Julian hosts the first public Solo Founders Office Hours, working live with five solo founders on their hardest current problems — ranging from monetizing an open-source gaming app with 1.25M users to navigating a fresh co-founder breakup, sequencing AI vs. workflow features in a wholesale platform, and deciding whether to lead with software or hardware in a printable robot kit.
Utkarsh Dalal (GameNative): monetizing 1.25M users of an open-source PC-games-on-Android app
The product has massive organic reach (1.25M users, 50K Discord, 10K GitHub stars) but only a few hundred paying subscribers for perks like Discord roles and early builds.
Perks are a fan club, not a product — people pay to support, not because the perks unlock core value.
User trust doesn’t matter if you don’t exist — sustainability requires revenue, and transparency about monetization decisions will retain the right users.
Charge for what you build next, grandfather the rest — don’t take away existing free features; create new, high-value capabilities worth paying for.
The core unlock is playing PC libraries on the go without a home PC — nothing else comes close, so the monetization lever should sit on that unique value.
Ryan (Crash Labs): one week into solo founding after a co-founder breakup
Customers are the orienting force; building is the addictive default — without paying customers, day-to-day focus drifts; talking to users clarifies priorities.
Ship a public benchmark so labs come to you — inbound from researchers beating your benchmark is stronger than outbound sales.
Show them the gap, then sell the fix — demonstrate a model’s specific deficiency (e.g., social reasoning in Mafia), then offer the curated RL environment that fills it.
Don’t be the best. Be the only. — differentiate on a niche no one else serves (continual-learning RL environments) rather than competing head-to-head.
Solo-founder credibility comes from proof of work — your Cohere background, the benchmark you’re publishing, and the specific problem you articulate matter more than team size.
Charlie Khan (Brio): the features customers ask for vs. the AI layer they don’t
Buyers and sellers think in familiar workflow features (tabs, upload buttons) because the plain-language AI layer didn’t exist until recently — they can’t imagine the alternative.
MagicSchool’s micro-app middle path — constrain the LLM into specific “jobs to be done” (lesson-plan generator, rubric builder) so users get immediate value without prompt engineering.
They’re asking for jobs to be done — each feature request maps to a concrete task; micro-apps deliver those tasks with the flexibility of AI and the specificity of UI.
Things that take long take long for a reason — the 2–3 week wholesale order cycle may serve hidden incentives; investigate before assuming compression is wanted.
For 200 years, buyers and sellers have communicated in natural language — Brio’s thesis: just “text Joel” and let every transaction make the next one faster and smarter, without behavior change.
Andy Wang (Infinomni): software or hardware first for a prompt-to-printable-robot kit
The snap-your-fingers question — if you had a partner and budget, you’d accelerate hardware testing (boards, actuators, wiring) and build custom character prototypes for early users.
Put the character on a phone first — run the “mind” (personality, conversation, memory) on a docked phone; the body arrives later as a “brain transplant” that frees the character into the physical world.
Draw the owl; don’t split your time — avoid parallel software/hardware tracks; ship the compelling character experience digitally now, iterate fast, add hardware when it’s the clear bottleneck.
The mind before the body — what makes a being a being is memory, communication, recognition — not the exoskeleton; dementia is tragic because the mind fades, not the body.
Physical articulation matters per character (Pikachu vs. Snoopy move differently) — but solve that after the character feels alive on screen; the body is a smaller part of the value in early days.
