One Founder, Zero Backend Engineers: How AI Agents Are Writing Production Code at Plënka

# One Founder, Zero Backend Engineers: How AI Agents Are Writing Production Code at Plënka

A pre-launch stock photo marketplace. One founder. No backend engineers. And an AI agent pipeline that drafts, reviews, and rewrites its own code.

That's Plënka — and its founder's recent post on DEV Community is one of the more honest accounts of what the one-person startup model actually looks like when the technical stack is handed to agents.

What Plënka is building

Plënka is a stock photo marketplace with a straightforward differentiator: photographers keep 75% of revenue. The industry standard — set by Shutterstock, Getty, and their peers — sits between 15% and 40%. That 75% figure isn't a rounding error. It's a direct attack on the incumbent model.

But what makes Plënka interesting for this publication isn't the licensing split. It's the engineering setup behind it.

The entire backend is written by AI agents.

What "written by AI agents" actually means

This phrase gets thrown around loosely. In Plënka's case, it means something specific.

The founder describes a pipeline where agents don't just generate code — they critique it. One agent drafts. Another reviews. A third rewrites based on that review. The output gets tested, flagged, and iterated on without a human touching every line.

This is multi-agent code review, and it's meaningfully different from a solo founder using Copilot to autocomplete functions. The distinction matters because it scales differently. A founder using autocomplete is still the bottleneck. A founder running an agent review loop is closer to managing a team — setting direction, reviewing outputs, and unblocking the pipeline when it stalls.

The founder is explicit about what this costs in attention: context-switching between product decisions and agent orchestration, debugging agent outputs rather than writing code, and maintaining prompts the way a CTO maintains engineering standards.

It's not passive. But it's also not a five-engineer backend team.

The day-to-day reality

The DEV Community post describes a working rhythm that looks less like "founder writes code" and more like "founder manages an AI development process."

On a given day: - Agents are generating route handlers, database schemas, and API logic - The founder reviews outputs for correctness and edge cases - Failed tests or broken logic gets fed back into the agent loop - Product decisions — pricing structure, search logic, photographer onboarding — still require human judgment

The founder's disclosure at the bottom of the post is worth noting: the post itself was drafted with AI assistance and published by the founder. Same setup. Same philosophy applied to content as to code.

This is the AI-native company model applied consistently — not just to one function, but to the operating system of the business.

Why this matters for revenue per employee

The economics are where it gets interesting.

If Plënka launches and reaches, say, $500K ARR with one employee (the founder), that's a revenue-per-employee figure that would have been structurally impossible five years ago for a marketplace with a functioning backend, search, payments, and creator payouts.

The capital that would have gone to two or three backend engineers — conservatively $300K–$450K annually in salary and overhead — stays in the business or never gets raised in the first place. For a bootstrapped founder, that's not an efficiency gain. It's the difference between needing investors and not needing them.

This is the core thesis behind the one-person unicorn prediction: AI doesn't just make founders faster, it removes entire cost centers from the org chart before they're ever added.

The honest limits

The post doesn't oversell the setup, which is why it's credible.

Agent-written code still breaks. Context windows still lose track of complex architecture decisions. Agents hallucinate dependencies. They write code that compiles but doesn't behave correctly under load. The founder is still the QA layer — and that's a real job.

There's also a harder question the post doesn't fully answer: what happens when Plënka scales? At 10,000 photographers and millions of assets, the agent pipeline that works for an MVP may need rearchitecting. That rearchitecting probably requires a human with deep systems knowledge, even if they're using agents to implement it.

The one-founder-plus-agents model isn't infinitely scalable. It's scalable enough to get to revenue, to product-market fit, and potentially to a point where the business can fund the engineering talent it eventually needs — without giving up equity to get there.

What founders can take from this

A few concrete observations from Plënka's setup:

Multi-agent review loops are worth building. Single-agent generation catches maybe 60% of obvious errors. Adding a review agent — even a simple one prompted to act as a skeptical senior engineer — meaningfully improves output quality.

Treat prompts like documentation. The founder's prompts are effectively the engineering standards for the project. Version them. Refine them. They're not throwaway inputs — they're the architecture of your development process.

Separate what agents decide from what you decide. Agents can generate the code for a photographer payout flow. They should not decide the payout percentage, the dispute resolution policy, or the photographer onboarding criteria. Keep the product logic in your hands.

Publish what you're doing. The DEV Community post brought Plënka more attention before launch than most pre-launch startups get. Transparency about the AI-native model is itself a distribution strategy.

The bigger picture

Plënka is one data point. But it sits alongside a growing number of companies being built by single founders with agent stacks handling work that used to require teams. Building a one-person startup with AI is no longer a thought experiment — it's a repeatable approach with documented examples.

The question isn't whether AI agents can write production backend code. Plënka answers that. The question is whether founders know how to manage that process well enough to ship something people actually pay for.

That's the only question that matters.

---

Is your company eligible? Submit to the leaderboard → onepersonunicorn.co/submit