Current thoughts on building and investing in AI apps
We have barely scratched the surface of the vertical AI opportunity
We have barely eaten into labor budgets ($11T; customer service is one of the few examples)
Today, AI is more of task automator than a job automator
But the rate of AI improvement makes it clear that job automation is within scope (many jobs are just a series of tasks)
[Voice AI may actually be the wedge into the labor budget across many industries]
The current best framing is not to think of “jobs to be automated” rather, it’s to think of “tasks to be supercharged” by AI
Despite the big SaaS selloff recently, it’s not yet clear if incumbents will lose staying power.
Incumbents will also add AI to their offerings (e.g. Epic with medical scribes)
Incumbent systems of record are clamping down on access to SOR data (e.g. Salesforce blocking Gong)
Getting to MVP functionality is easy with a vibecoded app. Being production-grade is another matter
OpenAI still runs on Workday
How do you look for an opportunity in the AI era? Think of being 10x better. Two ways to do so:
1)Find a workflow bottlenecked by human capacity. Turn something scarce into something abundant (e.g. answering every phone call, reading every document on the internet about a prospect before a meeting)
1)Instead of rebuilding the SOR, solve a completely AI-native problem where there isn’t an incumbent (e.g. AI embedded on your website that gives sales demos and product tours to visitors)
Being 10x better is actually the easier problem. Building defensibility is the harder one. Strategies for defensibility
Ensure you are doing hard things beyond the AI (e.g. you can create an image in ChatGPT, but you can’t edit it there)
All-in-one offering (e.g. Posthog vs Amplitude)
Build a network (effect) Ă e.g. Statusphere
Generate proprietary data from your app that actually makes the product offering better (e.g. Terrakotta has live building data from calls, Alex has live candidate data)
Deliver the end-to-end outcome
The threat for the big labs going after your opportunity is overstated. It’s easy to get to an MVP level solution, but in reality, the surface area needed to solve a problem well is often much deeper than first envisioned.
Once again, it forces you to ask yourself: what are you building that is hard that isn’t AI?
The new growth expectations are substantially higher than they were before and this isn’t changing any time soon
