From asking to delegating: four stages of putting AI to work
Most teams use AI at work the same way: type a question, copy the answer, paste it somewhere else. That’s only the first of four stages, and each stage after it changes exactly one thing about how work gets done.
Think about how a capable new person joins a team. On day one they’re sharp but know nothing about your business, so you hand them small, self-contained tasks. Then you give them your playbook. Then you give them logins. Eventually they own parts of the job outright, and the work gets done whether or not you remember to ask.
Using AI at work follows the same arc. Here’s the whole progression at a glance.
Four stages of AI at work, one change each
| Stage | Who starts the work | What it knows | How it reaches your tools |
|---|---|---|---|
| 1Ask | You | General knowledge | Through you |
| 2Brief | You | Your business | Through you |
| 3Equip | You | Your business | Directly |
| 4Delegate | A trigger | Your business | Directly |
AskThe brilliant stranger
You describe what you need, the AI responds, and you move the result to wherever it’s going: an email, a document, a spreadsheet. It’s excellent for first drafts, summaries, explanations, and thinking out loud. For most people, this is all AI at work means.
Two things hold it back. It knows nothing about your business beyond what you type in that moment, so its answers lean generic. And you are the only connection between the AI and everything else you use. Every input and every output passes through your clipboard. The task gets faster, but the workflow stays the same.
Get more out of it: say who the output is for, describe what good looks like, and paste an example when you have one.
BriefThe onboarded hire
Now you give the AI your own material: the price sheet, policies, standard procedures, templates, and a few things you’ve written yourself so it can pick up your tone. Most assistants let you save this as project files, custom instructions, or a shared knowledge base, so you brief it once instead of every time.
The model is the same one you used in stage one. The answers aren’t. Ask it to reply to a quote request and it uses your actual rates, your terms, and your way of putting things, instead of an average of the internet.
EquipThe colleague with logins
On its own, an AI model can’t see your CRM, your spreadsheets, your inbox, or your ticketing system. Most business software, though, has an API: a structured way for other programs to read and change its data. The Model Context Protocol (MCP) is an open standard that gives AI assistants one common way to connect to those systems, usually offered as ready-made connectors.
Once connected, the AI stops waiting for you to fetch things. It can look up a customer, cross-check a spreadsheet against your inbox, update a record, create a task, or draft and send a message. Picture a backlog of overdue accounts that needs to be matched against recent payments and marked up to date. By hand, that’s days of tab-switching and copy-paste. Connected, it’s one well-written instruction and a short wait.
This is the stage where you stop being the courier. You direct; it carries out.
DelegateThe standing process
Stage three still waits for someone to ask. Stage four doesn’t. The work starts from a trigger: a form submission, a new message in a shared inbox, a file landing in a folder, or a time on the calendar.
Workflow tools such as n8n, Zapier, or Make handle the plumbing, noticing the trigger and moving data from one app to the next. The AI sits at the steps that need judgment: reading a message, deciding what kind of request it is, pulling out the details, and drafting a response.
Say a maintenance request arrives through a web form. The workflow logs it in your tracker, the AI reads it, tags the urgency and the trade needed, and drafts an acknowledgment to the resident. A person approves anything marked urgent. Nobody opened a chat window for any of it.
What AI at work buys you at each stage
Asking makes you faster.Briefing makes the output sound like yours.Equipping removes the copy-paste.Delegating removes the request itself.
The real time savings live in the last two. In stage three, you give direction and the AI does the work. In stage four, the work runs on its own and your job becomes supervision. Yet most people never get past stage one, not because the later stages are hard, but because nobody showed them what the next step looks like.
Find your stage of AI at work
Answer these in order and stop at the first “no.”
- Does your AI already know your prices, policies, and tone without you pasting them in?If not, you’re at stage one.
- Can it read or update one of your systems directly?If not, you’re at stage two.
- Does any AI work start without someone typing a request?If not, you’re at stage three. If yes, you’re at stage four.
Move up one stage
- From stage one to twoGather the five documents you find yourself explaining most often and save them where your assistant can use them.
- From stage two to threeConnect one tool with read-only access, such as your calendar or a shared drive, and ask it the questions you’d normally look up yourself.
- From stage three to fourPick one repetitive, low-risk task with a clear trigger. Automate it with an approval step, run it for two weeks, and only then decide whether the checkpoint can go.
Getting real value from AI at work doesn’t mean sprinting to stage four. Each stage pays off on its own, and each one builds the habits the next depends on: clear instructions, good source material, careful permissions, and the discipline to check the work. Take the next step, not all of them at once.
Want more practical guides like this one? Browse other posts tagged AI at work, or head back to the home page.
Discover more from Boateng Kwadwo Agyenim
Subscribe to get the latest posts sent to your email.
