Hey all, I’m Mandeep. For the past two years I’ve mostly seen the Bay Area from the outside, despite living here. I moved here from London with my family to build Pump Up alongside my cofounder, and we did what you’re supposed to do at that stage: put our heads down and built.
Heads-down was the right call - classic PG advice. It’s also a habit that quietly becomes a dangerous default, putting everything else on hold and focussing on shipping regardless. However, we’re in the middle of the most interesting industry shift either of us can remember, in the city where most of it is happening, and consuming it all from the sidelines feels limiting. So, time to start writing.
Some background on me. I’ve been building products with a similar theme for twenty years. During my DPhil at Oxford I built a language and compiler so biologists could run serious simulations without touching systems code. At NStack we built a typed workflow platform in Haskell so data scientists could ship models to production without becoming infrastructure engineers. At Datapane we built an open-source Python library so analysts could turn their work into reports and data apps without learning a web stack. The through-line: letting technical folks run serious computation without becoming expert programmers.
Agents are that same problem 10x. At Pump Up we’re building multi-player agents that run business processes alongside the people who own them, and the audience isn’t data scientists or analysts anymore, it’s anyone with a repeatable process and a job to be done. Making that work in production, reliably enough that a business will trust it, is the most fun, fast-paced, and volatile engineering I’ve done.
So that’s what I’ll be writing about here. Three threads to expect:
Building with agents. How agents are rapidly reinventing how startups run, and dispatches from building ours: why our runtime has no job queue, what happens when you kill the process mid-run, and the ways LLM APIs surprise you in production.
Modern Java at an AI startup. We’re an AI company in SF choosing to build in modern Java in 2026, and it’s been a quietly excellent, if contrarian, decision. First proper post on this soon, just constructing the BlogPostFactoryFactory first.
War stories. A decade of building developer platforms and open-source tools left me with a stack of lessons that only became legible in hindsight: Haskell in production with eight engineers, growing an OSS library to 1.5k stars and learning what stars don’t buy you, pivots, building OSS communities / politics. The stories are better than the lessons, so I’ll lead with those.
Expect a longer piece every few weeks with random thoughts and musings in between. If any of this is your world too: say hello at mg@mandeepgill.net, or find me on X. And if you’re in SF, always happy to grab a coffee and a chat.