We work with AI faster than ever. At least it feels that way. Across eight episodes, Healthy AI First tells what studies really say about productivity with AI, where the time we gain disappears, and how to work with coding agents and AI assistants without burning out. For developers and anyone who writes, analyzes and plans with AI at work.
Listen on
Eight episodes, best heard in order. Every episode comes with a companion document listing all studies and sources.
16 experienced developers expect AI to make them almost a quarter faster. Measured, they are slower, and afterward they still believe they were faster. The opening episode of Healthy AI First tells why our gut feeling is a poor measuring instrument when we work with AI, what other studies say against it, and where the series goes over the next 7 episodes.
Listen and read the show notes
When AI writes the code, the work doesn't disappear. It moves: from writing to explaining, reviewing and fitting things together. This episode follows the trail of the time, shows why more code is a poor measure of success, and explains how a team can ship more every day and still make slower progress. Plus a tool tip on reading lead times from Git and GitHub.
Listen and read the show notes
A ticket made of a single sentence, an agent that builds a clean account deletion feature in 20 minutes, and a question from Jonas that reveals what the agent quietly decided along the way. Meanwhile, the AI writes a delivery date into Miriam's client proposal that nobody ever committed to. This episode is about which decisions have to stay with people and which ones the agent can safely make: load-bearing walls, one-way doors, locks instead of notes in the prompt, a lost Mars probe, and how much planning is really needed.
Listen and read the show notes
At 10:40 in the morning, 3 of Jonas's 4 agents reach for him at the same time, and a couple of rooms away, Miriam loses track of her own quarterly plan across 3 AI chats. This episode explains attention residue, takes apart the myth of the "23 minutes", and uses Little's Law to show why more agents with the same review capacity mostly mean longer waits. Plus a tool tip on separate git worktrees, a resumption point for every switch, and a suggestion to try out, for developers and for the office.
Listen and read the show notes
Miriam, the team lead, read a 12-page AI-written concept twice and still can't find the answer when Daniel asks in a meeting what happens if a payment notification arrives twice. This episode is about why a small diagram sometimes clarifies more than many pages of text, when duplicate explanations and voice help or hurt, and why explaining something to the AI yourself does more than just listening to it. With a rubber duck that asks back, a prompt to take away, and an experiment for developers and office work.
Listen and read the show notes
An agent reports "all checks passed", and everything is green. This episode asks what each check really proves, why an agent can bend its own tests, and why a second AI model as reviewer can share the same blind spot. Also: how to protect test files from the agent, mutation testing for changed lines only, risk classes as a team agreement, the ironies of automation, and the explain-back check.
Listen and read the show notes
We rewind Jonas's Tuesday and tell it again: one single change, an appointment reminder with a daylight saving trap, from sizing up the task to exploring, a 4-line contract, Daniel's review and a careful rollout. Alongside it, Miriam's day with a calendar that is only half hers, and what of all this still works for juniors and people on call. The workflow is a proposal to try out, not a proven optimal work cycle.
Listen and read the show notes
Miriam's department head wants a number by the end of the month: how much faster is the team with AI? Instead of one big shiny number, we follow Miriam, the team lead, with Daniel and Jonas for 4 weeks like a diary. Week one: only watching. Weeks two and three: one person works with the new rule, the other with the old one, and they swap after a week. Week four: looking at the results. They measure with a car dashboard of 3 gauges. Along the way, we meet Goodhart's law, a broken Thursday that a summary written by AI hides in the median, and a single page for the budget meeting. And one last time: your number.
Listen and read the show notesHealthy AI First is AI-researched and AI-generated, narrated in Björn's voice, recreated with AI, and reviewed by him. You will find all studies and sources in the show notes of every episode.
Björn is one of the founders of CLN Work. CLN Work helps companies adopt AI in small, verifiable steps: finding where real capacity is, which processes can be automated, and how to use AI interactively. The decisions stay with you.
Book a free 30-minute conversation