Building an autonomous business and closing in on $1M ARR

Amos Bar-Joseph has two exits under his belt, but he says he didn't do them right. This time, he's determined to build differently.

He and his two founders set out to build an autonomous company that brings in $10M per employee. And after roughly a year, Swan AI is nearly at $1M ARR with three cofounders.

Here's Amos on how he's doing it. 👇

The problem with venture capital

I built and scaled two startups the traditional way — raised venture capital, grew fast, landed two acquisitions. From the outside, it looked like success. From the inside, it felt sick.

Here's what "the playbook" actually looked like:

• Raise $2M-5M before you've found product-market fit.

• Scale to 40+ employees before you hit your first million in revenue.

• Burn through runway while rolling the bluff from one funding round to the next, hoping you figure out the business model before the music stops.

I did this twice. Both times, it worked — we got acquired. But both times, I knew we'd built on a shaky foundation, grinding through years of manufactured growth trajectories to satisfy investor timelines, rather than building something genuinely sustainable.

Now, I'm working on Swan AI. It's my attempt at completely reinventing how you scale a company from zero to hyper-growth.

It's Lovable for GTM. It's an AI GTM Engineer that helps sales, marketing, and founders turn any GTM process into an agentic workflow in seconds, from prompt to pipeline.

In 2025, we grew from 0-200 customers closing in on a 7-figure ARR with just 3 FTEs (the co-founders).

Targeting $10M ARR per employee

For a long time, staying lean meant staying small. That was the tradeoff. But AI agents finally make it possible to build an "autonomous business."

I started Swan because I'd seen enough of the old way to know it was dying, and I wanted to prove the new way could win.

We're not trying to be a lean, $5M ARR lifestyle business. We're targeting $10M ARR per employee with an ultra-small team, proving you can hit the same scale as traditional unicorns but without the bloat, bureaucracy, and burnout.

That's not a vanity metric, by the way — it's the constraint that shapes every decision we make. If a problem can't be solved with AI or systems, we question whether we should solve it at all.

Instead of the old playbook we're building a new one: Stay lean, use AI to amplify each person, and scale through intelligence rather than headcount. We're documenting everything publicly because if this works, it changes the entire game.

Building the wrong thing

We built the wrong thing first. Like everyone else in the space, we started with an AI SDR - automated prospecting, personalized emails, the usual playbook.

The workflow was simple: find prospects, filter with AI, personalize outreach, send sequences. It was clean on a whiteboard and easy to build.

Then, a customer asked us something that broke everything: "Can you make it work for webinar attendees instead of cold leads?"

Our beautiful workflow? Useless. We'd built automation pretending to be intelligence. Every new use case meant rebuilding from scratch.

That's when we realized we weren't building what we actually needed. We were running Swan's entire GTM ourselves — three founders, no SDRs, no marketing team. We were the use case. And we needed something that could adapt as fast as our strategy changed.

So, we scrapped the rigid workflows and rebuilt around a different idea: AI that could reason about any GTM motion, not just execute pre-defined steps.

The rebuild took weeks, not months. When you're three people with no buffer, you don't have the luxury of long cycles. We shipped fast, used it ourselves, and let our own pain guide every iteration.

Use constraints to force innovation

Constraints force innovation that comfort never would. And we had a big constraint: No hiring, period.

A great example is how our support ops evolved over time.

When support tickets hit 200/week, most startups would hire two support reps. We built a self-learning AI agent instead — and learned more about human-AI collaboration in four weeks than we would have in a year of "normal" scaling.

Here's how it evolved:

• Week 1: We gave our AI agent 20 documented answers and let it loose in Slack. It handled maybe 15% of tickets. Everything else still came to us.

• Week 2: Instead of trying to document every possible question, we built an escalation loop. When the AI didn't know something, it stayed in the customer's thread but pinged us internally. We'd provide the answer, the AI would deliver it. The customer never knew we were behind the scenes.

• Week 4: We noticed we were answering the same new questions repeatedly. So we added one feature: after we answered an escalated question, the AI automatically documented it for future use.

That simple loop changed everything. Within two weeks, we went from 15% to 70% autonomous resolution. The knowledge base grew from 20 to 180 solutions — not because we wrote documentation, but because the system learned from every human interaction.

The constraint of not hiring forced us to build something smarter than we would have if we'd just added headcount.

And it taught us something important: Don't try to make AI perfect before you ship it. Make it good enough to start, then build the feedback loops that let humans teach it as you go.

Breaking the model

The autonomous business model almost broke at 50 customers.

We'd built Swan to prove three founders could scale without hiring. And for the first six weeks, it worked beautifully — our AI agents handled support, onboarding, and pipeline generation while we focused on closing deals.

Then we hit the wall: 267 demo calls per week. Two founders were completely maxed out. 20% close rate meant we needed every one of those calls, but we physically couldn't do more.

The math was brutal. We'd proven that AI could handle 70% of customer interactions autonomously. But sales conversations — the highest-leverage moments — were still 100% human. And humans don't scale.

We had two options: Hire salespeople (breaking our model) or fundamentally change how we sold.

We chose the harder path. In seven days, we killed our entire sales-led motion and rebuilt around product-led growth. Self-serve for qualified leads. No demo required to get started.

It was the riskiest bet we'd made — our whole thesis depended on it working.

If I did it again, I'd build for self-serve from day one. We thought we needed high-touch sales to learn, but we actually learned more from watching people use the product without us hovering. The demo bottleneck forced a better model. I just wish we'd gotten there faster.

An AI-native stack

Our stack is built around one principle: AI-native from the ground up, not AI bolted onto legacy tools.

Engineering:

• Cursor - AI copilot that lets our CTO ship what used to require a 15-person team

• V0 by Vercel - Turns product concepts into working prototypes in hours

• Baz - AI code review that catches what humans miss

GTM & Sales:

• Swan AI - Our own product, turning anonymous traffic into a qualified pipeline

• Shortwave - AI email agent managing hundreds of conversations with perfect follow-up timing

• Unipile - LinkedIn automation API

Data & Intelligence:

• Explorium - B2B data foundation purpose-built for AI agents

• Sonar by Perplexity - Agentic web research that never stops

• Attio - CRM that's programmable, not just a database

Agent Building:

• n8n - Low-code platform for internal automations

• Retool Agents - AI agents on top of internal apps without engineering overhead

• Base44 - Builds interactive tools and websites through AI

The pattern: Every tool is either AI-native or built for programmatic automation. Nothing that requires a human to babysit it.

Growth via organic content

We grew Swan almost entirely through organic LinkedIn posts. No paid ads. No marketing hire. No PR agency.

We got 6M+ impressions in our first year and hit 200+ customers. All through, organic content being posted by the founders consistently.

LinkedIn became our entire top-of-funnel. Every deal we closed in the first six months started with someone seeing a post, visiting the site, or DMing me directly. The content didn't just build awareness — it built trust before the first conversation. And trust means higher conversion, lower churn, and customers who actually fit.

An AI-assisted content engine

Here's the system behind it:

I built what I call an AI content engine using Claude Projects. It's not about using AI to write faster. It's about using AI to think better.

The setup: A Claude Project loaded with my content pillars (the thematic angles I use to interpret any topic), brand guidelines, and examples of my best-performing posts. The system prompt tells Claude how to collaborate with me — run a creative process before responding, always give three options, wait for my input between phases.

The workflow: I come with a raw idea or insight. Claude helps me find the right angle, suggests hooks, and pressure-tests the structure. We iterate through strategy → narrative arc → execution. I never publish something Claude wrote alone. I publish something we developed together that sounds like me.

The result: A flywheel of content → trust → inbound → self-serve → happy customers → case studies → more content. Slow to spin up, but it compounds. This model is the moat.

The philosophy that makes it work: AI karma is real. If you're using AI to manufacture content without having anything original to say, it shows. The audience feels it. But if you're using AI to amplify genuine insights you've earned — through building, failing, learning — the content compounds.

I call this agent "Shakespeare." It runs my entire content engine now. But every post starts with something I actually believe. That's non-negotiable.

A skill vs a product category

The best advice I can give is this: Stop thinking about AI as a product category and start thinking about it as a skill you develop.

Most founders are waiting. Waiting for the right AI tool, the perfect use case, or someone to tell them where to start. Meanwhile, a small group is building. They're shipping ugly automations, breaking things, and learning what actually works. They're developing AI muscle.

Here's what that means practically: Get fluent with systems thinking. Learn to break processes into steps. Understand how tools connect through APIs and webhooks. Mess around with Make, n8n, or Retool. You don't need to understand LLMs — you need to understand workflows.

Then, pick something small and automate it. Lead enrichment. Meeting prep. Support triage. Doesn't matter what. Build a version that's 60% as good as doing it manually. Watch where it breaks. Fix it. Repeat.

The insight that changed everything for us was that AI agents don't need perfect instructions. They need feedback loops. Build systems where the AI can learn from corrections, and it gets smarter every week.

The founders who build this muscle now will have a compounding advantage. The ones who wait for AI to "mature" will spend the next five years buying tools from the people who didn't wait.

Start before you're ready. That's always the advice, but it's never been truer than with AI.

What's next?

Our goal isn't just to build a successful company. It's to prove a new model works — so others can follow.

The autonomous business thesis is simple: Humans with AI leverage can achieve outcomes that used to require massive headcount. But right now, it's still a thesis. There are early signals — Cursor, Bolt, Lovable hitting incredible revenue-per-employee numbers — but no one's documented the full playbook from zero to scale.

That's what we're trying to do with Swan.

This isn't about staying small forever. It's about proving that when you optimize for human-AI collaboration, every person on the team becomes a force multiplier. Maybe we stay at three founders. Maybe we eventually grow to ten people doing the work of a hundred. The point is the leverage, not the headcount.

No massive funding rounds before you've found product-market fit. No 50-person team to brute-force growth. No burning years of your life managing complexity instead of creating value.

If we pull this off, the playbook becomes real. Every framework we build, every system we create, every mistake we make — we're sharing it publicly so founders who come after us don't have to figure it out alone.

• Follow the journey: swan-ai.beehiiv.com ‚ my newsletter documenting how we're building an autonomous business. The wins, the failures, the frameworks. No polish, just what's actually happening.

• Try the product: getswan.com — see what an AI GTM Engineer actually looks like.

• LinkedIn: My profile is the home base. I post almost daily about human-AI collaboration, GTM, and the autonomous business playbook. Most of what I've shared here started as a LinkedIn post.

• Talk to my digital clone: Autonamos ‚— a GPT trained on everything I've written, built, and learned. Want to explore the frameworks or pressure-test an idea? Start a conversation with it.