Building a CustomGPT is the most boring thing in AI. It still puts you in the top 5%.

I'm anti AI-hype, but here's a claim I'll stand behind: the most valuable thing most teams can build this quarter is also the most boring thing in AI right now.

A CustomGPT. Or a Claude Project, or a Copilot agent; the name depends on your platform, the idea is identical. A chat assistant with three things baked in: standing instructions, your documents, and your rules. You set it up once, and from then on anyone on the team opens it and gets answers shaped the way your team needs them, without writing a prompt from scratch.

Nobody posts about this on LinkedIn. The posts are about autonomous agents running entire departments. Meanwhile, in the companies I train, the licences are paid for and most people are still typing "write me an email" into a blank window, getting mediocre results, and quietly concluding AI is overrated.

That gap is why the boring build puts you in the top 5%. Not because it's hard. Because almost nobody has done it.

What it actually is

Take one task your team repeats every week. Feed the assistant the standing context: the templates you use, the tone rules, the format the output must follow, the things it must never do. Then stop prompting from zero forever.

The clients I work with leave sessions with working examples: an assistant that drafts responses in the company's regulatory language, one that turns meeting transcripts into the exact follow-up format a team uses, one that answers new-hire questions from the onboarding documents. None of these will win a demo day. All of them get used the following Monday, which is the only benchmark that matters.

How to pick the task

Three filters. It repeats, at least weekly, because a one-off task never pays back the setup. It's text in, text out: drafting, summarizing, reformatting, answering questions from documents. And someone visibly hates it, because hated tasks create motivated users. If you've read about how we handle skeptics, you already know why that last filter does most of the work.

How long this takes

The first one takes about ninety minutes, most of it spent gathering the documents and writing down rules your team currently keeps in their heads. That step is uncomfortable in a useful way: half the value of building the assistant is discovering that your "standard format" exists in four versions.

The second one takes twenty minutes, because the pattern repeats. By the third, someone on your team builds one without you.

Why I keep saying it's boring

Because the boredom is the point. Impressive things get demoed and abandoned. Boring things get used every day, and daily use is where the skills, the trust, and the actual time savings come from. When people ask me whether they should wait for agents to mature instead, I ask what their team did with the chat window they've had for two years.

Build the boring thing. Put one repeated task inside it this week. The agent debates will still be there when you're done, and you'll understand them better for having shipped something.