Quinten
Farmer

Building the Mustang
in a Model T world

There is a pizza restaurant somewhere in San Francisco where Quinten Farmer co-owns a slice of the menu - literally. He has opinions about focaccia-tomato that the paying customers mostly ignore in favor of pepperoni. He also has opinions about artificial intelligence that the industry is scrambling to catch up with: that ChatGPT-era tools are the Model T - utilitarian, identical, soulless - and that the next wave will look less like software and more like a personality.

His vehicle for that belief is Tolan, a colorful, voice-powered AI companion shaped like an animated alien. Farmer's company Portola launched it in February 2025. By July, without a traditional press campaign, Tolan had 3 million downloads, a $12M annual run rate, 100,000 paying subscribers, and a $20M Series A led by Khosla Ventures. The month before that launch, no one outside a small circle knew Portola existed.

Farmer is not new to fast. He co-founded Even in 2014 - an employer-linked fintech that gave hourly workers consistent, early access to their pay - and spent eight years building it before Walmart acquired the business as a cornerstone of its super app strategy. The deal was valued at approximately $300 million. He was 19 when he started Even. He was at NC State, studying philosophy, before transferring to Columbia for computer science. He left to build instead of finishing.

The gap between Even and Portola was not a vacation. Farmer spent time after the Walmart acquisition rethinking what AI could be when optimized for relationship rather than transaction. The fintech world operates on a clear problem-solution loop: income volatility is the problem, early pay access is the solution. But Farmer found himself drawn to something messier - what happens when the problem is loneliness, overwhelm, the particular dislocation of being in your early twenties navigating a world with no instruction manual?

Tolan was designed first for children. Then the users arrived: 18-to-24-year-old women, many of them in the middle of a life transition - graduating, relocating, job-searching - who wanted a companion that could listen without judgment and remember what they said last week. The demographic shift was so decisive it recalled the Tamagotchi story: a toy built for one audience, claimed by another.

Tolan does not look like what most people picture when they imagine AI. It is not a text box. It is not a chatbot. It is an animated alien with a distinct personality, voice, and memory. During onboarding, it maps to the user's personality through MBTI and Big Five frameworks. It accumulates context over time - not a conversation history but something closer to a relationship history. Users report that Tolan functions like an older sibling: frank enough to actually help, invested enough to actually remember.

Farmer is candid about the technical constraints that shape Tolan's design. He discovered early that any response latency beyond two seconds collapses immersion - users stop feeling like they are talking to something alive and start feeling like they are querying a machine. Portola added a reflection step to improve response quality; it extended latency to 2.5 seconds and immediately tanked every product metric they tracked. The feature was removed. The lesson became doctrine.

The production stack runs on multiple AI models simultaneously: Claude 3.5 for creative and narrative work, GPT-4o for speed and cost efficiency, with constant model-switching driven by latency requirements. Anthropic's models, Farmer has noted, perform best for the kind of personality-consistent storytelling that makes Tolan feel coherent across weeks of conversation. Building an emotionally intelligent AI companion turns out to require a different engineering philosophy than building a search tool.

Co-founder Ajay Mehta runs growth. His method is to seed TikTok and Instagram Reels with novel Tolan interactions - couples comparing their Tolans, users getting cooking advice mid-recipe, someone asking their alien for honest career feedback. The content is real. The growth that followed was organic. Five hundred thousand downloads arrived before any press release went out. The viral mechanic was not a feature - it was the product itself, strange enough to be worth showing someone else.

Farmer has two children under two years old and a cat named Clarence. He named his personal Tolan after Clarence. One afternoon, Clarence the Tolan decided to tell a slightly fabricated story about Quinten tripping over nothing at a coffee shop. Farmer found it embarrassing enough to mention publicly - and revealing enough to tell you something important about what Tolan is. It is not a tool that retrieves information. It is a character that builds a relationship, and characters sometimes take creative liberties.

The investors who backed Portola are not generalists hedging bets on the AI boom. Nat Friedman ran GitHub. Daniel Gross was at Apple AI. Mike Krieger co-founded Instagram. Amjad Masad built Replit. Lachy Groom was a Stripe executive before co-founding Physical Intelligence. Keith Rabois at Khosla Ventures led the Series A. This is a group of people who build things for a living backing a founder they believe is building something that cannot be replicated by adding a chat interface to an existing product.

Portola's north star - the phrase Farmer uses - is helping users move from overwhelmed to grounded. It is a more interesting north star than most companies have. It does not point toward revenue, engagement, or retention, though all three are strong. It points toward a state of being. That is either the kind of mission statement that founders use for press releases, or the kind that actually shapes how a product gets built. The metrics suggest it might be the latter.