I spent a decade learning to build companies the hard way. AI agents just made most of what I learned obsolete - and I think that's the best

I'll start with the uncomfortable part.

For about ten years I built companies manually, in markets that, let's say, don't forgive mistakes. I scaled a cannabis retail brand from one store to fifty across Thailand. I built a holding structure spanning five companies in wellness and psychedelics. I raised $2.5M for client startups through crowdfunding. I operated across Ukraine, Poland, Thailand, and Hong Kong. All of it by hand - with people, payroll, coordination, and the thousand small operational fires that come with running real businesses on the ground.

Here's what a decade of that teaches you: the bottleneck is almost never the vision. It's the execution layer. The coordination. The ops. The unglamorous work that has to happen for an idea to become a real thing that makes money.

Then AI agents got good, and I had a moment I haven't really shaken since: this changes the economics of everything I just spent ten years learning to do.

Not because it makes me smarter. Because it changes what execution costs. The equation that just broke.

When you build manually, execution is expensive. You need a developer to ship. A marketer to grow. A salesperson to close. An ops person to keep the lights on. The cost of building a company is basically the cost of assembling that team.

Agents break that equation. For the first time you can have execution capacity without headcount. That's not a productivity tweak, it's a different cost structure for building anything.

And I don't think this stays a soft advantage for long. The companies that internalize it early get a gap that compounds. In 12–18 months, some teams will be operating at a scale their headcount shouldn't be able to support, and others will be running fifty people against competitors running eight people and a network of agents. I've competed in markets where a structural cost advantage is the whole game. This is that, for software companies.

What I'm actually building (and why it looks like three things).

I'm going to be honest about something that confuses people: I'm working on three products, and from the outside that looks unfocused. It's one insight at three scales.

1/ WorkNet is for founders at zero. No team, no budget, no infra. You describe what you're building, and a set of AI agents start working: shipping, marketing, reaching out. We've had founders launch a product, run campaigns, and close their first customers without a single hire. That's a kind of company that couldn't exist a few years ago.

2/ MaxWorker is for teams that already exist but are drowning in the operational layer - the reports nobody wrote, the follow-ups that slipped, the data scattered across six tools that never comes together. It's an AI coworker that lives in Slack and does the work about work.

3/ Workhold AI is the infrastructure underneath both, the operating systems and deployment frameworks for running a business on agents at scale.

One bet, three segments: the companies winning in five years are the ones that learned to run on agents instead of headcount.

Why I don't think this is hype (and I'm allergic to hype)

I've watched the operating numbers, not the funding announcements. Asana's research puts 58% of knowledge workers' time into "work about work." McKinsey pegs email alone at ~28% of the week. The average enterprise runs on hundreds of separate apps and workers switch between them over a thousand times a day.

None of those are AI numbers. They're costs that have been sitting on every company's P&L for years, quietly, with nobody treating them as a solvable problem. What changed isn't the problem. It's that we can finally build something that touches it.

On the "won't Microsoft/Google/Salesforce just do this" question - they're embedding AI as features inside existing products. Putting Copilot in Word makes Word better. It doesn't give you a participant that crosses the boundaries between every tool you use. The CRM still holds the data, the tracker still holds the tasks, Slack still holds the conversation — and the work of pulling all of that together is exactly the gap no single embedded feature can reach, because no single feature has the whole stack.

We run on our own product, including when it breaks

This is the part I care most about saying clearly, because I've seen too many "AI runs your company" pitches that are vapor.

Our own operations run on the same systems we sell. We use agents for marketing, analysis, and ops, and our team is small relative to what we're shipping precisely because we've swapped headcount for infrastructure. That's not a pitch-deck line, it's the only reason we can move this fast.

It also means we see what doesn't work before customers do. When you eat your own product daily, you develop a very unsentimental sense of where it's still bad. I'd rather build in public from that honest place than pretend the agents never get anything wrong. They do. The difference is whether you design for that or hide it.

Why I won't wait, and don't think you should either

Every month a company runs the old way, the ones running the new way pull further ahead, not in a vague "AI is the future" way, but in concrete cost-structure and execution-speed terms.

I think founders who move in the next 12–18 months will look back on this the way early SaaS founders look back on 2008 as the moment the ground shifted and a few people moved first. I don't want to be the person who explains, later, why they waited.

If you're building on agents already, I'd genuinely like to compare notes: what's working, what's quietly broken, what you've had to put a human back in front of. That's the conversation I'm here for.