A friend told me about a decision she’d gone back and forth on where she’d “asked ChatGPT about it,” the way you’d mention checking a weather app. One question, one answer, done.

That same week, I’d set up a rules file system in Cursor that runs an accessibility check and a QA pass automatically every time I finish a feature, on top of a custom MCP server I built for a repetitive process I got tired of doing by hand.

Same technology. Completely different relationship with it.

I didn’t say any of that to her. What she’s describing isn’t wrong, and it isn’t “behind.”

It’s actually the median experience of AI in 2026, and I’m the outlier in that conversation, not her.

She knows prompting is a thing. She knew it back when “prompt engineering” was everywhere online for a few months and then quietly stopped being a phrase people said out loud.

She’s never had a reason to go past the chat window, and nobody’s given her one that wasn’t a headline.

And it got my curiosity spinning on how different people in different professions across different ages use AI 👀

We’re not looking at whether people use AI (almost everyone does now), but how wildly different “using AI” looks depending on your job, your team, and even your age.

I went looking at the actual data, and the range surprised me more than I expected going in.

Explore: A Smart Free Chrome Extension That Upgrades AI Prompts

Why “Using AI” Stopped Meaning One Thing

Adoption isn’t the interesting question anymore. Doctors, lawyers, teachers, developers, your parents—most of them now touch AI in some form.

What differs is how, and researchers have a name for the two ends of that range: automation and augmentation.

Automation is handing AI a task and getting back a finished output. Think: write this email, summarize this document, one shot.

Augmentation is treating AI like a collaborator you push back on, iterate with, and use to stress-test your own thinking before you commit to it.

Anthropic’s Economic Index found augmentation has actually overtaken automation as the more common pattern on Claude.ai, but that split isn’t evenly distributed. It clusters hard by who’s using the tool and how long they’ve been at it.

My friend’s ChatGPT question was pure automation: ask, receive, move on.

My Cursor setup lives at the other end. Rules files, context, and hooks coordinating into something closer to a system than a conversation.

Neither of us is doing it wrong. We’re doing two different activities that happen to share a name.

My friend and I could open the same chat interface right now, and she’d still land on automation, and I’d still land on augmentation.

Note 💬
The difference lives in habits, not software. It comes down to whether you treat the first response as an answer or as a starting point.

What “Using AI” Actually Looks Like By Profession

The profession you’re in shapes which end of that range you end up on almost as much as your own curiosity does. Four professions made this obvious once I lined up the data.

Doctors Went From Skeptical to Daily Users Fast

Physician AI use nearly doubled in a year. 63% of U.S. doctors reported using AI professionally in early 2026, up from 47% the year before.

A lot of that jump is “ambient AI scribes,” tools that listen to a patient visit and draft the clinical notes automatically, freeing the doctor to actually look at the patient instead of typing.

The result is that physician daily AI use has climbed to about 37%, and once a physician adopts a tool, they stick with it.

That’s automation solving a real administrative problem, not augmentation. Nobody’s asking their scribe tool to challenge a diagnosis. It’s only freeing up time (and that’s important).

Lawyers Reach For AI Multiple Times a Day Now

Legal adoption is even further along. Over 90% of surveyed lawyers use at least one AI tool daily, most often for research, document analysis, and contract drafting.

Forty percent use it multiple times per day. Panelists discussing that survey at a Wolters Kluwer webinar noted that the gap isn’t the tools, but that better-trained users get significantly better output from the same tool everyone else already has.

That’s the augmentation gap showing up inside a single profession, not just between professions.

Developers Split Hardest of Anyone

Software engineering has the highest AI adoption of any sector, somewhere around 80% to 95%, and it’s also the widest range once you look past the headline number.

Some of us are still at “autocomplete finishes my function” 😬

Others are running agent orchestration:

• coordinated rule files
• shared context across a whole repo
• custom MCP servers plugging AI directly into an internal workflow instead of a chat window

Tip 💯
For anyone who’s only used AI coding tools for autocomplete or one-off questions, context files and rule files are the next real step up. They let you tell the AI how your project works once, instead of re-explaining it every conversation. Most editors with AI assistants, Cursor included, support this function (with slight implementation specifics).

That 80-to-95% figure covers a developer who occasionally accepts an autocomplete suggestion and someone running multiple coordinated agents against the same repo at once.

It’s closer to the gap between using a calculator and running a full spreadsheet model. Both are technically “using math,” but neither is the same activity, and neither pays off the same way.

The tools sit in the same category on a survey, but aren’t remotely the same job in practice.

Related: You’re Not Just Writing Code, You’re Architecting an Experience

Teachers Are Building With AI While Also Policing It

Teachers use generative AI to build differentiated materials, meaning versions of the same lesson adjusted for different reading levels or learning needs, and to draft instructional content faster.

At the same time, they’re watching students reach for the same tools to skip the thinking an assignment was supposed to build 🙈

Building with AI professionally while managing its downsides in the room, in real time, is a strange spot to work from, and it’s briefly touched on outside education circles. However, I don’t believe we’ve fully grasped the implications it’ll have in the next couple of years.

The Same Split, Sitting Right Next to Me at Work

I built a custom MCP server for one specific, repetitive process I got tired of doing manually.

Someone else on a different team built out an entire library of context files for their testing work, feeding AI the specifics of their test suite so it stops giving generic answers.

And a few teammates are only now bringing AI into their actual coding workflow at all. Mostly for troubleshooting, with nothing like an agent file or a shared rules setup yet.

That last group rarely gets pulled in because I explain the setup to them directly. It happens because they see a shared rules file system in action, or they’re reviewing a pull request I put up on an experimental repo and ask what a particular file in the diff is doing.

The conversation starts because they saw it working, not because someone announced a new best practice in a meeting.

Though it’s a small enough sample size, it does line up with the same pattern the profession data shows, where seeing the deeper version in action moves people further than reading about it ever does.

Proximity to someone already doing the advanced version moves people more than the tool getting easier or the trend getting louder. This is true at the team level, just as it is at the national level.

Age Doesn’t Predict Whether You Use AI Anymore — It Predicts How

The generational data isn’t about adoption but about posture.

1. Gen Z has the highest usage of any generation, with 70% or more using generative AI weekly, leaning toward curiosity and experimentation over caution.
2. Millennials use AI the most for actual work productivity, but SurveyMonkey found they also reach for it for hobbies and even mental health tips. That’s a comfort level that drops off sharply in older groups.
3. Gen X sits in the middle, described in that same research as approaching AI with a “balanced perspective” that recognizes the professional value while staying wary of it personally, largely over privacy and accuracy concerns.
4. Boomers are the most cautious of all, and what usage they do have often isn’t something they’d even call “AI.” Around 35% of boomers report passively using AI features baked into apps they already use. Alexa checking the weather. Google Photos sorting decades of pictures without being asked. (To some extent it’s AI; they simply don’t experience it as AI.)

Nobody in any of these groups is using less AI than they were two years ago.

What varies is how much trust and curiosity each person extends to it, and that variance runs independent of the calendar.

A boomer asking a smart speaker to play music and a Gen Z college student brainstorming an essay outline with a chatbot are both “AI users” by any survey definition.

However, neither is closer to augmentation just because of their birth year. A curious boomer who spends real time with a chatbot ends up further along that range than a Gen Z student who only ever asks it to finish a sentence.

Age predicts exposure and comfort level. It by no means predicts depth.

Do you know which side of that split you’d be on if someone ran the numbers on you specifically?

The Gap That’s Easy to Miss From Inside It

People who’ve used Claude for six months or more get about a 10% higher success rate out of their conversations than newer users.

The interesting part is that researchers couldn’t explain that gap by task, country, or which model someone used. It’s accumulated skill working with the tool itself.

A gap like that doesn’t close on its own as everyone eventually catches up. People already getting more value out of AI have the most reason to keep pushing further into it.

People still asking one question and taking the first answer have less reason to go deeper, and the tool getting more capable every few months doesn’t shrink that distance for them either.

I felt that gap concretely with my friend since the same six months that took my workflow from prompting to full orchestration barely touched her experience at all. Same span of time, wildly different trajectory, and neither of us did anything unusual. We had different reasons to keep going, and that’s really the whole explanation.

This same pattern shows up on teams similarly to how it showed up with her.

Two people can sit three desks apart, carrying the same AI subscription through work, and end up with completely different working relationships with it based on nothing more than how much time one of them spent experimenting early on.

A single training session doesn’t fix that kind of gap, because what separates them isn’t a fact you can hand over. It’s a habit built over months.

What Actually Separates Deep AI Use From Everyday Use

I’ll straight up tell you it’s not intelligence or access. Everyone in this post has access to roughly the same underlying tools (for the most part, location and a few life particulars matter, but we go with average for sanity).

What separates them is whether the first answer counts as final or as a draft.

Augmentation means:

1. pushing back when an answer feels generic
2. asking the AI to defend a choice
3. feeding it more context instead of taking the vague version

Automation means asking once and moving on, which is completely reasonable for a quick weather-app-style question, and pretty limiting for anything you actually need to get right.

Though neither one is a moral failing, only one of them compounds.

So which one were you actually doing the last time you opened a chat window?

My own first real use of AI at work looked a lot like my friend’s ChatGPT question. I posed one question, got one answer, and on to the next task.

The context files, MCP server, and coordinated rule setups all came later, built one frustrated afternoon at a time once a single-shot answer stopped cutting it.

Nobody handed that shift to me. I kept hitting the ceiling of asking once and accepting whatever came back, until building past it got easier than staying stuck at it.

Related: You Need To Work Smarter, Not Harder, With AI

It’s a Wrap

My friend and I aren’t using different technology. We’re using the same technology at wildly different depths, and the reason has nothing to do with intelligence.

It comes down to exposure, curiosity, and whether anyone ever showed us what the next layer down actually looks like.

If you’ve caught yourself in that same moment of quiet disconnect, talking to someone about AI and realizing you’re not even describing the same activity, you’re not alone. It’s normal, and I’d love to hear what that looked like for you.

A coworker? A friend or family? What was the actual gap you noticed?

I’ll leave you with a request to be curious and keep exploring.

See ya.