Stop Wrapping Failed Systems in AI

Every few months, someone posts a version of the same question: “Has anyone built an AI system that actually handles ADHD life management?” The answers are always the same mix of half-built prototypes, abandoned apps, and one person who says “it changed my life” without elaborating.

The thread that caught my attention nails the core problem perfectly: “The systems offered to people who are disorganised, chaotic, and forgetful are way too complex and bureaucratic, and the systems that are possible are usually too simple.”

This isn’t an ADHD problem. It’s a design problem that affects everyone building AI productivity tools.

The pattern is predictable. Someone takes a rigid task management framework — capture, categorize, schedule, review — and bolts an AI layer on top. The AI makes the capture step easier. Maybe it asks clarifying questions. Maybe it sorts things into categories. But the underlying assumption hasn’t changed: the user will show up consistently, review their lists, and follow the system’s logic.

One commenter put it bluntly: “You’ll use it intensely for 3 days then ghost it for a week. The system needs to handle that inconsistency, not assume perfect user behavior.”

That last sentence is the design principle most builders miss. They’re optimizing for the good brain day — the day you have energy, focus, and motivation to engage with a system. But the whole point of building the system is to help on the bad brain days, when opening an app feels like lifting concrete.

The few approaches that seem to survive contact with reality share three traits:

• Capture is zero-friction. Voice dump, text dump, no formatting required. Structure comes later, applied by the system, not demanded from the user.
• The system degrades gracefully. Miss a day? A week? It doesn’t pile up guilt. It picks up where you are, not where you “should” be.
• Tiny defaults beat ambitious schedules. Five minutes of piano beats a planned hour that never happens. The system’s job is to lower the activation energy, not optimize the calendar.

If you’re building AI tools for productivity — for any user, not just neurodivergent ones — ask yourself: does your system work when the user is at their worst? Or does it only shine when they’re already organized enough not to need it?