What Solo Founders Lack Isn't Hands, It's Someone to Make Them Tell the Truth
Julian's diagnosis of five solo founders points at almost the same thing every time: without a co-founder, you lose the person who forces you to face what you actually did today.
The video won't play here. Listen to the audio instead:
The argument · tap a timestamp to hear it
Hundreds of paying users means you haven't given them something worth paying for
Utkar's Game Native has 1.25 million users, but only a few hundred pay $5 to $10 a month for a Discord role, priority support and early builds. Julian's read is blunt: all of that is "bonus," not core value — it's essentially a Patreon-style fan club, where people pay to support you, not because it's genuinely a good deal. His test: a few hundred paying out of 1.25 million is precisely the proof that you haven't yet offered something truly worth paying for. And don't charge by taking existing features away — create something new in the future that users haven't used yet.
— JulianSolo founders have no accountability partner, and that's the deadliest part
Julian says solo founding is hardest from zero to one because "there's no accountability partner." He uses the gym as an analogy: if you've already agreed to train with someone every day, you're more likely to actually show up. Early on, with no customers, everything is fuzzy, and you're the only one judging whether you're spending your time right. His antidote: be a little kinder to yourself, while staying honest about what you're actually doing each day. He also names a default trap — writing code is "the addictive default option," because it always makes you feel like you're making progress, but the only thing that truly settles the question is paying customers.
— JulianLet the lab see its own gap first, then sell the solution
Ryan sells RL environments and benchmarks to frontier labs. Julian endorses his path: rather than leading with generic data, first use a public benchmark to get the lab to admit "we really do have this gap," and only then offer that you can fill it. Ryan says this is already standard practice among frontier-lab data vendors — show that some model fails at some capability, and a few months later the research lead comes to you asking whether they can train on your data. Ryan's benchmark is a bunch of frontier models playing Werewolf against each other; what it measures isn't who wins, but reasoning ability, who lies best, and who lies most.
— JulianYou don't need to be the best, you need to be the only
Ryan worries that being solo will make labs dismiss him. Julian says the worry is partly valid, but only if you're competing head-on with others. He quotes Kevin Kelly: you don't want to be the best, you want to be the only. So the question isn't "how does this team look," it's "is this the only person who can help us solve this problem." He also says being solo is "more in your head than in anyone else's" — you can simply answer: I'm just starting out, I only work with a few people, I'll be hiring, and I have AI leverage.
— JulianUsers don't want features, they want a button they can press
Charlie is torn between building the workflow features customers ask for and going straight at the AI brain that would eliminate all workflows. Julian tells the Magic School story: founder Adil was a former high school principal who wanted teachers to use ChatGPT, ran a whole-school training, and nobody actually used it. He judged that the blocker wasn't AI being too weak, but that a blank chat box is too scary — it can do anything, and therefore feels like it can do nothing. So Adil built dozens of "micro-apps," each doing one specific thing. First week: 100 users. Second week: 1,000. Julian's conclusion: when users request a feature, they're really telling you the jobs to be done.
— JulianA slow deal may not be an efficiency problem, it may be an incentive problem
Charlie says both sides of a wholesale trade know exactly what they want, yet it takes two or three weeks of back-and-forth emails and CSVs to close. Julian doesn't answer directly; instead he asks a counterintuitive question: if everyone wants to compress the time, why hasn't it been compressed? Could some incentive be deliberately stretching it out? He says sometimes things drag on for a reason — just not the reason everyone assumes. He also suggests using frequency × severity to decide which problem to tackle first.
— JulianBuild the brain first, don't start with the body
Andy wants to build "describe a robot, get a design and a 3D-printed kit," and is torn about whether to bring someone in to split software and hardware. Julian turns it around: if you were given a full-time person and enough money, what would you do differently? Andy says he'd put it all into 3D printing and hardware testing. So Julian suggests: could you skip hardware for now, run the character on a phone, put it on a desk as a base — the character can't move, but it exists; then the robot's first boot is releasing the character from the phone into the physical world. He uses dementia as an analogy: what makes a person a person isn't an exoskeleton, it's memory, a way of communicating, the look in their eyes when they recognise you. So early on, the body may matter less than the mind.
— JulianDifferent characters move differently, and that's exactly the hardware problem
Andy explains why he can't let go of physicality: Pikachu, Snoopy, a rover — they move through physical space in completely different ways. If you want to bring Pikachu or an owl into the real world as a robot, you have to find a solution for "different characters moving in different ways." That's why he weights physicality so heavily. Julian concedes the point holds, and says if he built an owl it would be "a real hoot." This is the only moment in the session where a founder pushes back on Julian's advice head-on.
— AndyIn their own words · checked verbatim
But that is so what we do uniquely over GameHub is that we are open source, right? We are transparent, and people really care about that.
Utkar5:05
I think that um user trust doesn't matter if you don't exist.
Julian6:06
being a solo founder is not easy. like it is really hard uh in a lot of ways. Um particularly it's hard in the zero to one phase before you have customers before you have a team.
Julian11:08
I think that building is like the addictive default, right? Uh because it always feels like you're making a little bit of progress, but really the rubber only meets the road if you have, you know, paying customers.
Julian12:09
You don't want to be the best, you want to be the only.
Julian21:16
it was really intimidating because it could both do anything, but by the ability to do anything, it could also kind of feel like it couldn't do anything at all.
Julian26:20
the thing that makes a person a person isn't just their sort of exoskeleton right it is who they are is their memories. It's the way that they communicate.
Julian42:33
Figures
| Game Native users | 1.25 million | 1:03 |
| Game Native paying users | a few hundred | 3:04 |
| Game Native subscription price | $5 to $10 per month | 3:04 |
| GMV of Charlie's customer leaving Shopify | $15 million | 23:17 |
| Magic School's user growth in its first two weeks | 100 in week one, 1,000 in week two | 27:22 |
Glossary
- RL environment
- A training setting in which an AI agent repeatedly tries, fails and gets feedback.
- jobs to be done
- What a user actually wants to accomplish, rather than the feature they say they want.
- GMV
- The total value of goods transacted on a platform.
- ERP
- Back-office systems companies use to manage orders, inventory and finances.
How to listen
Founders building solo and stuck on whether to build product or find customers; and anyone who wants to see how an investor breaks down an early business model on the spot.
The host's intro from 0:02 to 1:03 can be skipped — go straight to the founder questions.