Solo Founder Raised $20M Before Figuring Out What to Build
She sees Slack as a fifteen-year-old office building designed for humans—one where AI agents have no place. So as a solo founder, before fully clarifying her strategy, she raised $20M across two rounds to rebuild a messaging platform from scratch.
The video won't play here. Listen to the audio instead:
The argument · tap a timestamp to hear it
No competitors doesn't mean no opportunity—it means no one has the patience
She says she's also unsure whether it's early or late, but sees that Slack and Teams are moving too slowly on adapting their platforms for AI agents. She believes team messaging products need enormous patience to build well, and in today's AI gold-rush environment most people want quick returns, so the market is less crowded than it appears—the potential exists but the shape hasn't yet emerged.
— Sara DuInvestors back the person, not the fully-formed plan
She admits she raised capital as a solo founder before fully clarifying ‘this is the thing’—mainly because one investor shared her judgment on how AI and tool use will reshape work, and believed in her way of operating. She sees this as what early-stage investors do well: assess people, not the plan itself.
— Sara DuSlack is an office building built for humans; AI agents can't live in it
Her core claim: Slack's data model has never treated AI agents as entities equivalent to humans. Agents can't truly understand which channels they can join, how to have side conversations with people, or how to start group chats. She compares it to placing a superintelligent being into an office not built for it—the doors and hallways don't work, it has to operate with minimal context from people, and it's blamed when things go wrong.
— Sara DuHide the AI agent's thinking process
She strongly opposes showing AI thinking traces in the UI because watching a progress bar jump from one second to sixty is deeply frustrating. Ando deliberately avoids this interface because the goal is to treat AI teammates more like human ones—people need hours or even days to ‘warm up,’ not instant mind-reading.
— Sara DuHumans show zero tolerance for agent mistakes, endless patience for human ones
She observes that people are intolerant of AI error in a way that mirrors the opposite of human trust-building: trust between humans grows slowly but becomes resilient once earned. Trust in AI grows just as slowly but collapses instantly. She thinks this is a hard problem platforms face—deciding how much interface design can help an AI agent earn human user trust, where the design itself bridges that trust gap.
— Sara DuBuilding a messaging platform is closer to game design than B2B SaaS
She says what surprised her most is how much human feeling matters—this is more like making a game than standard business software, which is why messaging founders historically come from games. The reason: what users feel as they move through an interface is something large models currently struggle to simulate, whereas human designers excel at it.
— Sara DuBefore finding product-market fit, fewer people makes for faster learning
She's become radically minimalist in hiring and execution. After easy fundraising, many founders think ‘more money means hire faster, do more,’ but she believes the only thing that matters before product-market fit is learning velocity, and more people slow learning down—too many cooks. So focus on core metrics: what exactly are you validating?
— Sara DuThe cost of solo founding: no one to hold your full context
When asked about solo founding's downside, she says the biggest risk is overthinking from isolation—there's no peer to press on difficult questions with, even if smart people outside the company could help. The upside: a company becomes its own entity with a soul; a solo founder can more clearly hear what it truly needs next, and decide who deserves to become co-founder.
— Sara DuIn their own words · checked verbatim
It's truly like work life integration. If you want to start a company, especially solo, you just have to like love it so much. You're like addicted to it.
Sara Du4:01
Slack, you know, it's like this big office building that was built 15 years ago for humans.
Sara Du32:08
You're literally just dropped into this office building that was not built for you. Like you can't really use the doors or hallways properly.
Sara Du33:09
I'm actually very anti-thinking traces because I think it's so crazy that, like, imagine if between you and I, like, you could see, like, Sarah working one second, two seconds. It's 60 seconds now.
Sara Du36:18
humans are so intolerant of agent mistakes versus humans. It's like the inverse of that, like human trust takes a while to gain, like the speed of that is slow. And then the speed of losing trust is fast with agents.
Sara Du42:33
It's, like, more like building a game than a B2B software company. So that's why if you study the founders of messaging platforms from previous eras, they're all actually gaming founders, because it's not a standard B2B SaaS build.
Sara Du52:43
I think a lot of solo founders probably overthink things, right? You don't really have many people you can bounce ideas off of, or more sensitive topics off of, because like even very smart people outside the company don't have the full context.
Sara Du1:09:06
companies are entities. They have souls of their own, and if you're a solo founder, you can actually listen to what the company needs based on how it's growing, and you can choose who might be worth being a co-founder later on in the journey
Sara Du1:10:52
Figures
| Product build time | ~1 year | 30:02 |
| Beta users | 100+ | 30:02 |
| Countries in beta | 14–15 | 30:02 |
| San Francisco team share in beta | 15–20% | 30:02 |
| Startups that fail due to people issues | ~65% | 18:41 |
Glossary
- Model Context Protocol
- Protocol released December 2024 that makes it easier for large language models to call external tools.
- Thinking traces
- UI that displays an AI agent's active reasoning process, such as progress indicators.
- Nerd sniping
- Using a small, intellectually interesting task to draw someone in part-time until they can't stop.
- Product-market fit (PMF)
- Stage where a product has proven itself needed by its target market.
- RFC
- Short document written before work begins to clarify a feature's intent.
How to listen
Founders and product managers deciding whether to start alone, or designing human-AI collaboration tools.
53:43–58:49 covers childhood and athletics with low information density.