceptional at specific tasks.

Think about it like this:

1. General-purpose agents will handle broad tasks

2. Specialized agents will excel at specific functions

Looking at agent.ai, where we're building the platform where you can discover, build, and hire specialized agents, some of our most popular specialized agents are:

• Flux Image Generator: 102.1K executions

• Domain Valuation Expert: 40.7K executions

• Company Research Agent: 25.2K executions

These numbers aren't just statistics – they're early proof that users want agents that are incredibly good at specific tasks. And this is before the integration of computer-use capabilities (we’ll be adding those features as soon as they’re useful).

For developers and entrepreneurs, this means:

• Rethink your AI product roadmap with computer use in mind

• Look for opportunities in specialized agent development

• Start preparing for the integration of these new capabilities

The next few months are going to be exciting (and probably a bit chaotic) as these technologies mature. Remember, we're at the GPT-3 stage of computer use capabilities – impressive, but just scratching the surface of what's coming.

If you had an AI agent that could use any software on your computer, what would you automate first? Let me know in the poll below—I'm genuinely curious.

-Dharmesh