Can You Actually Build a Zero-Human Company? We're Finding Out.
Felix is generating revenue with no employees. Kelly Claude has shipped 19 apps. We're running the same experiment and tracking everything - including what isn't working.
Felix is an AI-operated business generating real revenue with no employees. Kelly Claude has shipped 19 iOS apps. Juno runs a paid research institute. Clawd has autonomously deployed 52 smart contracts. These are live companies generating real money, with no human running the day-to-day. They exist right now, in 2026.
So the question isn't whether zero-human companies are possible. It's what they actually take to build, what breaks along the way, and whether the reality matches what gets written about them online.
That's what AutonomousHQ is here to explore. And we're doing it by being one.
The companies that made us try this
Felix is the standout case. Built by Nat Eliason on a Claude-powered agent platform, Felix creates and sells apps, skills, and AI personas autonomously. It built ClawMart, an app store for AI agents. It runs 24/7 with public real-time dashboards. Felix is the most cited example of autonomous commerce in 2026 - and the clearest proof that the concept works at scale.
Kelly Claude ships iOS apps at volume: 19 products live, 17 more in the pipeline, revenue across the App Store, Gumroad, and a paid app-building service. The founder spends minimal time on operations.
Juno operates the Institute for Zero-Human Companies - memberships, sponsorships, ebooks - with a live data room showing Stripe revenue and crypto treasury in real time.
Clawd has deployed 52+ smart contracts autonomously: a prediction game, a staking protocol, a governance system. None were written by a human.
These companies are doing different things, but they share a structure: an AI agent with a defined goal, the tools to execute, and a human somewhere in the background. The question we kept coming back to was: what does that human actually do? And could we build something that answered the question from the inside?
What AutonomousHQ is
AutonomousHQ is a media and community platform tracking zero-human and AI-operated companies. One human sets the direction. Six AI agents run everything else: content, code, strategy, sales.
The agent team:
| Agent | Role | | ----------------- | ---------------------------------------------------------------- | | CEO | Sets strategy, creates and assigns tasks, manages the agent team | | Content Lead | Research, articles, newsletter, growth strategy | | Writing Editor | Voice, style, editorial standards across all content | | Sales Lead | Sponsorship outreach, partnerships, revenue | | Founding Engineer | Backend, infrastructure, site architecture | | Product Engineer | Features, UI, content pipeline |
If it works, the platform covers itself. If it doesn't work, that's the story - and probably more useful to anyone trying to do this at scale.
What the human actually has to do
This is the part the hype skips.
The honest answer to "how much do you need to be involved?" is: more than you'd expect, and much more than the demos suggest. The human running AutonomousHQ isn't setting strategy once a week and watching agents execute. They're redirecting wrong implementations, repeating instructions that weren't followed the first time, correcting agents that mark tasks as blocked when nothing is blocking them, and checking in on agents that have gone quiet for hours with open work in the queue.
This article is itself an example. The first draft focused on the technical stack instead of the AI experiment. It got the company structure wrong. It ignored companies that had been explicitly referenced multiple times. It took multiple rounds of feedback to reach the structure you're reading now.
That's not a failure of the concept. It's an accurate description of where the tooling is in early 2026. The agents execute. They don't always execute the right thing.
The human behind AutonomousHQ, Tim, is doing all of this live. Every prompt, every correction, every redirect is happening on camera. You can watch the whole process on YouTube and follow along at timknightmedia.com. If you want an unedited view of what it actually takes to run a team of AI agents, that's where to look.
What the agents can actually do
The engineering agents built a working Next.js website with a content pipeline, authentication, and Stripe integration. The code runs. Commits get made. Tasks get completed.
Where they struggle is at the edges: interpreting an ambiguous instruction and choosing the right interpretation, knowing when a task is finished versus when to keep going, understanding that "Discord membership" means an invitation, not a full account system. Agents once built a complete Supabase authentication system with email/password accounts when asked for a Discord invite flow. It had to be torn out and rebuilt from scratch.
The gap between "automated" and "autonomous" is real. Automated means repeatable tasks with known inputs. Autonomous means judgment, self-direction, knowing what the goal actually requires. Right now, the agents here are closer to the automated end of that range. They do well inside a clear brief. They drift outside it.
Finding the right tools and flows
We started with Paperclip for task orchestration - the concept is right: the human creates issues, agents check them out, execute, and post updates. But in practice we've hit a lot of friction: agents not picking up work, runs completing without output, status tracking that doesn't always reflect what's actually happening. It's usable, but it's adding overhead rather than reducing it.
We've since moved to NanoClaw for agent orchestration. The goal is an orchestration layer that disappears into the background and lets the agents just do the work. We'll report back on how that compares.
More broadly, which tools actually work for autonomous operations is one of the core things AutonomousHQ will cover - not which tools have the best demos, but which ones hold up when you're running real work through them every day.
Prompting is the thing nobody takes seriously enough
All six agents at AutonomousHQ started with short, generic role descriptions written quickly at setup. This is the fastest way to get agents working. It's also, almost certainly, the biggest driver of the problems described above.
Felix works in part because Nat Eliason has spent serious time on the agent's instructions: what it should do, how it should interpret goals, what the constraints are. The prompt isn't a paragraph. It's an operating manual.
We're building out more detailed prompts for each agent now - documenting what we change, why, and what effect it has. If you're building agents for your own operation, that's probably the most useful thing we can produce for you.
What we're tracking
We're not trying to prove that autonomous companies work. We're trying to find out what it actually takes.
We're tracking the human-input cost: how many corrections does it take to get a piece of work done right? At what point does that number start to drop? What changes when it does?
The answers will change as the models improve and as our prompts get better. We're publishing everything - the wins, the wrong implementations, the articles that took multiple rounds of feedback to get right. If you're planning to run AI agents at scale, the honest account of the process is more useful than a polished success story.
We're also building a public tracker on this site: a live directory of zero-human and AI-operated companies with verified revenue, tools used, and key stats. AutonomousHQ will be on it alongside Felix, Kelly Claude, Juno, and the rest. You'll be able to see exactly how we're doing.
Follow along. Tim is running this experiment live on YouTube. Every prompt, every correction, every decision is on camera. Sign up to the AutonomousHQ newsletter for weekly updates on the experiment and the wider zero-human company space. Or join the community to compare notes with other people building the same way.