Loading...
Loading...
Please wait
Building software has never been easier. Keeping it alive has never been harder. Most founders adopting AI right now have only priced in the first half of that sentence.
Richard White is the founder and CEO of Fathom, the top rated AI note taker on G2. He started the company just before 2020 on two bets almost nobody agreed with: transcription costs would fall to zero, and AI would get good enough to do something useful with what it heard. Both were right. He breaks down what actually changed, what didn't, and why the maintenance cycle is the part nobody warns you about.
A new frontier model lands every three to six months. The other side of that coin is that a model gets deprecated every three to six months too. Build on one version and you have about six months before you rebuild on the next.
Richard explains why Fathom is moving workloads off frontier models and onto open source, not to save money, but because the upgrade cycle is unsustainable for anything you intend to maintain. He walks through why a purpose-built pipeline running five or six models still beats a single general purpose call, what happens to accuracy when you're searching for something that appears in one percent of your meetings, and why the GPT-5 release that landed flat commercially mattered enormously to anyone solving retrieval problems.
Then he flips it. Fathom operates like a Formula One team because it competes at the frontier and throws away the engine after every race. A normal business isn't in that race. Move your build from one model version to the next and it'll be slightly worse and close enough that you won't care. The maintenance cost is real. It is not a reason to wait.
This is for founders and operators making real decisions about AI inside a business that already generates revenue, and for domain experts sitting on knowledge they've never been able to productize. Software markets that were never worth raising against are now buildable in a weekend by the person who already understands the customer.
Questions Answered
Why has building with AI become easier while maintaining it has become harder?
Why is Fathom moving from frontier models to open source?
How should a business owner adopt AI without it becoming a full time job?
What replaces the meeting when AI captures and routes the information for you?
Why doesn't dumping all your transcripts into a chatbot work?
What does managing AI agents have in common with managing people?
How does model capability map to what you can safely delegate?
Can a domain expert now build profitable software without funding or a team?
Is headcount still a useful proxy for company size?
Looking to dive deeper into these conversations and connect with our host and guest?
Follow Richard White:
Follow Kayvon:
Want to go deeper with Kayvon?