You don't need an AI strategy. You need to know which problem you have.
Most AI projects fail before anyone opens a code editor. Not because of the technology. Because nobody asked the right question first.
Every week I talk to someone who wants to “use AI in their business.” They've read the articles. They've seen the demos. They know it's happening and they don't want to be left behind.
The conversation usually starts the same way: “We need a chatbot” or “We want to automate something” or “Can you help us implement AI?”
My first question is always the same: what problem are you actually trying to solve?
The answer to that question determines everything — what to build, how to measure success, how long it takes, how much it costs, and whether AI is even the right tool at all.
The three types of AI problem
After building MindHunt AI from scratch and working through the Anthropic certification program, I've come to think about AI projects in three categories. Not because someone told me to — because I kept seeing the same patterns in what worked and what didn't.
Type 1: Your people need a smarter tool
This is the most common entry point. Your team does knowledge work — research, writing, analysis, decisions — and Claude (or another model) would make them faster and better at it. No new systems. No custom development. Just people with a better tool.
A law firm deploying Claude across 600 lawyers for research and contract review. A consulting team using it for client deliverables and meeting prep. Each person uses it however they find most useful.
The first time I used Claude seriously was in 2023. I was writing a candidate brief that would normally take me two hours. It took twenty minutes. I remember thinking: this is not a productivity tool. This is a change in what one person can do in a day.
I didn't build anything. I just had a better tool. That was enough — for a while.
Type 2: A specific workflow needs to be automated
There's a defined process somewhere in your business that takes too long, costs too much, or requires too many people. It has clear inputs and clear outputs. Claude can run it end to end, with a human reviewing the result.
This is where the real engineering work lives. And it's where I see the most value created per dollar spent. The key is specificity: a vague “automate our operations” project will fail. A “take this form submission, pull data from these three systems, generate a risk summary, and route it to the right person” project can be built in days.
Type 3: Claude becomes part of your product
This is the most complex and the highest bar. AI embedded directly into something your customers interact with. Not a feature — the engine. It has to work reliably, at scale, in production. Customers will notice when it doesn't.
MindHunt AI is a Type 3 project. Every search, every candidate score, every personalized email — Claude is doing the work. Not as a helper. As the product itself.
In November 2025 I asked “what is a terminal?” on day one of building it. By July 2026 it had 268 registered users, a paid subscription tier, and a security audit behind it. Built on Claude Code and the Anthropic stack.
I tell you this not to brag but because Type 3 is genuinely hard and I want you to know the bar before you decide it's what you need.
Why most AI projects fail
They fail because someone decided on the solution before understanding the problem.
“We need a chatbot” is a solution. “Our support team is answering the same twelve questions five times a day and customers who write on weekends get no response until Monday” is a problem. Those two framings lead to completely different projects — different architecture, different success metrics, different investment.
The chatbot framing leads to a generic interface bolted onto your website that nobody uses. The problem framing leads to an automation that runs the twelve answers automatically, routes the edge cases to a human, and works around the clock.
One of those is a project that succeeds. The other is a case study in why AI is “overhyped.”
The question I ask before every project
Which type of problem is this?
Type 1: your people need Claude. Deploy it. Train them. Measure productivity.
Type 2: you have a specific workflow that's too slow, too expensive, or too manual. Define the inputs, outputs, and success criteria. Then build.
Type 3: you want Claude in your product. Budget accordingly. Set the quality bar high from day one. Plan for iteration.
Sometimes the answer is none of the above — sometimes a rules engine or a simple database query is genuinely the right answer and AI would add cost and complexity without adding value. I'll tell you when that's the case.
That thirty-minute conversation — the one where we figure out which problem you actually have — is the most valuable hour of any engagement. Everything else follows from it.
Not sure which type of problem you have?
That's the most common answer. Let's figure it out together.