Taste, simplicity, and AI-era product judgment with Michael Shimeles

Kent talks with Michael Shimeles - Rasmic - about building products in the AI era without losing the engineering judgment that makes software useful.They cover AI consulting, why clients often ask for the wrong thing, how to push conversations back to the real problem, why simplicity is a product advantage, and why developers are more valuable when they combine technical skill with taste, domain knowledge, and user empathy.(00:00) - Meet Michael Shimeles (01:16) - Product studios and AI consulting (04:35) - Training teams and building custom agents (07:37) - Uncovering the real problem (12:33) - Questions that cut through AI hype (16:47) - Do it the hard way first (19:01) - Finding ideas in products you use (21:03) - Taste, quality, and care (24:10) - Simplicity beyond design (28:04) - Domain knowledge and feedback (31:11) - Developers are expected to own more (35:35) - Do not fall in love with the solution (39:14) - Homework: read an old engineering book Michael Shimeles is a full-stack engineer, DevX engineer at Convex, creator, and the person behind a product studio and AI consultancy. In this conversation, he and Kent talk about what happens when every company suddenly wants agents, automation, and custom AI tools - and why the right answer is sometimes "you do not need AI yet."A major thread in the episode is problem discovery. Michael describes clients who come in excited about subagents, custom harnesses, or the newest tool from a YouTube video, while the real job is to slow down, ask what problem they are trying to solve, and decide whether a simple workflow, an off-the-shelf tool, or a custom product is actually warranted. Kent connects that to The Mom Test and the value of doing things the hard way first so you know where existing solutions fall short.They also dig into taste and simplicity. Michael argues that AI has made it easier to ship software, but not easier to care. Taste comes from shipping, noticing products you wish you had made, learning from feedback, and making the path from a user's problem to a solved problem as short as possible. The episode closes with a practical challenge for engineers using agents: go back to the old engineering books. Syntax is cheaper now, but engineering, architecture, first-principles thinking, and simplicity make you better at directing AI.HomeworkPick one classic engineering book you have ignored or dismissed and read it with AI-assisted development in mind.Notice how the engineering principles from that book change the way you prompt, review, and steer coding agents.Treat agents as code-writing accelerators, not replacements for architecture, simplicity, and first-principles thinking.ResourcesRas Mic - Michael ShimelesConvexFabrikaThe Mom TestClean CodeGuest: Michael ShimelesCompany: ConvexGitHub: @michaelshimeles𝕏: @rasmicHost: Kent C. DoddsWebsite: kentcdodds.com𝕏: @kentcdoddsGitHub: @kentcdoddsYouTube: kentcdodds-plusPodcast: epicproduct.engineer