Definition of Done for AI Projects: How to Know When an AI Build Is Finished

If you have an AI project that feels almost finished and keeps dragging on, you need a definition of done for AI projects. A definition of done is the finish line you write at the top of your plan that says, in plain language, this is how we know this thing is complete. Write it before the work starts and the project gets a finish line you can actually cross.

I saw this play out on a Monday creating call with a founder I coach and their team. One of their operators had been moving the company's records out of personal drives and old software and into Notion, so I asked a simple question. How do you know when you're done? The answer came back fast. "Well, it's never done."

That answer is honest, and it's exactly the trap I wanted the team to see.

What Is a Definition of Done for AI Projects?

A definition of done for AI projects is a short list, written at the top of the plan, that names the exact conditions that make the project complete. It answers one question before any work starts. What has to be true for us to call this finished?

Here is how I put it to the team. "One of the keys of any macro level plan in AI is to make sure you've created the plan that says use the steps and when we know we're done. That's the clear definition of done, especially in these big migration projects."

For their migration, I asked them to get specific. Has every person moved their working files into Notion? Is anything they need still sitting in the old software? Can we confirm that each person on the team put everything in? When those answers are yes, the migration is done.

Why Do AI Projects Need a Definition of Done?

Open-ended is the default.

"It's an easy thing to just like have open-ended," I told them. A project with no finish line keeps asking for one more check-in and one more round of asking around to see if anyone still needs something. It stays open until something tells it to close.

I'm direct with my clients about this. I want the migration done today. "I don't want the migration hanging."

A finished foundation is also what lets you build the next thing. I'm living the same challenge right now with the agentic version of my own second brain. What does it need to run an entire digital team, and how do I know when that Notion is done? Being done on my end is what triggers the client version, the blog posts and the wiring. In order to get it to my clients, I have to be done on my end first.

How Do I Write a Definition of Done for AI Projects?

Start at the top of the plan. Before the first task, write the finish line in a few plain sentences. "The answer is you create a plan at the top that says, this is what I know this bad boy is done."

Then split the project in two.

Zero to one is the build. The migration itself. It has an end.

One to two is the ongoing. Using Notion as your single source of truth every day after the move. That part keeps going, and that's a good thing. If you want the bigger picture on that second phase, here's how to use Notion as an AI operating system.

When you separate the two, the open-ended pieces stop blocking the finished ones. On that call there was one data source the team wasn't sure they could ever pull in. Cool. That goes into the ongoing bucket, logged so nobody loses track of it, and the migration still gets to be done.

Once your foundation is finished, keep it finished by giving every repeatable process a home. This is where it pays to organize your SOPs in Notion in the same format every time.

How Do I Know When an AI Migration Is Complete?

You know it's complete when every condition on your list is checked and you can prove it without asking around.

For the team, that means every person has moved their files into Notion and nothing they need is left behind in the old software. Once that's true, they get to stop migrating and focus on organization.

I ran the same play on my own security. I moved every password I own into a new password manager, created a brand new password for every single thing that touches me, my bank accounts and my business, and set up two-factor authentication on all of them. It took four hours on a Friday. I had to create a plan for when I knew I was done. Once it was finished, it became the system I run moving forward.

That's the pattern. Zero to one gets a finish line. One to two becomes the habit.

How Should I Prompt AI for a Big Project?

Give AI the goal and your finish line, then let it figure out the steps.

On that same call we kicked off the first version of a brand new project for the team, live. I opened the prompt with the end goal of what we wanted to create, described who it was for and why it mattered to them, and let Claude go figure out how. "Give the goal," I told the team. "It can go and figure out how in the world we're going to do it."

Then I closed with the qualifiers I put at the end of almost everything. Be world class. Impress us. Simple and easy. Step by step. "The fundamentals work in doing something like this."

Your definition of done is the other half of that prompt. The goal tells AI where you're headed. The finish line tells it what complete looks like. If you're still building the habit of handing AI real context, start with your first win with a new AI tool, then graduate to documentation your AI agents can read.

Your Next Steps

1. Pick the AI project that has been hanging the longest.
2. At the top of its plan, write a few plain conditions that make it done.
3. Split it into zero to one (the build) and one to two (the ongoing).
4. Move every open-ended item into the ongoing list so nothing gets lost.
5. Hand AI the end goal, the finish line and your qualifiers, then let it map the steps.

The Finish Line Is the Plan

Every big AI build I run starts with the same question I asked that team. How do you know when you're done?

Answer it at the top and the rest of the plan has somewhere to go. Finish the zero to one, and you get to spend your energy on the one to two.