You're the one at work who gets AI

You know who you are.

You're the one people message when the chatbot hands them something useless. You've got a doc somewhere full of prompts that actually work, and someone in a meeting has said, "Ask them. They're good with AI."

Nobody hired you for this, and it isn't in your job description. You got curious early, and now people walk over to your desk.

If that's not you, it's probably someone you work with. Either way, this is about what that person is really doing.

What you're actually doing

Think about the last time someone asked you for help.

Say a teammate got a bad sales email out of the AI. You fix the prompt, then ask them what a good one sounds like, and what the best rep on the team does that nobody else does.

Then you write that down. Plainly, in a form the model can follow and the next person can reuse.

You're getting what the experts know out of their heads and into a form the team and a model can run. The rules they follow without thinking, and the taste they can't explain until you ask the right question.

You might call that a playbook or a style guide. Here it's called building frameworks.

A framework is what good looks like, written down so it can run again. A plan runs once. A framework runs every time the job comes up.

You've probably been building them for a while already.

Why it makes you hard to replace

Models handle more of the doing every year. First drafts, summaries, code, slide decks. That part keeps getting faster and cheaper.

But a model has no way to see what good looks like at your company until somebody writes it down. It can write ten versions of a proposal before lunch. Nothing in it says which one your biggest client would sign, or where your legal team draws the line.

That knowledge lives in people's heads. Usually a few people, and usually the busiest ones, who never have time to write it down.

Someone has to get it out and say what good looks like and what has to hold, no matter who's doing the work or which model is doing it.

That's you. And it's never finished. The experts keep learning, and the models keep changing, so the written version needs someone who can keep pulling the next piece out. That's the part that's hard to replace.

Remember prompt engineer?

A job title can show up fast. That doesn't mean it lasts.

In January 2023, Anthropic's job board listed a role called Prompt Engineer and Librarian. The salary range was $250,000 to $335,000. The work, as the posting described it, was to find the best ways to prompt the model, then document them so other people could learn them.

That's close to the work you're already doing.

Two years later, Microsoft surveyed 31,000 workers in 31 countries for its 2025 Work Trend Index. The report lists the top ten AI roles leaders said they were considering hiring for. Prompt engineer isn't on that list. The most common workforce strategy leaders named was training the people they already have in AI skills, at 47 percent.

Around the same time, two researchers looked at more than 20,000 LinkedIn job postings. They found 72 prompt engineer jobs. That's under half a percent.

Read together, those numbers say the title hadn't become common by 2025.

So if you're waiting for "framework builder" to show up as a job title before you take this seriously, you could be waiting a long time. And if it does show up, it could fade the same way.

So is this a job?

Not yet, not as a title.

It becomes a profession when companies hire for it by name, in real numbers. Prompt engineer got hired by name. Two years on, it was still under half a percent of postings. Framework building could go either way.

Until then, you can do this work in any role: sales, operations, design, finance, support. You don't need a new title to start, or anyone's permission.

And it's the work that makes you the one people keep coming back to.

Your first move

Start with the things you'd never compromise on, in your own work, on any project. The stuff you'd still fight for when the deadline says skip it.

Write them down, and you'll find two kinds of material. First, the rules of your craft that you can look up, like a style guide or a coding standard. Second, your own taste, the calls you make because you know they're right even when no rule says so, like cutting the slide that's correct but too busy. That list is your first framework.

Sources: the Anthropic job posting as archived by the Wayback Machine on January 18, 2023 (the live posting is gone); Microsoft, 2025 Work Trend Index, survey fielded February 6 to March 24, 2025; An Vu and Jonas Oppenlaender, "Prompt Engineer: Analyzing Hard and Soft Skill Requirements in the AI Job Market," arXiv 2506.00058, postings collected April 7, 2025.