Eng stack for a typical 2-person startup today
2 comments
Glad to see real outcomes from small teams.
> Once Claude familiar with your repo (ask it to make copious notes), async coding is the next level of productivity. I now ask it to write a lot of the repetitive code paths.
Can you share what is your approach to context engineering this ?
Also what AI tools you use on daily basis ?
> Once Claude familiar with your repo (ask it to make copious notes), async coding is the next level of productivity. I now ask it to write a lot of the repetitive code paths.
Can you share what is your approach to context engineering this ?
Also what AI tools you use on daily basis ?
Hate to break it to you but if nextJS is your front end you basically are vendor locked to vercel..
I think we’ve essentially unlocked unlimited engineering capacity for routine tasks, thanks to these AI tools.
First, we built everything serverless and chose Vercel. The bigger cloud vendors are great, but unless you have hundreds of customers, they’re often overkill. With serverless, unless there’s usage, there’s no cost. At some point the cost will exceed the value, but we’re not there yet. Vercel is serverless-first. This is not an ad, our usage has not crossed $20/month (compared to thousands somewhere else).
Second, DuckDB has become our in-memory engine for analytics where, in the past, we would’ve needed to run something like Snowflake. Even for medium analytical tasks (parse and analyze a massive CSV), DuckDB is brilliant. We’re also experimenting with making this scale by building our own DuckDB cluster and trying tools like MotherDuck to see how the cost compares.
Third, we heavily use Claude Code Web. Once Claude familiar with your repo (ask it to make copious notes), async coding is the next level of productivity. I now ask it to write a lot of the repetitive code paths. I put my kid to sleep, kick off 10 tasks, and wake up with usable code the next morning. I’m sure Codex is good too.
Fourth, we made everything testable and runnable by LLMs. Every new piece of functionality comes with clear debug routes that an LLM can understand and iterate on. We tell the LLM: you build it, you test it, you fix it, you Q/A it, come back with a report.
We also built everything in containers, so we’re not locked into a single cloud vendor and can deploy into a customer’s own environment easily. LLMs also understand these services the best, because they’re open-source.
So what does this mean?
There’s a lot of talk about human vs AI productivity right now, but I’m actually super bullish about the future. I think we’re going to see many more David-vs-Goliath stories. Small teams will win more often. We will have more time to build things the right way, and more time to talk to customers and learn.
Sharing this in case it’s useful to others building early.