OpenClaw Just Joined OpenAI

My family was out of town this long weekend. (President’s day in the U.S.)

Which means I did what any reasonable person would do with 72 hours of uninterrupted time: a marathon coding weekend (with AI, of course).

Not necessarily because I have to or I have an urgent deadline. But because right now building with AI is more fun than building’s ever been. And trust me, I’ve been building software products for decades.

This weekend turned into why these are such fine times. I coded. I invested in a startup in 7 hours. And I watched one of the most interesting open-source AI projects get acquired by OpenAI in real-time.

I thought it might be useful to do a behind-the-scenes of what it's actually like working in AI right now:

• My AI coding marathon buildout

• How a surprise investment happened in record time

• Why OpenClaw joining OpenAI is a bigger deal than it looks

—@dharmesh

Quick note: This is the first time I'm running a sponsored section. I only want to share resources that are genuinely useful, so I'm starting with this free playbook from the HubSpot team on AI agents. Note: I don’t ever plan to accept third-party sponsors, but HubSpot’s kinda my baby. Thanks for indulging me. Let me know in the poll at the end if this kind of thing is useful (or not).

Your 2026 Guide to AI Agents

Inside: what makes agents work, proven use cases across marketing/sales/ops, and how to build effective human-AI partnerships.

Access the free playbook here.

Day 1: HubSpot Dev Platform + OpenClaw

The first day of my marathon coding weekend started with digging into the HubSpot developer platform. Not the APIs (which I use all the time), but the UI extensions part of the platform. HubSpot has long been a powerful platform to build on and extend.

UI extensions let you build custom experiences right inside HubSpot's interface in a way that makes them look “native”. I've had some ideas bouncing around for a while on how we might take some of the developments in vibe coding (what I call agentic coding). This weekend finally gave me the chance to tinker on those ideas.

My setup: Claude Code for the heavy lifting, Codex for when I run into issues with Claude Code (which happens) both running inside Warp as my terminal (though I don’t use the Warp’s “agentic development environment features).

If you're not sure where to start with AI coding tools, I broke down the entire AI coding tool landscape here: The 2025 AI Coding Tool Landscape. It probably needs an update soon, but it’s still a good place to start if you’re a beginner.

And yes, I'm also tinkering with OpenClaw (which used to be called ClawdBot and then MoltBot — but this is the last time I’m going to refer to the old names — we can just call it OpenClaw).

Running experiments. Testing capabilities. Seeing what breaks. Watching it solve problems in ways I wouldn't have thought to approach them.

Don't worry, I'm running it in a VPS… and it does not have access to any of my primary accounts. Still don't trust it enough for that.

But it's genuinely fun to work with something this capable. The kind of fun that makes you lose track of time.

Day 2: I Made a Surprise Investment in 7 Hours

So, as a planned extension of Day 1, I was coding away on a new agent (which I might launch in early alpha today — maybe even by the end of this newsletter). 🙂 

What wasn't planned: signing docs on an indie investment.

Time elapsed between knowing they were open to investment, getting details, figuring out terms, and signing the docs: 7 hours.

Here's how it played out:

1. I've been using a new startup's product as part of this new project I’m working on and ran into some issues — so reported them.

2. I had not expected to hear back until Tuesday (it's a long weekend for those in the US). But I heard back within minutes, from the founder.

3. The issue got fixed and pushed to production within an hour.

4. I sent more issues. Heard back again -- in minutes.

Went back and forth like this for a few hours — on a Saturday afternoon — during a long-weekend.

That's when I asked if they were interested in an indie investment from me.

They were. We chatted about it over email, they sent the docs, and I signed them Saturday night. Total time: 7 hours.

Like most of my investments, I've never met the founders, talked to them on the phone, or had a Zoom meeting.

Basically, this startup is building tools that make it easier to get to hard-to-get data. The product is early, but the responsiveness and speed of iteration tell me everything I need to know about the team.

Can't share the name yet (they're in stealth), but stay tuned.

Moral of the story: You don't have to be working on the weekend. You don't have to respond immediately to emails. But it definitely doesn't hurt. For both sides: founders and investors that want to get into the best startups.

OpenClaw Creator Joins OpenAI

A couple of weeks ago, I wrote about OpenClaw -- the open-source project for creating a personal AI agent that took over everyone's timeline.

I said the capabilities were impressive. I also said most people should wait to use it because of security risks and complexity.

And I made a prediction:

"My guess is that the frontier model companies (OpenAI, Anthropic, Google) are going to offer something along these lines soon -- with better security models, easier setup, and recognized brands."

But it happened way faster than I expected!

This weekend (on Sunday!), Sam Altman announced that Peter Steinberger, the creator of OpenClaw, would be joining OpenAI and that the OpenClaw project would become part of a foundation with support from OpenAI.

Though I'm a bit sad that Peter's not going to be the benevolent dictator of OpenClaw any longer, I think this is a win for Peter, a win for OpenAI, and a win for the project.

This is important because when I wrote about OpenClaw, one of my concerns was that it was too technical and too risky for most people. You had to run it on a separate physical machine (hence the uptick in Mac Mini sales) or a VPS (Virtual Private Server), manage security yourself, understand terminal commands, and trust an open-source project with access to your accounts.

It’s easier to explain to your IT people that you're using something from OpenAI than "OpenClaw." For reference, when I told my wife about OpenClaw, she thought the whole lobster themed naming was cute, but that OpenClaw was a bit creepy — made her think about someone having a large open claw and grabbing all your data.

Now, the OpenClaw project is getting some much needed backing. Word is it that Peter (the creator of OpenClaw) was losing $15k-$20k a month operating OpenClaw (and he had figured out a monetization model yet).

As a result of this move, OpenAI gets the talent (Peter’s awesome — I don’t know him but watched the 3 hour episode he did with Lex Friedman), and a head start on personal AI agents. Peter gets resources and distribution. The open-source community gets a foundation with real backing.

And users? They'll ultimately get a version of this with better security models, easier setup, and the OpenAI brand behind it.

Congrats to all, but especially to OpenAI. I honestly did not see the acquisition coming this fast.

I'll bet Anthropic didn't either.

As Sam wrote, "...the future is going to be extremely multi-agent."

I agree. And now one of the most interesting multi-agent projects is in the hands of one of the most important AI companies in the world.

Why This Matters (And Why There's Never Been a Better Time to Build)

In sum it up, it was quite the weekend -- both on the coding I’m doing, the investment I made and the OpenClaw news.

I've been at this for 30+ years now. We went from assembly to C to C++ to higher-level languages. Each step lets us build more with less. Work at a higher level of abstraction.

But over the last couple of years, AI has taken that up a massive notch.

As builders, we can now take the ideas manifested in our heads and -- with a relatively high degree of fidelity -- translate that into a working product. We can prototype quickly. We can iterate more quickly. We can produce and even launch products more quickly than we did before.

Even if the models stopped improving tomorrow (which they won't), we have enough technology in our hands as builders to last us another decade just to absorb the impact what we hold in our hands today.

That’s why I'm still building.

Not because I have to. Not because we can't hire people to do it. But because building things is more fun and more accessible than it's ever been.

Projects like OpenClaw -- built by one person, used by thousands, now joining one of the most important AI companies in the world really show you what's possible.

It’s time to build! :^)

—Dharmesh (@dharmesh)

P.S. Let me know in the poll below: Should I include occasional sponsored content in future newsletters? Only if it's genuinely useful, of course.