The Math Behind $100M ARR – It's More Achievable Than You Think

01 /The Concept

Feltsense is building AI agents capable of operating as fully autonomous startup founders.

"We believe our future will be shaped by an army of AI founders that can understand human needs and build what satisfies them. These AI founders will identify our pain points, create digital or even physical infrastructure to address them, and make that infrastructure accessible to people."

Beyond this vision statement, the website offers little additional detail. Despite that, Feltsense raised $5.1M in early February from a group of venture funds and angels – including the founders of the startup intelligence database Crunchbase and the creator of Moltbook, a recently launched social network for AI agents.

02 /Why It Matters

What follows is a summary of the key ideas from the founder's most substantive public statement – found on Twitter, paraphrased here in third-person.

Feltsense was founded in January 2025. Within a few months, the team had launched 10,000 AI founders at a cost of virtually pennies per startup. Launching was the easy part. Nearly a full year went into figuring out how these AI founders could actually find product-market fit for their products.

In December 2025, something clicked. Weekly signups for products built by Feltsense's AI founders jumped to a number that had been the monthly target – and the team knew they'd finally found something that worked.

The future, as Feltsense sees it, will be defined by armies of AI founders competing for market share – advancing to take ground from incumbent players, and defending what they've claimed.

This framing matters. Most AI optimists have focused on AI's ability to build *better* products. But great products don't sell themselves. The real job of an AI founder isn't to "create better products" – it's to compete for market share. That means finding product-market fit first, then investing in marketing to expand the position that's been won.

Why might AI founders be better at this than humans? Because AI can identify opportunities faster, test ideas faster and cheaper, and learn from every step without the ego friction that causes many human founders to double down on wrong assumptions instead of pivoting.

Feltsense's AI founders handle the full startup lifecycle: sourcing ideas from social media signals, developing strategies, building products, and taking them to market. They can even hire humans to cover tasks where people outperform AI.

The idea of AI hiring humans is already gaining traction. Moltbook and Rent a Human have reached similar conclusions. You could call such systems "delegators" – entities that route work to others based on who can do it best.

The thing is, delegators already exist. Uber is a prominent example: its algorithms hire human drivers to complete trips with minimal wait times and optimal pricing. That's a software system directing human labor.

The concept extends to other domains. What's particularly interesting is that in pilot programs built on this principle, the people doing the work often don't realize they're working for an AI.

They receive instructions via messages or app prompts – and in most cases, there's no way to tell whether they came from a person or an AI system. Twelve months ago, that would have seemed impossible.

Blind evaluations go further: workers rate their experience with AI employers *higher* than with human ones.

This leads to a provocative conclusion: humans are no longer the "accelerators" in a human-AI system. They're the bottleneck.

Because an AI agent listens more attentively, explains more clearly, teaches more effectively, and nudges more skillfully than most human managers – which means it can help a person achieve their goals faster than a human supervisor could.

The implication: you don't need to wait for AI that thinks like humans. The right "AI + human" pairing – where the AI is the delegator – already outperforms either working alone.

03 /Opportunities

A lot has changed in the past 12 months.

At the start of last year, Rocketable ([related review](/review/v-obshhem-sluchaje-jeto-poka-fantastika-a-v-chastnom-vozmozhnost-na-milliard)) raised $6.5M through Y Combinator for a similar idea – but reflecting the understanding and technical capabilities that existed at the time.

Rocketable's stated goal was to build a software holding company with many products, operated by a single human founder. AI would handle all ongoing development: processing user feedback, fixing bugs, shipping new features.

But at the time, Rocketable's founder was clear about one thing: AI could only improve existing products – not invent new ones. That meant the business model required acquiring already-operating products from human developers: minimum two years of operation, stable or growing revenue between $500K and $2M per year, at or near profitability.

Translation: product-market fit could only come from humans; AI was there to maintain and extend it. And even that felt radical.

Feltsense takes one more step: it claims AI can now generate ideas, test them, and find product-market fit independently. Only then does it bring in human collaborators – to help develop and improve the product alongside the AI founder that conceived it.

The emerging pattern: AI can already autonomously create and grow products, fully independently or by recruiting humans as needed – depending on the specific task.

That generalization, like all generalizations, probably isn't universally true. The more useful question is: for which specific cases do AI-built products create more value than human-built ones?

Inception Point AI ([related review](/review/edinstvennaja-uspeshnaja-kontent-strategija)), which raised $5.05M in late January, points in an interesting direction here.

The startup claims to have built the world's largest independent podcast publisher, running more than 4,000 shows and releasing 3,000 new episodes per week. The entire production pipeline is AI-driven.

Feltsense has pushed the cost per episode below $1. At current podcast advertising rates, each episode breaks even if just 20 people listen to it.

This enables something no human podcaster could sustain: ultra-niche shows built for audiences of a few hundred people. Human podcasters avoid tiny audiences because the economics don't work and the effort isn't justified. They inevitably widen their scope to survive – destroying the very specificity that made the content valuable.

"Making content for everyone means making it for no one" But making it for a hundred people won't pay the bills either. So humans rarely pursue this. Or their initial enthusiasm fades fast.

Now flip the math: launch 1,000 AI founders, each building a product for 1,000 people. That's 1 million people in aggregate. At $10/month per subscriber, that's $100M per year.

The unusually promising direction right now: building entire portfolios of similar AI products, each targeting a narrow audience segment – where the economics only work at scale, and AI is the only way to achieve that scale.

"AI product" here means a product that's almost entirely run and evolved by AI at a cost that stays near pennies per unit. What portfolios of AI products could you design and launch?