Blog · AI maturity
Copilot is the start. Here is the next level.
How to get more out of Copilot today, why it is not the same as Claude, Codex or an agentic platform, and how your organization moves up one wave.
- Wave 1 Copilot in your apps · You are here
- Wave 2 Agents: Claude, Codex
- Wave 3 Agentic platform + Enterprise Brain
- Most people meet AI in Copilot. Early movers are changing how their industry works.
Where most people meet AI
For most people at work, AI has a name: Copilot. It sits in Outlook, Teams, Word and Excel, it is already licensed, and it is approved by IT. That makes it the right place to start, and for many teams it is the only AI they are allowed to use.
But most of the value of AI is not in wave 1. Copilot helps one person in one app at a time. The organizations that pull ahead let agents do whole pieces of work, and then build the platform that lets those agents run inside their processes. This post does two things: it shows how to get more out of Copilot today, and it shows what the next level looks like, so you know when you have reached the limit.
Wave 1 · Copilot
Four ways to get more out of Copilot today
None of these need a new tool or a new license, just a different way of asking.
Tip 1
Ask for paths, not commands.
“Put this into OneNote” gets you exactly that, and nothing better. Ask instead which way fits, and let the assistant lay out the options. You stay the one who decides, and you often hear about an option you hadn’t thought of.
Copilot
In your Office apps
- A Loop page in the team channel: everyone edits live in the meeting; weak at finding old decisions.
- A list in the team’s SharePoint site: one row per decision with owner and date; searchable and filterable.
- A OneNote section per quarter: quick to start; hard to keep consistent across three teams.
For three teams that need to look things up later, the list is the safest choice. Shall I draft the columns?
Tip 2
Give it the whole folder.
One document gives you a summary. Twenty documents give you patterns: put the retro notes, status reports or customer mails of a quarter into one folder and ask what keeps coming back. That is where AI sees what no single reader sees.
Copilot
In your Office apps
Three themes come up in five of the six retros:
- Late requirements: stories change after sprint start.
- Test environment: waiting for a free environment, mentioned in every retro since sprint 3.
- Unclear ownership of bugs between two teams.
The test environment was raised five times but never got an action item with an owner.
Tip 3
Sort your inbox by who is waiting on you.
Most mails that need you never say so. Ask the assistant to split your week into direct asks (“can you send…”) and inferred asks: the things people expect from you without writing them down.
Copilot
In your Office apps
Direct asks (4):
- Budget sign-off for the training plan, requested Tuesday, due Friday.
- Feedback on the vendor shortlist.
- Two meeting requests without a reply yet.
Inferred asks (2): the release thread assumes you will brief the steering group, and the onboarding mail expects you to name a buddy for the new colleague. Nobody asked you outright.
Tip 4
Prepare the leadership call two days early.
Before a leadership review, ask for a risk view of everything that is blocked. Two days ahead, not two hours: then you still have time to unblock what you can, and you walk in with a plan for the rest.
Copilot
In your Office apps
Two of the seven blocked items touch the release:
- Interface to billing: waiting on a test account for nine days; one mail to the owning team could unblock it.
- Data migration script: blocked by a missing approval; it needs a decision in the call.
The other five can wait until the next sprint without moving the date.
The chats are illustrative examples written for this post. Real answers depend on your data and on what you can open.
Why it is not the same
Copilot, Claude and Codex, and an agentic platform
They all have a chat box, so they look alike. The difference is who does the work. With Copilot you drive every step. With an agent like Claude or Codex you set the goal and the agent does the steps. On an agentic platform, agents do the work inside your processes, with your company’s knowledge, and people approve where it matters.
| Copilot in your apps | Agents: Claude, Codex | Agentic platform | |
|---|---|---|---|
| What it does | Answers and drafts inside the app you are in | Plans and carries out multi-step work: reads, writes, runs tools, checks the result | Runs agents as part of your processes, around the clock |
| Who drives each step | You, prompt by prompt | The agent, towards a goal you set | The process; people approve where it matters |
| What it knows | The document, mail or chat in front of it, and what you can open | The files, repository and tools you hand it | Your company’s knowledge, through the Enterprise Brain, within each user’s permissions |
| What it is good for | Writing, summarizing, finding | Building, analyzing, automating a task end to end | Changing how a process works: tickets, customers, operations |
Copilot is adding agent features, too. The line is less about the brand than about the way of working: as long as you prompt every step yourself, you are in wave 1.
Waves 2 and 3 · Agents
What the next level looks like
You stop asking for text and start handing over work. The agent reads the ticket, asks for the approval it needs, acts, and writes down what it did.
- An access-request agent: read the ticket, request approval, act.
- A help agent that opens a ticket by itself when it cannot answer.
- One agent per customer, with the whole history of the relationship.
- A public website chat that answers only from your published documents.
Your agent
Connected · Service desk
Three requests are open. I read each ticket and checked it against the access policy:
- Read access to the reporting space: standard role, approved by the space owner. Granted, ticket closed.
- Admin rights on the build server: needs the platform lead’s approval. I asked for it in the ticket and will act once it is approved.
- Access to the finance share: outside policy for this role. I declined and explained why in the ticket.
Every step is in the ticket history.
How to move up one level
- Collect what works. A one-pager per team with the prompts that worked and their results. It spreads good habits faster than any training.
- Show managers real agents. One workshop where leaders see an agent do a piece of their own work changes the conversation from “AI writes emails” to “what could we hand over?”
- Start with your most modern team. Let one team build with agents (vibe coding, agentic coding) and make it the example for the rest.
- Tidy up the knowledge. Agents are only as good as what they can find: a maintained wiki, a SharePoint that is not a junk drawer, clear owners. That is the groundwork for the Enterprise Brain.