How to launch your AI startup in 2025 | Alexander Gorny

Alexander Gorny: From Strategy Chief to Founder Former strategy director at Mail.ru Group, now founder of the Mo app and Academy, Alexander Gorny builds and scales AI-driven products. He approaches AI entrepreneurship with a pragmatic lens focused on defensibility, sales, and real economics. The guidance that follows distills actionable routes and constraints for launching AI ventures today.

Two Practical Paths to Start With AI One route is small, fast-to-build AI tools that solve narrow tasks and sell to consumers or very small businesses. The other is a long-game B2B startup that automates specific roles inside enterprises. The first earns “here and now” but is short-lived; the second demands years of focus and deep integration to become valuable.

Quick Wins With Small AI Tools Transcribing videos, simple bots, tiny plugins, and similar utilities can be built quickly and monetized at low monthly prices. Many niches exist and can be filled rapidly for immediate revenue. These projects are inherently temporary, because general-purpose tools or browsers will eventually subsume them.

Small AI Project Economics Are Power-Law Outcomes range from losing money to earning modest sums, with rare outliers making very large monthly revenue. Most projects top out in the low five figures per month at best, while many operate at a loss. A minority achieves significant scale, often by targeting international markets.

Durable Startups Automate B2B Roles Enduring value emerges by automating monotonous, high-cost roles within a specific industry. Integrations with legacy systems, compliance, and domain nuances create a moat that generic prompts cannot replace. The payoff comes from tangible cost savings—reducing headcount and capturing a share of the value created.

Sales Mastery Determines B2B Success Selling expensive software into companies requires specialized B2B sales skills. Either partner with someone who can sell or commit to learning through practice, content, and hard-earned iterations. There is no shortcut course that turns a developer into an enterprise seller overnight.

Russia’s B2B AI: Sber Moves, Few True Startups Sber actively sells GigaChat-based solutions across professions and shows traction, though at a small scale relative to its whole business. Genuine, defensible B2B AI startups inside Russia are scarce. More Russian-speaking founders have built successful AI startups abroad than domestically in this category.

Let Ambition Pick Between Pet and Decade-Play Pursue a pet project if the goal is enjoyable work, learning, and possibly a few million over time. Skip pet projects if the aim is a billion-dollar outcome. Ambitions, not trends, should choose the path.

Three Simple Idea Filters That Work If many similar builders already earn with the same simple tool, a fast follower can likely earn too. If a tool trends in the West but hasn’t reached Russian channels, localize it first and move quickly. If you own or can access an audience, build what they will buy and sell directly into that channel.

Russia’s Venture Is Sporadic, Not a Market Founders and friends fund thousands of projects, while angel checks cover only dozens. Few funds make a very small number of truly commercial deals. Corporates rarely invest or acquire; 2021 was an exception when Sber bought companies roughly weekly.

Don’t Count on Easy VC or Fast Buyouts A handful of venture successes exist, but there is no functioning market handing out checks for pitch decks. If there is a specific, defensible reason you can be one of the very few winners, try—but expect nothing by default. Otherwise, the dream of easy funding and quick exits will disappoint.

Corporate Appetite Comes and Goes Building for a strategic buyer can work if the insight and demand persist for years. Buying waves are cyclical: Mail.ru once acquired many education projects, then stopped. Reliance on a single buyer’s mood introduces material risk.

Realism Beats Pessimism About “Unicorns” Tech, revenues, and opportunities are growing, but not every year produces new giants. Avito took around 17 years to reach scale; Wildberries existed 15 years ago but was small. Expect progress without assuming frequent birth of mega-corporations.

Venture Thrives in Few Geographies—Accept It A true venture market exists in the United States, with partial echoes in the UK, Germany, Israel, and a few others. Structural factors—like long periods of low interest rates—built that ecosystem. Russia lacks those conditions, so builders must plan without them.

Three Build Strategies Now: Pet, B2B SaaS, Strategic Sale Pet projects can produce income and experience; B2B SaaS offers compounding value with focus and time. Building to sell under a specific strategic can work but is murky because buyers change course. A notable sale built for a strategic did occur, but it involved clawbacks and legal friction.

Layers on LLMs Need Vertical Moats Most AI apps are layers on OpenAI and peers, with no guarantees against platform replication. Defensibility comes from domain data, industry workflows, and integrations that platforms won’t prioritize. Vertical depth turns a replaceable layer into an irreplaceable system.

Platforms Leave Room, But Control Differs App Store history shows platforms could have internalized most categories but didn’t. Apple folded only a few areas in-house and left the rest to developers. Still, today’s AI differs because backend models are swappable and less gatekeeping than phone ecosystems.

Own the Customer and Swap the Model Apple and Google own user access and take large revenue shares, but LLM vendors typically don’t own the end customer. Multiple high-quality models exist and can be switched with minimal friction. When customers pay you directly, platforms have less leverage to squeeze.

Competition and Open Source Push Prices Down Model pricing has dropped sharply, and open models solve a large share of real tasks on self-hosted infrastructure. Commodity performance forces commercial providers to align pricing closer to cost. The ability to change backends keeps pricing pressure high.

General Layers Get Eclipsed; Go Specific Generic AI layers are vulnerable to being overtaken when platforms release broad features. A generalized assistant can quickly leapfrog custom wrappers built on the same backend. Specific domain depth is the safer route to durability.

Complex, Regulated Niches Are Safer The more contracts, licensing, legacy tools, and quirks a domain has, the less likely a platform will replicate it. Messiness becomes a moat, making copycats hesitate. Specialization increases stability and keeps competition thin.

Few AI Exits; Many Micro-Flips Only a few meaningful AI exits happen each year domestically. Meanwhile, tiny flips—like selling a small bot for a few thousand dollars—occur more often but amount to about a month’s developer salary. Builders should calibrate expectations accordingly.

B2C Growth Uses Ordinary Channels Paid ads, creator placements, podcasts, and partnerships remain the core acquisition levers. There is no AI-specific shortcut in consumer marketing. Unit economics must work; otherwise, growth stalls.

What Traction Looks Like for Consumer Apps About three million people have downloaded the app at least once. Year-over-year growth recently landed around two to three times. Tens of thousands currently maintain active paid subscriptions.

Virality Is Rare; Budget for Growth People share only extraordinary products; most tools won’t spread for free. Small free exposure can drive tiny businesses, but serious startups should plan to pay. Standout products like ChatGPT, Cursor, and Hippocratic demonstrate how real utility can still earn organic buzz.

AI Will Reshape Work Far Beyond Juniors Large IT teams already saw layoffs, and developer hiring has become harder for candidates and easier for employers. These are early breezes before a much larger storm across many professions. The next two to five years will bring far bigger shifts than today’s headlines suggest.

Coordination Will Shrink; Outcomes Are Uncertain Tools let fewer people orchestrate more, reducing the need for large teams and their coordination overhead. As staffing shrinks, additional roles become redundant, amplifying cuts. Visual Basic once promised fewer coders yet created more jobs; massive change is certain, the final balance isn’t.

Use AI Only Where It Truly Unlocks Value Add AI when it solves new, previously impossible tasks—not as decoration. The best opportunities are born from capabilities that didn’t exist five years ago. In Mo, AI features exist but remain peripheral to the core product.

Fortify the Right Flank and Plan Steadily Forward Specialize deeply so platforms won’t take what makes your product valuable; that right flank is the moat. Continue compounding in focused products—Mo, Academy, and a growing community—and make occasional angel investments when they fit. Persistence, specialization, and measured expansion set the course.