Made it to a YC interview, and found out we love the grind
In February a friend and I took BoBe, our AI assistant, to a ten-minute YC interview. We didn't get in, and it was one of the best things to happen to us.
On 24 February at a quarter past seven in the evening, a friend and I joined a video call with a partner at Y Combinator. It was booked for ten minutes. We were there to talk about BoBe, the AI assistant we had been building for a few weeks.
Project managers for bots
The idea came from our own days, which we both spend inside AI tools. In the application we wrote that our job “had shifted from doing work to managing AI tools”. We were opening the right chat, restating context and copying answers from one app to another. “We’d become project managers for a team of bots that forget everything between sessions.”
BoBe was the answer we pitched. It would live on your computer, notice what you work on and act without being asked. “If you spend an afternoon debugging across a codebase, docs, and Stack Overflow tabs, you wake up to a PR with the fix already drafted.” Everything it saw and remembered would stay on your own machine, in a memory you can look at.
Tools that watch your screen make people nervous, and they’re right to ask where all of it goes. We started from that worry: the assistant’s memory belongs to you, sits on your disk, and you can read it.
Two weeks, then an application
We sent the application on 10 February, with a demo video and a short video of the two of us. It was honest about where we stood: “a working demo and a PoC”, built over “a couple of weeks”. The app was an Electron window talking to a Python program on the same machine, which ran open models through llama.cpp.
We had talked to about 30 people, from researchers to friends who don’t care about AI tools at all. We described something that finishes the half-written documents and forgotten follow-ups on their computers. In our words, “every single one said ‘I need that.’” We promised a version people could use by the end of February.
Ten minutes, then a week
The invitation came on 21 February. The call itself was over by 19:25, which is how YC interviews are meant to go. Afterwards we sent the partner links to our demos, and a week later the answer came.
It was a no, and a generous one. They liked the idea of an agent that doesn’t wait for you to ask, and wrote that they believe that’s where the future is headed. Their concern was distribution. Large companies were building the same kind of proactive assistant into products that millions of people already use. Being local-first and private by default was a real difference, but they weren’t convinced it could win that race.
What we took from it
It wasn’t a bad day. The questions made us look hard at the difference between what we wanted to build and what people would choose to use. We had been building for ourselves. That’s a fine way to make a tool, and a risky way to start a company.
It also pushed us into a pivot that turned out to be the right one. And it showed us something about ourselves. The application had said it comes naturally to us “to lock in, ship, and debug at 2 a.m.”
The weeks around that interview proved it. We love that grind.
It didn’t go to waste
BoBe didn’t become a company, but it didn’t stop either. I kept modernizing it: a Rust core and a native Mac app, then a move onto an agent harness in May. Its memory became one Markdown file you can read and edit. It learned to decide when to speak first without turning into Clippy. It got a voice that runs on your own Mac and a body made from a toy robot.
We shared it with friends and family, and today it’s a personal assistant we use ourselves. It’s open source under the MIT license and reached version 0.9.2 in October. In March, Hermes Agent came out, an open-source agent “that grows with you”. It’s close to what BoBe turned into for us.
Seven months later
In September and October, the biggest names in AI shipped the idea we had pitched in February. Meta introduced Muse on 8 September, a personal agent for everyday tasks and longer-term goals that keeps working after you close the app. OpenAI announced dots at DevDay on 29 September, a “personal agentic assistant” that works on your goals in the background. On 21 October Anthropic opened a preview of Claude Duet, which “monitors connected applications, surfaces priorities and carries out assignments in the background”.
So we were early, by about seven months. The partner was right too. They never doubted the direction, only who would win the distribution. All three launched inside apps that millions of people already use: WhatsApp, ChatGPT and Claude.
BoBe still does one thing differently. Its memory stays on your disk, and its voice runs on your own Mac.