Now

Most of my attention goes to a single question: how do you take fraud and trust defences from "good enough" to an order-of-magnitude better without adding an order of magnitude more people? That means fewer, sharper bets — small, self-contained teams that own a problem end to end, and detection that recovers value rather than just flagging it.

A working set of AI agents I use every day, with a runtime that dispatches them on demand. Public on GitHub. It's how a lot of the rest of this list gets done.

A four-reader resume analyser I built and shipped end to end in a single Claude Code session. Live at resumelens.vaibhav.sh. It's the proof, for me, that the line between what a product person can specify and what they can ship has moved.

A small tool for thinking about financial independence as a destination you keep, not a one-time answer. Live at firefly.vaibhav.sh.

A practitioner series for product people who want to actually use these tools, now on a slower cadence.

A newer publication on fraud, trust, and marketplace integrity in the AI economy — the systems side of the same coin. theintegritybrief.substack.com.

A book in active draft, on how companies accumulate structural trust failures through small, sensible-at-the-time product and organisational decisions.

One idea underneath most of the above: trust erodes quietly, in the gaps between people who each did their job. The interesting design question is rarely "was the gate strong enough" — it's "what does a breach actually buy, and how long does the window stay open." That's the thread I'm pulling on in both the writing and the day job.