ADHD: When Coding Bots Stop Being Total Mess

The project is literally called ADHD (yeah, the name is intentionally cheeky), and it basically teaches AI agents built on Claude to think like actual creative humans: not a straight line, but a fan of thoughts branching in parallel. Picture an AI that instead of marching forward step-by-step (1, then 2, then stop) explores five different directions at the same time, each with its own mental frame.

How does it work? The agent spins up divergent thought branches, scores them, then becomes a ruthless gardener: axe the ones that look like logical traps and dig deeper into the survivors. It's tree-of-thought with aggressive pruning, if you want the fancy term. Basically, it's like having an internal jury saying "yep, this path makes sense" or "nah, this is a dead end, let's try another lane".

The whole thing runs on Claude's SDK, which means developers can start using it tomorrow. You don't build from scratch; you grab Udit Akhourii's code, throw it into your project, and suddenly your bot can actually handle creative and weird mixed-discipline problems without melting.

Why should you care? Because until now, AI agents were kind of like a kid with a checklist: do step 1, then 2, then 3, no detours allowed. ADHD makes them way better at messy real-world problems where the right answer isn't obvious. Take a design project mixing logic, aesthetics, and function: a basic bot gets stuck, one with ADHD generates five parallel approaches and knows which one wins.

The code is open-source on GitHub, so anyone can grab it, learn from it, remix it. It's the type of quiet drop that in six months you'll find baked into a dozen different tools because it just works.