meta

The 6 stages of using AI — where are you right now?

2026-06-18 · Ascendy Engineering

TL;DR

Split using AI into six stages: not using it → asking → delegating → handing off the work → building → making AI do the work. Up to Stage 3 (one-off delegation) you’re using a tool; from Stage 4 (re-architecting the workflow) you’re changing how you work. That takes a shift in mindset, not skill — and most people stop right there.

👉 Jump to the diagnostic → — if you’d rather find your stage before reading the whole thing.

Background

AI is read as moving from “tool” to “partner.” But the depth of that relationship differs from person to person.

Some people only ask instead of searching; others build services without writing code. Same “using AI,” entirely different realities. We needed a frame that makes the difference visible.

If the ability to use AI is roughly linear, then the spots where people get stuck should follow a pattern too. Name that spot and describe its symptoms, and from wherever you stand the next step becomes visible.

This post offers that frame.

The six stages

Stage 1: Not using it

You haven’t brought AI into your daily life. This includes people who haven’t felt the need yet, or who tried it, didn’t like the result, and drifted away.

There’s nothing wrong with this stage. It’s just where the biggest jump still lies ahead.

Stage 2: Asking

You start asking AI instead of searching. You use it as a “smarter search box.”

“How do I write this regex in Python?” — you ask, an answer comes. You eyeball whether it’s right, then use it.

AI is the tool, you’re the driver. Verifying that it did what you asked is on you.

Stage 3: Delegating tasks

You start handing off one-off tasks. “Can you draft a reply to this email?” “Can you summarize this paper in three lines?”

The result comes back and you use it as-is or tweak it lightly. Felt productivity jumps.

This is where the most people stop. Productivity is already up, they’re satisfied, and they don’t feel a reason to go deeper.

Stage 4: Handing off the work

Up to Stage 3 it was “I work, and when I get stuck I ask AI.” Stage 4 is the opposite: “I rebuild the work itself so AI does it.”

For example:

• If you used to pull a monthly report by hand, now an AI script pulls it automatically and emails it to you.
• If you organize customer feedback, an automation sorts incoming feedback by category and you only read the summary.

It’s not skill that has to change — it’s mindset. Not “which part of this work do I hand to AI?” but “how do I re-architect the flow of this work itself?”

A zone appears where you delegate without checking. You have to trust the AI.

This is the hardest wall. It’s where people stop most, and even people who are good with AI sometimes never cross it.

Stage 5: Building

You go beyond consuming AI to producing with it. You build tools or services yourself.

Developers do it in code. “Can you write a script that automates this?” — and working code comes back.

Non-developers do it with no-code and vibe-coding. They wire services together with Zapier, Make, or a script written by Claude. “Could I build this as an automation tool?” becomes an actual tool.

Stage 6: Making AI do the work

You build systems where AI works on its own.

Simply, that’s automation (pull a report every Monday at 8am and post it to Slack). Elaborately, it’s an organization of agents (a system where multiple AIs split roles and pass work back and forth).

This stage is wide too. Even within Stage 6, there are layers.

The hardest stretch: 3 → 4

Why most people stop here

1) Anxiety about verification

Up to Stage 3 you receive the output, check it with your eyes, and use it. Right or wrong, you see it. That’s reassuring.

Stage 4 is different. A moment appears where you say “AI handles this work; I just look at the result.” You don’t fully verify. And accepting that is hard. The “is this really okay?” anxiety lingers.

2) The satisfaction problem

Stage 3 alone already lifts productivity a lot. “Honestly, isn’t this enough?” — you think, and stop. The motivation to go deeper is weak.

3) Lack of metacognition

Decomposing your own work into “AI’s part / my part” isn’t easy. To judge “how far can I automate?” or “is this delegatable?” you first have to see the flow of your own work — but most people just do it.

What changes

Crossing into Stage 4, productivity doesn’t rise in a straight line. At first it may even drop.

Because re-architecting the workflow takes time. You think through “how do I automate this?”, set up tools, test… and meanwhile the existing work can get slower.

But once it’s set up? From then on you barely touch it. The cost of repetition approaches zero.

Diagnostic: where are you right now?

Use the tool below to find your position. You don’t pick a stage directly — it back-calculates from 5 everyday behaviors.

How to move to the next stage

→ To Stage 2: just ask once

Start with things a search doesn’t answer well. Instead of searching “why won’t this setting work?”, ask AI.

→ To Stage 3: start with one-off tasks

Email drafts, meeting notes, simple translation. Delegate the tasks you’re confident about first. Check the result and use it.

→ To Stage 4: look at the flow of your work

List the things you repeated over the past month. “Which of these could AI do?” “How could I reshape this flow so my hands touch it less?”

It doesn’t have to be perfect at first. Automate small things. As that experience stacks up, you’ll see the larger flows.

→ To Stage 5: build something

Non-developers: start with Zapier, Make, or simple Google Sheets automation. Developers: start with scripts.

The experience of turning an idea into an actual tool is what matters.

→ To Stage 6: think in systems

Wire several automations together and you get a system. Beyond simple automation — the mindset of making AIs pass work to each other.

Related reading

• “Beyond the six stages — the AI organization and its blind spot”: what do operators past these stages worry about?

Authorship & citation: Written by Ascendy Engineering; quotable with attribution. Found something wrong? Let us know via a GitHub issue.

Tags: ai, productivity, workflow