Founders lose momentum because AI has cut the cost of a new start to nearly zero — Lovable, v0 and Bolt turn an idea into a working prototype in an evening, and dopamine hits the start, not the finish. Stulberg and Magness's The Passion Paradox names this obsessive-passion trap. The fix is a two-rule barbell protocol: cap live projects at three, and require a shipped user before you touch the next start.
Six months ago I had eleven half-built AI projects on my laptop. Not eleven ideas — eleven actual prototypes, each with a domain name, each with a working demo, each running. I have shipped exactly two of them. This is not a discipline problem the way it used to be; something structural has changed, and I only figured out what after re-reading a coaching book I first opened five years ago.
The structural change is that AI has driven the cost of starting a new project to something close to zero. Lovable claims 3.9 million users building apps from a text prompt as of mid-2026. Bolt.new gives you a working Next.js app in the browser in under two minutes. Vercel's v0 turns a Figma screenshot into deployable React. The threshold that used to protect founders from their own ideas — "I would have to spend two weekends before I know if this is real" — is gone. Now the threshold is one strong coffee and a good prompt.
Brad Stulberg and Steve Magness named the underlying trap in The Passion Paradox, which I originally read as a book about athletes. The relevant sentence, which I now have on a Post-it above my monitor, is this: "We don't get hooked on the feeling associated with achievement, we get hooked on the feeling associated with the chase." Their argument is that dopamine, which drives passion, is released during pursuit rather than after completion — the same neural mechanism that powers world-class performers also powers slot-machine addicts. In pre-AI founder work, the "chase" was slow enough that dopamine had to spread across weeks of building, and finishing was the only reliable next hit. With AI-native tooling, the chase collapses to a single evening. You can get the dopamine of a "new project" almost every night. That is not a productivity boost. That is what Stulberg and Magness call obsessive passion — motivation by the hit, not the work — and it hollows out finishing.
The other Stulberg-Magness idea that reshaped what I do is the barbell strategy: keep one heavy anchor of stability while you allow measured risk on the other side. The research they cite — including Adam Grant's finding that founders who kept their day jobs while starting companies were 33% less likely to fail — is not really about jobs. It is about the psychological function of a stable anchor: it stops you needing every new thing to work, which is exactly the pressure that makes you abandon the last thing when a shinier prompt appears.
I now run a two-rule barbell protocol. It is short on purpose, because a five-step protocol is itself another new-project dopamine hit.
Rule 1 — Cap live projects at three. One "anchor" (the main business, non-negotiable, gets the first two hours every morning), one "learning bet" (a real project pushing me into a new domain, timeboxed to six weeks), and one "wildcard" (whatever the newest AI tool wants to build with me tonight, hard-capped at four hours a week). If I want a fourth project, I have to ship or formally kill one of the three, in writing, with the date. No exceptions. The formality is the point — Stulberg and Magness's line about "rewriting your story" applies here: I have to explicitly say goodbye to the abandoned project, not just let it rot in ~/projects.
Rule 2 — No new starts until the last one has a real user who is not me. Not a signup form. A person who used the thing in the last seven days without me pinging them. This is stolen from Kenneth Stanley's Why Greatness Cannot Be Planned as much as from Stulberg-Magness — the stepping stones only work if you actually stand on them. AI makes it trivial to build the next stone; the discipline is refusing to touch it until you have proven the current one bears weight.
The honest failure mode of this protocol: it works for me because I have a co-founder who enforces the "no new starts" rule, and it fails in the weeks where I travel alone. Twice in the last quarter I found myself explaining to her that I had "just briefly" started a fourth project on a plane. Both were dead within two weeks. The rule is directionally right but it is not self-enforcing — it needs a peer, a coach, or a public commitment. Stulberg and Magness are clear about this too: self-awareness is the only real defense against passion's inertia, and self-awareness in isolation is genuinely hard for founder-brains. That is the whole reason coaching exists as a profession.
The deeper thing I am watching, and I do not have a clean answer for yet: cheap starts might be making founders individually worse (more scattered, less shipped) while making the ecosystem collectively better (more shots on goal, more surprise winners). Both can be true. What I know is that the tools got faster before the discipline caught up, and any founder telling you they have "AI-native focus" figured out after nine months of Lovable is selling you something. I am running the two-rule protocol, I still break it, and I ship more than I did before I wrote it down. That is honestly the whole claim.
Sources: Brad Stulberg & Steve Magness, The Passion Paradox (Rodale Books, 2019); Lovable AI, official user-count claims (lovable.dev, mid-2026); Bolt.new and Vercel v0 product pages (verified Aug 14, 2026); Adam Grant's research on side-project founders cited in Passion Paradox; Kenneth O. Stanley & Joel Lehman, Why Greatness Cannot Be Planned (Springer, 2015).
Related: How to Find Your Passion · Best Self-Improvement Books · How to Make Better Decisions · AI Coach App — Building It in 8 Hours