Craig Stanley
Home / Capabilities / Start here: decision models / Four jobs AI tools do for a decision

Four jobs AI tools do for a decision

A way to sort Microsoft's AI products by the job they do for a decision (find, draft, act or score), so the names matter less.

· 3 min read · Craig Stanley
In short, explained

Computer helpers do four kinds of job. Some find things out. Some help write things. Some do jobs for you. Some help pick between a few choices and say how sure they are.

Microsoft's AI products change names often, so I sort them by the job they do for a decision. Some find and summarise information. Some draft things people edit together. Some take actions. Some score a fixed set of options. Knowing which job you need narrows the choice quickly.

Classify AI capabilities by decision role: retrieve and summarise (Copilot Chat, Microsoft 365 Copilot, Notebooks), draft and share (Pages, Office in Copilot), act (Agent Builder, SharePoint agents, Copilot Studio, Cowork, Foundry agents) and score (decision models such as Microsoft-Decision-1). Choose by role first, then by licence, governance and cost.

Why sort by job

I started this section wanting one page per Microsoft product. The trouble is that the names move. In 2026 alone, Microsoft renamed Microsoft 365 Copilot to Microsoft Copilot in its documentation, merged Chat and Cowork into a new Home experience, and renamed its Scout agent to Autopilot. If I organise my own understanding around product names, it goes out of date every few months.

What doesn't move as fast is the job a tool does for someone making a decision. So I sort by that first.

The four jobs

JobWhat it does for the decisionExamples in Microsoft's range
Find and summariseGathers what you need to know before you chooseCopilot Chat, Microsoft 365 Copilot, Notebooks
Draft and shareProduces something people edit and agree onCopilot Pages, Office in Copilot
ActCarries out steps, sometimes on its ownAgent Builder, agents in SharePoint, Copilot Studio, Cowork, Foundry agents
ScorePicks between fixed options and says how confident it isDecision models, such as Microsoft-Decision-1 in Foundry

Some products do more than one job. Cowork, for example, can research, draft and act within one task. I put each product where it does most of its work and note the rest on its own page.

How the jobs map to risk

The four jobs carry different kinds of risk, which is why I find the split useful beyond tidiness.

When a tool finds and summarises, the main risk is that a person acts on a wrong or partial summary. The person is still the one deciding.

When a tool drafts, the risk is that a draft gets accepted without real review. Again a person decides, but the draft shapes what they see.

When a tool acts, the decision has moved to the system, at least for that step. This is where permissions, limits and logs matter most. The Risk section treats these as decisions in their own right.

When a tool scores, it gives a number that someone or something will compare with a threshold. The risk is in the threshold and in whether the scores are well calibrated. Setting a threshold you can defend covers that.

A worked example

This scenario is illustrative.

A team decides whether to approve unusual supplier invoices. Each job could help in a different way. Microsoft 365 Copilot could summarise the supplier's history from documents the officer can already open, and Copilot Chat could do the same if the officer uploads the files. A Copilot Page could hold a shared note of the case for a second approver. A Copilot Studio agent could chase missing paperwork. A decision model could score each invoice against the options approve, query or reject, and send low-confidence cases to a person.

Seeing the four jobs side by side makes it easier to pick one to start with. In this case I'd start with the summary, because it helps the person without moving the decision away from them.

Where I got stuck

The line between drafting and acting is getting blurry. Microsoft describes Cowork as a way to delegate a task that "runs end-to-end to return a completed result", which is acting, while the output is often a document, which is drafting. I've kept Cowork under act because it takes steps without a person approving each one, but I'm open to changing that.

Sources

Read next

A question to take awayWhich of these do you already pay for and not use?

About me

Craig Stanley

Microsoft AI consultant and technical architect, based in Whitley Bay. Over the last few years I've delivered Microsoft 365 Copilot, Copilot Studio agents, Microsoft Foundry (formerly Azure AI Foundry) work and governance for UK public sector and financial services organisations.

What interests me is the decision underneath the tool: what it costs, what it risks, and whether a small, transparent model can make it better. I write the methods up here and on Substack so anyone can use them.

I write this site to learn in public: explaining each idea simply is how I check I understand it. Why I write this site.

Find me