Is the $1B solopreneur "dream" now possible? | TWS #044

plus xAI merges with SpaceX, Opus 4.6, Intel builds a GPU, and much more...

This is a FREE EDITION of the newsletter. Every week, get the latest curated signals on news, uncommon insights and perspectives that help you become a better builder and thinker. Paid subscribers get full access to all newsletters, deep-dives, essays, field guides, and curated podcast Q&A notes. Please fill out our anonymous survey so we can deliver the best content to you.

---

It’s been an interesting couple of weeks to say the least when it comes to AI stuff.

I don’t usually drag on about particular news as I like to keep things fresh, but for this time only, I wanted to double-click on everything that’s been happening with AI agents because it was a telling sign when a friend reached out to me just a few days ago (who is by no means tech savvy) and wanted to see if I could help install Clawdbot for them on a VPS (Virtual Private Server) 😳. I mean, if this isn’t a signal that something like this has gone totally mainstream, I don’t know what is…

In any case, I said sure.

He’s now got an assistant called “Felix” helping with his day-to-day tasks. WILD!

But it also got me thinking about what this means for people wanting to start their own business or startup. We’re now in an age where building a product and potentially a company is becoming extremely granular.

What I mean by that is that we’re rapidly moving from a world where “scaling” meant adding headcount, to a world where scaling just means adding compute.

And if my friend is now thinking about spinning up autonomous agents on a VPS, then the barrier to entry has just entirely collapsed.

It’s probably a strong signal that we finally need to have a serious conversation about the $1B business (unicorn) run by a solopreneur, that for the last decade, has been nothing more than clickbait.

For years, Silicon Valley had this somewhat idealistic view of the “one-person unicorn” — basically a startup built by a single founder that would reach unicorn status without hiring a single full-time employee.

Naval Ravikant was one of the earliest and most vocal proponents, predicting a future where “code and media are permissionless leverage”. This means that it allows an individual to scale towards infinite value.

It was kind of a beautiful and romantic idea. It was also, for the better part of a decade, pure vaporware.

I guess most of us just nodded along, but deep down, we knew the reality. You inevitably hit the “human ceiling”. You could build it alone, sure (thanks to AWS, GCP, Vercel, etc.). But you couldn’t sell it alone. You couldn’t handle support alone. You couldn’t scale operations alone. The “10x Engineer” was real, but the “10x company” required…well, a company.

That was effectively true until about six months ago (it’s now February 2026).

For me, I get the sense that the infrastructure has shifted.

We’re now moving from the era of Software as a Service (SaaS) to Service as Software. And because of this, I really believe the unit economics of the firm could be fundamentally disrupted.

So it’s time ask, from first principles: is the one-person unicorn finally possible?

Continue reading in Part 2 below… 👇

—

But for now, let’s dive into this week’s newsletter…

---

---

Claude Opus 4.6

I don’t really mention model releases here a lot. But this week is the exception as Anthropic and their cash-cow Claude have been making waves both at the Super Bowl, across the markets (I’ll dig into this next week!) and on the news circuit.

In any case, Opus 4.6 is their latest release, and it’s the gold standard right now when it comes to coding and even general-purpose AI. Everything about Opus is just good—I use it on the regular, switching back/forth between Gemini 3 Pro.

Now they have “Adaptive Thinking”, where the model autonomously decides to pause and “think” deeper about complex problems, rather than just blurting out the first response.

Also, they’ve increased their context window to 1M tokens, so now you have a model that can finally hold an entire codebase or legal archive without hallucinating the details.

I’m going to do a deeper dive on Anthropic at some point because the way they build technology and their brand is quite interesting.

Until then, let the updates keep on coming…

Apple just made its second-biggest acquisition ever

Looks like Cupertino just dropped $2B on a company you’ve probably never heard of. Apple just acquired Q.ai, which is an Israeli startup specializing in audio AI and the detection of “facial micromovements” to interpret speech and emotion.

Apple doesn’t take acquisitions lightly. They prefer to do everything in-house, but that’s slowly changing around how they collab with folks like Google and now in the hardware space with Q.ai. This is their second-largest acquisition, trailing only Beats (Dr. Dre’s headphones brand), which they purchased for $3B in 2014.

All the tidbits around what Q does are intentionally vague, but it looks like the tech hints at some sort of massive leap in human-computer interaction. It’s the idea that devices now aren’t just listening to what you say, but they can read your lips and your emotional state to understand how you say it.

I guess I’m pretty skeptical whenever I hear “emotion detection”, but obviously Apple doesn’t spend $2B on vaporware. It feels like the missing sensory layer for the Vision Pro or whatever AR glasses come next.

If your device can read your micro-expressions (and others too!), the interface completely disappears. But that said, the privacy implications of a device that knows you’re lying before you do? That’s gonna be fun.

Intel will start making GPUs

The company is waking up to the reality that CPUs aren’t the center of the universe anymore. CEO Lip-Bu Tan announced at the Cisco AI Summit that Intel is officially pivoting to produce GPUs, directly challenging Nvidia’s monopoly.

The strategy is explicitly built around “customer needs”, which is basically code for “we will build whatever you want if you stop buying Nvidia chips”.

It’s a massive shift for a company that has historically tried to force the market to march to the beat of x86 chips (which still power your computers and laptops).

I actually want to root for the underdog here, but Intel has been “entering the GPU market” for what fees like years. I guess the difference this time seems to be the humility in their approach, which is focusing on specialized customer needs.

The good thing is that Intel controls its own foundries (the facilities that actually build the chips), so that’s a good supply chain advantage to have.

SpaceX acquires xAI in $250B Deal

This wasn’t a complete surprise, to be honest. This will value SpaceX at $1.25T, probably making it the largest IPO in history (when it happens). The logic behind this is that they haven’t been shy about doing orbital data centers in the near future. The rationale is that right now we’re running out of power and cooling capacity on Earth for AI, so the plan is to launch massive AI computer clusters into orbit, which will be powered by solar energy and cooled by the vacuum of space.

The projection is that space-based computing is projected to be the cheapest method within 2-3 years. It’s a vertical integration of energy, launch (Starship), and intelligence.

I think for many people, this sounds pretty insane. But fundamentally, there is a method to the madness because of the sheer thermal output of a GPU cluster. It’s fair to say that we’re boiling oceans to train models right now, so moving that heat waste to space where you have infinite solar power and a near-infinite heat sink is arguably the only physics-compliant way to scale AI indefinitely. You would basically train the AI in space, then send it back to Earth for deployment and inference.

At first, it’s going to be a logistical nightmare, but just like anything, once you have the foundation and infrastructure in place, it becomes a very scalable thing.

Google’s Project EAT?

Google is finally taking its own medicine to fix its velocity problem. EAT (which is like “eating your own dog food”) is an internal initiative to forcefully inject AI tools into Google’s own workforce. The irony is real: they’ve been selling AI to the world, but internal adoption has apparently been…sluggish.

The goal is to use their own models to “supercharge” employees, specifically targeting developer velocity and code quality within the infrastructure teams. Obviously, it makes sense and is probably a recognition by them to stay ahead since they need to be the power users of their own tech.

It’s actually kind of shocking that this wasn’t already happening. The fact that Google needs a formal, named initiative to get its engineers to use its own AI explains a lot about the disconnect we’ve seen recently between research breakthroughs and their product. In any case, Google engineers’ debugging Google tools usually leads to better tools for the rest of us.

Meta is doing the same thing, but being more brash about it.

---

---

The physical reverse Turing test

To prove that it wasn’t human, they literally had to cut off the fabric and peel back the syntethic skin to reveal it was a humanoid!

AI Agents to Boost Productivity and Size of Software Market

“Our technical deep dive illustrates the potential for agents to become the new user interface for knowledge workers,” Borges writes. By 2030, the agent portion of the software market may account for more than 60% of the total. In other words, the profit pool is going to shift to agents, but the entire market for software will be larger.

Poland’s path to remarkable prosperity

According to data from the International Monetary Fund, Poland’s GDP in 1990 was a mere US$6690 in current dollars. By 2024 it grew almost eight-fold to US$51,630 in terms of purchasing power parity. All that in just three decades, or one generation. And it goes on. According to the European Commission’s forecast, in 2024–25, the Polish economy will be the fastest growing large economy in the European Union.

An AI agent on Moltbook sparked up a storm with a controversial post

---

Thanks for reading so far. 

If you liked the content so far, you can check out our full TWS archive, plus more in-depth content: under the hood and field guides.

If you’re a founder, apply here (Metagrove Ventures) for funding.

I’d love to get some feedback. If you have some time, fill out the following survey.

Want to get in front of our audience? You can partner with us. 

Upgrade to the premium version for full access to all content.

Follow me on LinkedIn or X for more insights.

---

---

Note: If you’re reading this via email, click here to read the full post in case your email client shortens the full piece.

---

The One-Person Unicorn

The Strawman

Let’s start with why this is likely to fail. The elephant in the room.

Despite all the hype around Openclaw (fka Clawd/Molt-bot) spinning up entire backend architectures in minutes, the skeptic’s case is stronger than most admit.

The primary killer of the solo-founder dream is something called the “90% trap”.

You see, current agentic workflows are really, really good at getting you 90% of the way there. They’re able to draft the email, they generate the boilerplate, and they analyze all the data. But the final 10% which are all the nuances, the strategic “judgement calls”, and the high-EQ sales negotiation requires a human touch and some intervention.

In a traditional company, you would delegate that 10% to a VP or a Senior Manager. But in a one-person unicorn, the founder is the exception handler for every single process. They have to. There’s no one else.

So if your agents are 90% accurate and you run 100 of them, you have created fires that you, personally, must put out every morning. Instead of the CEO, you actually become the world’s most overworked debugger.

Not only that, but you also face the “Trust Paradox”, especially if you’re building and selling in the enterprise space. To scale to $1B, you also need enterprise contracts. And we all know that enterprise customers buy trust, not just code. They want a “throat to choke”.

And so the question becomes: can a single human legally and operationally be the counterparty of a Fortune 500 equivalent? Probably not (but it depends…)

The friction of procurement departments alone, who demand audits, SOC2 types (which agents can actually handle), and quarterly business reviews, is a massive barrier.

Finally, there is something called “Organizational Entropy” (read the piece from First Round for more info). But the general gist here is that agents don’t have a semblance of culture (yet…). They don’t have some level of “shared alignment” unless explicitly programmed.

With humans, they self-correct & align based on the overall vibe and mission; a swarm of agents effectively maximizes for their individual prompt constraints, which usually leads to a really chaotic and disjointed product experience.

The Steelman: Infinite Digital Interns

Now, let’s look at the reality of what just happened with Moltbot.

The argument of the one-person unicorn rests on an important premise: that the marginal cost of cognitive labor is trending towards zero.

An interesting graph from McKinsey shows how AI is being adopted across business functions:

Source: McKinsey (Quantum Black)

In the old world, if you wanted to double your output, you had to roughly double your headcount. You would add “cost” (like salaries, benefits, office space) to gain “concurrency” (people working together in parallel).

In the post-2026 world, concurrency is decoupled from headcount. You now add compute, not people.

A single founder using AI agents isn’t really “coding faster”, but they’re acting like the API router for a swarm of specialized, digital interns.

For example:

• If you need a marketing department? You don’t hire a CMO anymore. You spin up an agent instance dedicated to SEO scraping, content generation, and social distribution, running 24/7.

• If you need a sales team? You don’t hire SDRs. You now deploy agents to personally handle thousands of outbound sequences, and you treat every lead with the researched depth that a human SDR would take 30 minutes to do.

All of this allows for the emergence of “Service as a Software”.

The next $1B company might not be a Salesforce (which is a tool you use to sell). But rather, it will be an agent that does the selling for you.

So if you’re a founder who builds that agent, you don’t need a massive team. You’re selling the outcome (booked meetings), not the tool (CRM).

The Revenue Per Employee (RPE) metric goes from the traditional $200-250K to effectively infinite.

But hang on…didn’t I just say that in the Strawman section that the 90% trap is real?

Yes, absolutely. It’s real for a variety of products that are positioned as tools, not the outcome. This is because you’re building a tool like a CRM that requires a human to operate it effectively (providing that final 10% of skill).

However, there is a case to be made (still to be yet be fully proven out) that the market is now slowly demanding and pricing products based on outcomes.

The Steelman argues for building a product (an agent that does the selling). By selling the outcome (booked meetings) rather than the tool, the agent assumes 100% of the operational responsibility.

In terms of “judgment calls”, there’s also a case to be made that if the agent can simulate the deep research and personalization previously reserved for humans, it effectively automates that “final 10%” of quality that was previously the bottleneck.

I recently came across this idea: human-on-the-loop vs. human-in-the-loop.

To reach 100% automation of the workflow (so the founder isn’t the bottleneck), the model shifts the human’s role. So instead of being “in the loop” (fixing the last 10% of every task), the founder is “on the loop” (acting as the “System Architect”). Now, the founder would manage the machines that do the work, rather than doing the work that the machines failed to finish.

Ultimately, the “Service as a Software” approach model covers the final 10% by deploying agents that are sufficiently advanced (“Manager Agents”) to deliver complete outcomes (i.e., deep research, signed meetings) rather than just partial outputs (drafted email), which in turn removes the human from the execution loop entirely.

How might this work?

Well, firstly, I don’t think you can follow the traditional playbook anymore. The rules are now inverted.

Firstly, don’t build a feature; you have to build workflows instead. The one-person unicorn can’t win by having just a better “UI”. It needs to win by integrating deeper. Instead of just building out a dashboard, you have to build out a system that replaces the person who looks at the dashboard (or close to it).

The next one is something called aggressive asynchronicity. Basically, you must design your business to be async-first. Real-time support is impossible for one person. Therefore, your product must be technically flawless (self-healing code, handled by agents) or your service model must be fundamentally asynchronous.

Finally, the Founder becomes the “System Architect,” and not the “Doer.” Your job changes. You are no longer the “Chief Executive Officer.” You’re the “System Architect.” Your daily routine isn’t “writing code” or “taking calls”; it is reviewing the logs of your agent swarms, tweaking their system prompts, and optimizing the “Constitution” that governs them. You manage the machines that manage the work.

This sounds super futuristic to a lot of people, but there are people out there who are starting to build out the infrastructure to do exactly this.

—

Ultimately, the question I sought to ask was “Is the One-Person Unicorn possible?”

In 2025, I would have said no. In 2026, with Openclaw and Moltbot in the wild, the answer is: Yes…but do you really want to?

Even if you could build a company towards a $1B completely solo, where’s the fun in that? Building a business solo can be a very lonely journey. There’s no shared highs, no high-fives in the war room, it’s just you and the machine.

Personally? I think building with others is just so much more fulfilling.

The magic often happens in the collaboration, not just the output.

So yes, the tools are here, and the capability is real. But is it worth it once you reach the top of the mountain alone?

---

If you liked this, please consider sharing this with your network. Thanks for your support!