I talked with Michael Shimeles - Rasmic - about what product engineering looks like when everyone suddenly wants AI in everything.
Michael works at Convex, creates videos about AI and software development, and runs a product studio and AI consultancy. That combination gave us a really practical conversation about the difference between building what sounds cool and building what actually solves a problem.
Start with the problem, not the agent
One thing I appreciated about Michael's perspective is how willing he is to push back. A client may come in asking for subagents, a custom harness, a local model setup, or whatever tool was impressive in a recent video. His job is not to say yes to the spec. His job is to figure out what problem they are actually trying to solve.
Sometimes that means building a custom system. Sometimes it means setting up training. Sometimes it means telling them they just need an existing tool and a few workflows.
That kind of honesty is product engineering. It protects the client, protects the relationship, and keeps the team from spending months shipping something impressive that nobody needed.
Taste is care stacked up
We also talked about taste. Michael had a line I loved: AI has made it much easier to ship apps, but taste, quality, care, and empathy are still missing from a lot of what gets launched.
Taste is not only visual design. It is the whole path from "I have this problem" to "the problem is solved." How quickly can someone get value? How obvious is the flow? How much did you simplify because you understood the person using it?
That is where domain knowledge matters so much. If you used to work in a pharmacy, you have a better shot at building useful software for pharmacists. If you do not have that domain knowledge, you need someone close to the problem who can give you feedback before you spend six months building the wrong thing.
Engineering still matters
Michael closed with a point I think a lot of engineers need to hear right now: syntax is getting cheaper, but engineering is not going away.
Agents are great at writing code. They are not great at engineering unless you bring engineering judgment to the conversation. Architecture, simplicity, first-principles thinking, and the ability to review what was generated are all more valuable when the code comes faster.
His homework was simple: pick one of the classic engineering books you have ignored or dismissed and read it again through the lens of AI-assisted development. The goal is not nostalgia. The goal is to become better at steering the tools we now have.
