--- title: "What's left when the code is solved" description: "The VC checklist wanted a hacker, a hipster and a hustler. AI has filled the hacker seat. On what it would take to solve branding and selling too, and why the T-shaped generalist might be the best-placed founder now." date: 2026-07-31 image: "/uploads/2026/07/hipster-hustler-hacker.jpg" categories: - "AI" - "Startup" - "Future" --- There's an old piece of startup wisdom that VCs used to treat as a checklist. A founding team needs three people. A hacker to build the product, a hipster to make it look and feel right, and a hustler to sell it. Three seats, and if one stood empty you'd hear about it in every pitch meeting until you filled it. One of those seats now fills itself. The hacker seat. There was no single moment when it happened. It has been creeping up on us over the last six months, faster with every model release, until building software quietly became close to solved for most of what an early startup needs to build. Not because engineers stopped mattering, but because AI got that good. I know because I sit in that seat myself now, and I'm not a hacker. Which leaves two seats. Branding and selling. The interesting question is no longer whether AI comes for them too. It's what solved would actually look like, because they won't fall the way the hacker seat fell. ## The rarest co-founder became a subscription I wrote a few weeks ago that [the code was never the hard part](/thoughts/the-code-was-never-the-hard-part). For an early startup it barely qualifies as a part at all anymore. Agents plan, build, test and ship end to end. The technical co-founder, the person every VC insisted you couldn't start without, the seat founders spent years of equity filling, is now something you rent for a few hundred dollars a month. I feel this personally. I'm a designer by background, a generalist by nature, and I can now pretend to be the hacker convincingly enough to take a product from idea to production. Not at the level of a great engineer building infrastructure at scale. But at the level an early startup actually needs? Yes. The pretending works. ## Why the other two seats resist Here's the thing about code. It has a compiler. It has tests. When you're wrong, the machine tells you within seconds, and an agent can loop on that feedback a thousand times without you in the room. That tight loop is exactly why AI ate the hacker seat first. Brand and sales don't have a compiler. They're graded by humans, slowly and ambiguously. A brand decision proves itself over months. A sales motion proves itself over a pipeline of real conversations with real people who rarely tell you why they said no. The feedback loops that made code solvable barely exist here, and that difference explains almost everything about where AI is strong today and where it still isn't. So the two remaining seats resist differently. Branding is partially solved. Selling barely. ## What solved looks like for the hipster AI can already produce endless competent brand work. Identities, sites, copy, campaigns, all of it polished, all of it fine. That's exactly the problem. Everything is fine, and almost nothing is chosen. The output is plausible, an average of everything good that came before, when a brand needs to be a specific point of view held with confidence. Taste is choosing. It's saying no to nineteen decent options because the twentieth is right for this company, this market, this moment. And it's consistency, making decisions the same way for years so they accumulate into something people recognise. That's what a hipster co-founder actually did. The deliverables were never the job. Solved, then, looks like this: an AI that holds a brand position over time. One tuned to a single company instead of the average of the internet, with persistent memory of every decision made and why, aware enough of the market to know what everyone else looks like and steer away from it. Distinctive on purpose, not by accident. With a design and marketing background I can get maybe 80 or 90 percent of the way there today by directing AI hard. The models do the production, I do the choosing. The missing stretch is exactly that judgment layer, and I'd expect the labs to close a lot of it within a couple of model generations. Taste as a service sounds wrong, but brand memory as a service doesn't. ## What solved looks like for the hustler Selling is the hard one, and it's the seat I struggle with most myself. The mechanical layer is already gone. AI researches prospects, personalises outreach, writes follow-ups, keeps the CRM honest, runs a decent demo. But that layer was never really the sale. And the most visible AI sales tactic, flooding inboxes with generated outreach, killed itself. When every cold email might be a machine, every cold email is worth nothing. The value of a genuine human signal went up, not down. Because selling, at its core, is someone deciding to trust you. With their money, their time, their reputation inside their own company. That decision is social, and it moves at human speed. I can see three shapes solved could take. The first is the product selling itself, where AI makes the product so good and so instantly useful that distribution collapses into word of mouth, and the hustle mostly disappears. The second is agent-to-agent selling, where procurement runs through the buyer's AI and selling becomes making your product legible to a machine. Transparent pricing, real documentation, verifiable claims. Sales as engineering. The third is the narrow version, where agents run everything around the sale and the founder shows up only for the trust moments, the twenty minutes where a human decides to believe another human. My honest guess is that all three arrive, in that order, and that the trust moment stays human longest. Which means the hustler seat doesn't get solved so much as shrunk. From a full-time role to a few hours a week of the founder being genuinely, personally credible. ## The T-shaped orchestrator Where does that leave someone like me? I'm somewhere between hipster and hacker without being elite at either. A T-shaped generalist with a design background. In the old trio model that was a polite weakness. You were a partial person on a three-person checklist, and investors could see the gaps. I think it's quietly becoming the strongest position on the cap table. Because the job is changing from being excellent in one seat to orchestrating all three. Directing an AI hacker requires knowing what good software feels like, not writing it. Directing an AI hipster requires taste, not production skill. Even the shrunken hustler role rewards range: understanding the product deeply, the brand honestly, and the customer's world well enough to earn those twenty minutes of trust. Generalists with taste have spent their careers being told to specialise. It turns out we were early. I want to be honest about where that leaves me right now, though. The hacker seat I can cover with AI, genuinely. The hipster seat I can mostly cover myself, with AI doing the heavy lifting, though not to a hundred percent. The hustler seat is still hard for me, and no model currently on the market fixes that. The gap between what I can orchestrate and what I still have to be is exactly one seat wide. ## A trio of hats The trio isn't disappearing. It's becoming three hats one person wears while directing machines that do the rest. And the machines run on [intelligence on tap](/thoughts/intelligence-on-tap-is-addictive), which gets better every few months while the hats stay the same. What's actually scarce now comes down to two things. Taste, the ability to choose one right thing from infinite competent options. And trust, the ability to make another human believe you. Everything else on the old checklist is becoming a subscription. The code is solved. What's left is getting someone to care and getting someone to believe you. That was always the real company. Now it's the whole job.