The solution might be cancelling my AI subscription

Noted developer Simon Willison recently shared a blog post by David Wilson that struck a chord with many AI users. You open Claude to “write a quick script,” and an hour later you have a polished-looking project with tests and docs—yet the original itch was never scratched. This “project bubble” driven by AI coding agents is becoming a new attention sink.

The trap of cheap rewards: why we can’t stop

AI has driven the cost of building close to zero. A project that used to take weeks of design, coding, and debugging can now materialize over a lunch break. But the cognitive capacity to maintain and absorb these projects hasn’t kept pace. The result: we pile up plates at an all-you-can-eat buffet but digest nothing. As one Hacker News commenter quipped, their GitHub is turning into a “graveyard of ideas.” More insidiously, the instant positive feedback—code runs, UI renders—acts as a cheap reward, short-circuiting deliberation and hooking us on an endless loop of starting the next thing instead of finishing anything.

An unexpected twist: when ADHD meets AI

Yet in the same thread, a very different narrative emerged. Multiple people with ADHD reported that AI helps them sustain focus for the first time. “I’m finishing side projects because I can get them working before I get bored.” Another wrote: “I used to need intense EDM to concentrate; now I sit in silence and talk to my agents. I maintain inbox zero and keep up across projects—I feel like I have a support team.”

This seeming paradox reveals how AI tools interact with cognitive styles. An ADHD brain often struggles with task initiation and sustained effort, but AI’s low-latency feedback and continuous interaction provide external scaffolding. It turns building into a high-stimulation, low-friction game, making flow states more accessible. For some, AI acts as a cognitive crutch; for others, it’s an accelerator of distraction.

A deeper trend: the two faces of productivity tools

This points to a broader insight: AI is dividing users not by technical skill, but by cognitive traits. We used to assume that efficiency tools benefit everyone equally; now we see that the same hammer can drive a nail or smash a thumb. Future AI products may need to incorporate “intent management”—for instance, a gentle nudge before spawning yet another project: “Are you sure you want to abandon the last one?” Or a focus mode that limits concurrent tasks.

Practical advice: learning to brake in an AI-powered world

Rather than canceling subscriptions, we need to cultivate a new meta-skill: AI discipline. Try a “think first” rule: define the problem on paper before opening the chat window. Set a weekly cap on new projects, and force yourself to write a maintenance plan for each one—even if it’s just “I will delete this in two weeks.” Or use AI’s metacognitive potential: ask it to periodically check whether your current task aligns with your long-term goals. If you have ADHD, you might intentionally design high-interaction, short-iteration AI workflows, treating the agent as a thinking partner rather than a code machine.

In the end, this debate isn’t about whether AI is good or bad; it’s about how well we know ourselves. When a tool is powerful enough to simulate a whole dev team, the real bottleneck returns to one of humanity’s oldest challenges: managing attention, impulse, and intention. That might just be the most valuable skill in the age of AI.