I spent thirteen years as an engineer and architect before I wrote my first product spec. Long enough to pick up a reflex: when someone hands me a confident framework, I want to know whether they’ve ever had to ship the thing they’re describing.
I’m Rajat Gupta. These days I build product at SAP, on the team behind the code-generation agents in Joule Studio, the build layer of the SAP Business AI Platform. My own patch of that work is the part nobody puts on a launch slide: the evaluation and governance gates that decide whether an agent’s output can be trusted, plus a portable skills framework that lets any coding agent extend the platform. Most of my week goes to a single question. When an agent writes code or takes an action on its own, how do you actually know it did the right thing?
That question is most of why ProductShifu exists.
What this is
ProductShifu (the Shifu’s guide to product kung fu) is where I write for working product managers. You, specifically. The PM who’s read the hype piece on AI agents and now has to decide what goes on next quarter’s roadmap, defend a pricing model, or tell a stakeholder no.
I write about AI products, agent design, pricing, and the parts of the job that never fit on a Jira board. The aim is to show you how to think about a hard problem, not to hand you a checklist. Every post carries at least one real example and, where I can manage it, a moment where I got something wrong. I’ve shipped features nobody used. Those show up here too.
Where I’m coming from
Eighteen-odd years in tech. Thirteen of them as an engineer and architect, a move into product in 2021, an executive MBA from Mannheim and Tongji somewhere in the middle. I’m a principal PM by trade: an individual contributor who sets the standards other PMs build against, not a manager running an org.
If something here earns its place in your Monday standup, use it. If you reckon I’ve got it wrong, even better. Tell me. The best conversations I have tend to start with a reader who disagreed.