5 Claude Code sessions open. 47 files touched. Shipped? Basically nothing.

vigneshwar udhayakumar — Sat, 25 Apr 2026 13:28:23 +0000

It’s 3:00 AM. staring at a terminal filled with AI-generated code. I’ve touched dozens of files across three different projects. I’m exhausted, my brain is buzzing, and I realize: I haven't actually shipped the one thing I sat down to do.

That’s what ADHD drift looks like in an AI coding workflow.

The Problem: The "Capability Trap"

AI coding tools like Claude Code make the cost of "let me just try this real quick" feel like zero. It isn't. The cost is your day, your focus, and your momentum.

I’d start with a simple task: "Fix the header."
Twenty minutes later, I’m refactoring the entire component library because a button variant felt "off." Then I’m tweaking design tokens. Then I’m rewriting a utility file from three weeks ago.

The worst part isn’t even the lost time—it’s the abandoned context. I’d wake up the next day, stare at those 5 sessions, and have zero memory of where I was.

The Solution: Nilai

I built Nilai because I needed a way to stop the drift. It's a lightweight, open-source tool that acts as a guardrail for your focus.

How it works:

Before you start, you define:

One concrete task
1–5 "done" criteria
A time box

ADHD-First Features:

Tangent Flagging: When you start drifting into a refactor that wasn't on the list, Nilai flags it in the moment.
Parked Thoughts: That "let me fix this too" impulse? Saved as a parked tangent for later.
Context Recovery: When you close a session, it auto-generates a summary. Pick it up days later and see exactly where you left off.
Zero Bloat: No cloud, no tracking, no gamification. Everything lives in your repo as plain files.

I built this for my own late-night sanity, but I realized others might need it too.

Check it out here:
👉 https://github.com/vignu10/nilai

I'd love to hear how you handle "AI rabbit holes" in your workflow!