How to Find Your First AI Project When You Have No Ideas

Most people who say they have no ideas for an AI project are wrong. They have plenty of ideas. What they lack is a framework for recognizing them.

The raw material for your first AI project is already in your work. It’s hiding in the task you do manually every week that you’ve never stopped to question. It’s the thing you copy from one place and paste into another. It’s the answer you write from scratch every time someone sends you a particular kind of message. It’s the report you build by hand on the last Friday of every month.

You don’t need to generate ideas. You need to notice what’s already there.

This is a practical guide for finding your first AI project: where to look, how to evaluate what you find, and how to choose which one to build first.

Why “I have no ideas” isn’t true

The feeling of having no ideas usually comes from looking in the wrong place. People search for impressive ideas, novel applications, or things that sound like they belong in a demo. They’re looking for something that would make a good LinkedIn post, not something that would save them 45 minutes every Tuesday.

That filter is the problem. It screens out everything worth building.

Your first AI project doesn’t need to be clever. It doesn’t need to scale. It doesn’t need to impress anyone outside of the specific context where it’s useful. It needs to solve a real problem you actually have. That’s the entire bar.

Once you lower the bar to “real problem I actually have,” you realize you’ve been sitting on ideas the whole time.

Where to look for your first project

There are three reliable places to look.

Your recurring tasks. Anything you do more than twice a week that follows a predictable pattern is a candidate. These tasks have the right properties for a first build: they’re well-defined, they happen often enough to justify the effort, and the value is visible immediately. The question to ask is: if a tool did this instead of me, what would it take as input, and what would it produce?

Your frustrations. The tasks you actively dread. The things you put off. The work that never feels done. Frustration is a signal that something is taking more effort than it should. It doesn’t always mean AI can help, but it’s a reliable place to start looking.

Your copy-paste patterns. Whenever you find yourself moving the same information between tools, reformatting something by hand, or recreating the same structure repeatedly, you’re looking at a candidate. These tasks feel like they should already be automated. That feeling is usually right.

Four categories to focus on

Once you start looking, it helps to have categories. Not because every idea fits neatly into one, but because having categories gives you a way to think systematically about what you’re looking at.

Internal tools. Something you or your team uses to do work. A form that routes incoming requests to the right person. A dashboard that consolidates information from multiple places. A generator that produces a first draft of something you make regularly. Internal tools have a short feedback loop: you build it, you use it, you know immediately whether it works.

Automations. A workflow that currently requires a human to trigger it. If there’s a task that starts when something else happens, and you’re the person who makes that connection manually, that’s an automation candidate. The clearest sign: you’ve said “I have to remember to do X every time Y happens.” That’s a workflow waiting to exist.

Content systems. A repeatable process for producing or transforming content. Not a one-time generation task, but a system. You feed it a transcript and it produces a summary in a specific format. You give it raw notes and it produces a structured brief. The key word is repeatable: the same input type always produces the same output type, consistently.

Client-facing tools. Something a customer or prospect interacts with. A calculator that helps them understand their options. An assessment that surfaces a recommendation. A configurator that narrows choices based on their answers. These have a double benefit: they reduce friction for the customer and they generate useful signals about what people need.

Three filters for the right first project

Not every idea is a good first project. Once you have candidates, run them through three filters.

Filter one: have you done it manually more than twice? Once is an edge case. Twice might be a coincidence. Three or more times means it’s real and recurring. Recurring problems are worth solving. One-off problems are not worth building tools for.

Filter two: can you describe the output specifically? Vague problems produce vague tools. “Something to help with content” is not a project. “A tool that takes a sales call transcript and produces a three-bullet summary in under 100 words” is a project. If you can’t describe what the tool would produce in one specific sentence, the problem isn’t defined enough yet.

Filter three: are you the user? Your first project should solve a problem you personally have, not a problem you imagine other people have. You know your own problem from the inside. You understand the edges of it. You’ll know immediately if the output is good or bad. That feedback loop is essential when you’re learning how the build process works. Build for yourself first. Build for others second.

What good candidates look like in practice

Here are examples of first projects that pass all three filters.

A marketing manager who sends a weekly performance summary to their leadership team. They spend 90 minutes every Monday pulling numbers from three different tools and writing the same section headers. A tool that pulls the data and formats the summary cuts that to 15 minutes.

An account manager who writes new business proposals. Every proposal follows the same structure with different specifics. A generator that takes deal notes and produces a first draft cuts the time from two hours to 30 minutes.

A consultant who runs recurring workshops. After every session, they write a follow-up email summarizing the discussion and next steps. A tool that takes their raw notes and produces that email saves an hour per workshop.

None of these are impressive. All of them are real. All of them pass the filters: recurring, specific output, the builder is the user.

The mistake that kills most first projects

The most common mistake is choosing a project that’s too ambitious.

People decide they want to build a customer-facing product, an internal platform, or something that would require integrating six different systems. They start building. They hit friction. They realize it’s more complex than expected. They stop.

The right first project is the simplest possible version of something real. Not the full vision. Not the long-term product. The smallest build that actually works and produces genuine value.

You can always add to something that works. You can’t finish something you abandoned.

How to go from idea to project brief

Once you have a candidate that passes the three filters, write two sentences before you build anything.

Sentence one: what this tool would replace. Be specific about the manual thing, the friction, the time cost.

Sentence two: who benefits when it works. Usually that’s you. Sometimes it’s your team. Write it down.

That’s your project brief. Two sentences. It’s enough to start building, and it’s enough to evaluate whether what you built actually works.

If you can’t write those two sentences clearly, you don’t have a well-defined problem yet. Go back to the filters and narrow it down.

You have more ideas than you think

The problem is almost never a shortage of ideas. It’s a shortage of the right question.

Start with: what am I doing manually, repeatedly, that a tool could do instead? Give yourself ten uninterrupted minutes with that question. Write everything down without filtering.

Then apply the three filters. Recurring. Specific output. You’re the user.

You’ll have a first project by the end of that exercise. Almost everyone does.

If you want a structured way to work through this process, the 5-day AI Build Challenge walks you through it step by step: one idea per day, one short exercise, no tools required. By Day 4 you’ll have a specific project and a two-sentence brief. That’s everything you need to start.