The prior-art check: make your AI attack your idea before you build it

If you read the kanban story, you already know the disease: months spent building something the world already had, assisted by an AI that was never asked the rude question. Here's the cure, as a repeatable discipline. It takes minutes. It should hurt a little.

The core move: make the machine switch sides

Your AI defaults to builder mode — you describe, it constructs, it encourages. The prior-art check forces it into skeptic mode with an explicit assignment. Before any real time goes into an idea, open a fresh conversation (not your build thread — that one's already loyal to the idea) and give it something like:

"I'm considering building the following. Your job is to convince me NOT to build it. Search your knowledge for existing products, open-source projects, papers, and established concepts that already do this or most of it. Name names. Tell me what the category is called, who dominates it, and what a newcomer would actually have to add to matter. Do not encourage me. If this is a solved problem, say so bluntly."

Then, and this matters, take the names it gives you and verify them on the live web — search the product names, read the comparison articles, check the GitHub repos. Models can miss recent entrants and occasionally invent products; the check is model-first, web-confirmed.

The three questions the check must answer

• What is this called? If your idea has an established name — kanban, CRM, spaced repetition, reverse auction — the game changes instantly. Named categories have literatures, leaders, and failure post-mortems you can read in an afternoon instead of re-living in a year.
• Who already does it? Not "are there competitors" — there are always competitors — but specifically: which existing thing could a user adopt today to get 80% of what you're planning?
• What's genuinely left? Sometimes the answer is "nothing, drop it" — a great outcome for three minutes' work. More often it's "the category exists, but here's the sliver that's actually unserved." That sliver is your real project, and you just found it months early.

When to run it

At the start, obviously. But also every few weeks mid-build — projects drift, and the feature you added in week six ("what if it also did scheduling?") deserves its own check. And always after any session where you felt the nobody has ever thought of this glow. That glow is lovely and it is precisely the feeling the validation loop manufactures on demand. Treat it as a trigger, not a signal.

Why this is a wellness practice, not just a business one

Because the habit generalizes. The prior-art check is a specific instance of the only real countermeasure to the loop: deliberately routing your ideas through something incentivized to disagree. People who build this reflex for product ideas start applying it to their theories, their plans, their 3 a.m. certainties. It's intellectual hygiene disguised as market research.

When we open, this ships as one of our downloadable best-use skills — a ready-made version of this discipline you can drop into your AI setup, alongside the others we're building for keeping yourself sharp inside heavy AI workflows. Until then: fresh thread, rude assignment, verify on the web. Three minutes. Before the five months.

Touch Grass Wellness publishes practical writing about heavy AI use and mental health. Clinical claims link to their sources; nothing here is a diagnosis or a substitute for assessment by a qualified professional.