Toxic productivity with AI is output driven by fear and external validation; healthy productivity is driven by the work itself, the line Brad Stulberg and Steve Magness draw between obsessive and harmonious passion in The Passion Paradox. The tell is what you do with the time AI saves: refill it or keep it. I run a five-question check every Friday against Rize's automatic time data (from $12.99 a month).
The most productive quarter I ever had by output was also the one where I stopped being able to tell whether I wanted any of it. AI made the output cheap. It did not make the wanting any clearer, and I have since watched the same pattern in three founders I coach: more shipped, more written, more decided, and a rising background hum of dread.
The frame that finally made this legible was Brad Stulberg and Steve Magness in The Passion Paradox. They separate obsessive passion, driven by external results, validation and fear, from harmonious passion, driven by the activity itself, and note that the two are indistinguishable from the outside because they come from the same biology: dopamine is released during the chase, not on arrival, and tolerance builds like any drug. That is the line between healthy and toxic productivity, and AI is a tolerance accelerant. It shortens the chase, so you need more chases. Two current numbers make the mechanism visible. Resume Now's workplace survey, reported by ZDNET on September 16, 2026, found that roughly a quarter of workers use AI for research, 14% for emails, 13% for summarising, and just 3% for strategic or planning advice: the uses are overwhelmingly volume uses. And the tool I use to see my own behaviour, Rize, an automatic time tracker whose individual plans run $12.99 to $39.99 a month ($9.99 to $29.99 billed annually), shows me the same thing in my own week: hours saved by AI on Monday reappear as extra tasks by Wednesday, not as white space.
Here is the contrast as I now use it, because the abstract definitions did not change my behaviour and this table did:
| Signal | Healthy (harmonious) with AI | Toxic (obsessive) with AI |
|---|---|---|
| What happens to saved time | Stays empty or goes to one deep task | Refilled with more AI-assisted output within a day |
| Why you open the model | A specific question you already tried to answer | Anxiety, a blank moment, or to feel movement |
| How output feels afterwards | Satisfaction that fades slowly | Relief that fades in minutes, then the next task |
| Metric you actually watch | Whether the work is getting better | Volume: messages, drafts, tasks closed |
| Response to a day without AI | Slower, fine | Withdrawal: irritation, compulsive re-checking |
The Friday check that goes with the table is five questions, answered against Rize's data rather than memory, because memory is the one witness obsessive passion always bribes. Where did the hours AI saved this week actually go, by category? How many times did I open ChatGPT or Claude without a question written down first? Which piece of work got better, not just done? What did I refuse to do? And, the Stulberg and Magness self-distancing move, what would I say to a friend who showed me this week? I write the answers in the third person, which sounds precious and works: "he refilled eleven saved hours with drafting" is harder to argue with than "I was productive." Their mastery mindset supplies the target state, drive from within, focus on the process, be the best at getting better, embrace acute failure, patience, be here now, and the Friday check is just a way to score the week against those six instead of against a task count.
What I actually changed. I moved every AI session behind a written question in my daily note; if I cannot write the question, I do not open the model, and Rize tells me that cut my ChatGPT and Claude window time by roughly a third without cutting output I could name. I keep a one-line "got better" log, because "done" is the toxic metric and "better" is the harmonious one. And I protect the saved hours physically, a walk with no phone, because the tolerance mechanism does not care about my intentions.
The limits are real. Stulberg and Magness are explicit that balance is an illusion and that every genuinely passionate person is unbalanced; the goal is self-awareness about the trade, not a calm week, so a brutal launch month is not automatically toxic. Rize measures windows, not intent, and some of my "AI time" is thinking. The Resume Now numbers are self-reported and a resume company has an interest in workers looking careful. And the hardest case is the one the table cannot settle: a founder in a survival year whose fear is accurate. There, the honest answer is that AI-driven volume may be the right call for a season, provided you know it is a season and have written down when it ends.
Sources: ZDNET, "What workers are really using AI for in 2026 – and what they aren't," September 16, 2026, reporting the Resume Now workplace trust survey (zdnet.com); Resume Now, "Workplace trust in AI" survey report (resume-now.com); Rize, pricing page, individual plan ranges as of September 2026 (rize.io/pricing); Brad Stulberg and Steve Magness, The Passion Paradox (Rodale, 2019), chapters on Vallerand's dualistic model, the mastery mindset and self-distancing.
Related: How to Find Your Passion · Best Self-Improvement Books · How to Make Better Decisions · What University Will Not Teach You