The Goldilocks Problem · Cloud Champion
Thanks for coming.
Most enterprise AI failure isn’t capability mismatch. It’s persona mismatch. You came for a comparison of three tools and got an argument about three people instead. The map is below, along with the parts there was never time for.
Chris Landtiser · Microsoft® Most Valuable Professional · M365 Copilot
The setup
Goldilocks never asked which porridge was best
She asked which one was right for her. That detail gets dropped from every retelling, and it happens to be the only part that matters here.
Microsoft shipped three things that look like they belong in the same comparison. They don’t. Copilot Studio wants you to build an agent. Cowork wants you to delegate to one. Scout wants to run on your machine while you’re in a meeting. Three different verbs, aimed at three different people — and almost every evaluation I see starts by asking which of them is best.
There’s a reason the confusion is so widespread, and it isn’t anyone’s fault. The three arrived in roughly inverse order of how many people they suit: the tool built for the rarest persona landed first, and enablement followed the ship order instead of the population. A product catalogue is rarely a well-kept garden path.
The archetypes
Three people, three tools
How many of each you have is the whole rollout question. Almost nobody checks.
Builder — rare
Wants to construct and scale. Hands them a blank canvas and they’re happy: they want to wire the thing up, own the logic, and ship it to other people.
Delegator — the vast majority
Wants the work done. Not a canvas, not a build — the report finished. They aren’t less technical than the Builder, they’re differently motivated.
Operator — a small crowd
Wants to innovate and collaborate. Scripts, shortcuts, a personal toolchain nobody asked for. They automate their own job before anyone offers.
The map
The persona map
Read it by column, not by row. The first is fit, the second is what the tool quietly costs you, and the third is the thing that has to be true before you hand it to anybody.
Builder
Rare
- The tool
- Copilot Studio
- What it assumes about you
- You want to construct, maintain, and govern — not just use.
- What it costs you
- Ownership. You now maintain a product.
- Readiness gate
- A named lifecycle owner before it ships.
Delegator
The vast majority
- The tool
- Copilot Cowork
- What it assumes about you
- You need the work done, not a tool built.
- What it costs you
- Skill. Management nobody taught you.
- Readiness gate
- Delegation training, and a meter someone watches.
Operator
A small crowd
- The tool
- Microsoft Scout
- What it assumes about you
- You understand business process and don’t panic when a terminal command pops up.
- What it costs you
- Readiness. Accountability and a local blast radius.
- Readiness gate
- A governance plan and understanding of the risks.
Not one of these is a better version of another. They were never competing.
So route people to their rung
The enablement play was never “drive adoption.” It’s routing. Put all three in the same all-hands deck and you generate three groups of confused people and one group of frustrated ones.
- Builders Studio, plus a named lifecycle owner.
- Delegators Cowork, plus delegation training.
- Operators Scout, when governance and accountability are locked in.
One caveat the map can’t show: real people are complicated and rarely sit wholly inside a single persona. Plan for cross-tool curiosity rather than pretending it won’t happen.
The delegator
Nobody trained anyone to delegate
We handed the workforce a direct report and no management training. Then we called it an adoption problem.
Writing handoffs
Describing the work well enough that someone else can do it. Every bad prompt is a bad handoff wearing a different name.
Setting acceptance criteria
Deciding what “done” looks like before the work starts, so there is something to judge against.
Reviewing unsupervised work
Assessing output you didn’t watch being produced — and noticing what’s confidently, quietly wrong.
Running feedback loops
Closing the loop on purpose, so the second attempt is better than the first.
Four management skills, and not one of them is an AI skill
Which is the good news, because it means the curriculum already exists. Delegation, expectations, feedback, reviewing work — that’s the first-time-manager course, and it already has a budget and an owner. Whoever runs your manager training is your AI enablement partner. They just don’t know it yet.
Teach prompting
A syntax that changes every six months.
Teach scoping, acceptance criteria, and review
Transfers to every tool that ships after this one.
Readiness
Where each one actually stands
This is the part of the talk with the shortest shelf life, so it’s dated. If you’re reading it long after the stamp below, re-verify before you repeat it.
Copilot Cowork — the production-ready one
Generally available worldwide. The compliance surface is complete: admin controls and spending policy, off by default until someone deliberately turns it on. Of the three tools in the talk, this is the out-of-the-box one. If you take a single line to your architecture review, take that one.
Microsoft Scout — not yet, and that’s fine
Preview, Frontier only, attestation required, and unavailable in government clouds. That’s the whole caveat list, and it’s also the least interesting thing about the product — the release state will change, but the kind of person it’s built for won’t.
The meter is the tell
Cowork bills by consumption rather than by seat, and that changes the rollout maths before any feature does. Per seat, adoption is free once you’ve paid and your only job is to get people using it. Metered, every task has a cost and “get everyone using it” stops being a strategy. A metered tool assumes somebody can tell whether the output was worth it — which loops straight back to the four skills above.
Verified as of 19 August 2026. Preview features ship weekly, so anything here about release status has a shelf life. I date-stamp it — you should too.
What worries me
Persona mismatch doesn’t file complaints
The empty chair
It shows up as silent abandonment — the person who was handed the wrong tool and never told anybody. They don’t appear in your feedback, because the failure never felt like something worth reporting.Nobody opens a ticket
There is no support category for “this tool was built for somebody else.”
They try it twice
It doesn’t fit the shape of their job, so they quietly go back to the old way.
It reaches you as a flat line
Six months later, when somebody asks why utilisation is low.
Slides & further reading
The full deck from the session. If you want more on Cowork specifically, I wrote about working with it alongside Claude earlier this year.
Fit is manufactured
In the story, the third bowl is just right by luck. That’s the part of the fairy tale we quietly kept — waiting for a tool that arrives just right out of the box. Nothing does. Somebody has to own persona routing, readiness gates, and the meter, and in most organisations nobody owns any of the three. That’s a function, not a one-time project. If you’re wondering what to do on Monday: find out who owns it. If the answer is nobody, you’ve found your gap.
A short, occasional email on enterprise AI adoption — what actually happened, not what was promised. No hype.