Somewhere in your tenant, right now

An employee you have never met is building something you cannot see.

You do not know their name, and this is not a hypothetical. Someone is doing it right now.

The instinct is to lock it down. For the next few minutes, I am going to argue that locking it down is the most dangerous thing you could do, and show you what to do instead.

Meet Maria

What the frontier is already doing

Leading organizations are democratizing access, with governance in place.

This is the innovation unlock. The leaders are not trading speed for control, and they are not funding a giant gatekeeping team to get it. Governed democratization is how scattered usage becomes compounding innovation.

Platform figures: Microsoft FY25 Q3 earnings. Customers: Deutsche Bahn's Microsoft story, AbbVie (Power HUG #10), Boston Scientific (Power HUG #11).

Start here, before anyone builds anything

One Pledge. Every employee.

Before we talk about makers, let's talk about everyone. Shadow AI is the new shadow IT. Anyone can paste customer data into a chatbot, whether or not they ever build a single app or agent. So the first move is not a training path. It is an attestation.

You do not earn your way into the Pledge. You start there. It is Level Zero, and everyone is already on it. The bare minimum that comes with it is a short AI literacy path, not a course catalog.

The one idea that changes everything

Guardrails over gates.

Here is the heresy. You believe more governance means slower makers. The opposite is true, and the fastest maker programs on earth are also the most governed.

Same platform. Opposite posture. And trust runs both ways: the CoE extends trust to the maker by saying yes safely, and the maker trusts the governance process in return. That is the deal. Everything else in this framework is just that one shift, applied again and again.

The core idea, made visible

Progressive enablement. Increasing trust.

One question builds every rung: what is the smallest privilege we can grant safely, and what must a maker prove to earn the next one? Nobody is forced up this ladder. Some makers stay at Bronze forever building small team tools, and that is a complete success. It measures demonstrated trust, not ambition. Maria starts at the bottom rung.

Times are a defensible starting point, not gospel. For a regulated payer, keep Bronze narrower and Silver to Gold reviews stricter if your risk profile calls for it.

Six tracks, one ladder

The checkpoint at each rung.

You do not invent a new ladder per product. One vocabulary, applied across all six Microsoft AI tracks a maker might pick up. Each cell below names the smallest concrete thing a maker does to earn that rung, and links to where that learning lives. The blanks are deliberate, not gaps.

← Swipe the table to see every rung and link →

In the Copilot Studio Agent Academy, Recruit, Operative, and Commander are the progressive lesson tiers, beginner to advanced, which is why they map cleanly to Bronze, Silver, and Gold. Every track sits on the universal Responsible AI Pledge, and the harness choice, standard versus GitHub Copilot, appears at Silver and Gold.

The one rule that keeps this humane

The bar is competency, not mastery.

Here is the discipline that keeps every checkpoint honest. At each rung, the question is never "are you an expert." It is "can you build one real thing, safely, and publish it once."

Trust is earned in the smallest verifiable increment, then the ladder hands you the next one. Set the bar at mastery and nobody climbs. Set it at competency and everybody moves.

The maker's path

A visible path, idea to production.

Before anyone talks about governance, give the maker a map. Every solution, an app or an agent, moves through the same seven steps. This is not a line of approval gates. It shows a maker what they can do on their own, what guardrails apply, and when to bring in help.

Start small, grow on a predictable path. A maker can ship a small, low-consequence tool, then follow the same steps as it gains users, connects to more systems, touches more sensitive data, or takes more autonomous actions. A good idea is never blocked or thrown away. It just picks up the right guardrails and the right help as its stakes rise.

Operating-model concepts here and in the scaling sections that follow are adapted from Productive Edge's AI-native delivery perspective. productiveedge.com

Building your program

Five decisions that turn this into your program.

You already made the first decision, the ladder itself. The other four turn it into a running program. Everything before this was concept. Everything from here is construction.

Building your program, decision 2

Govern by zone, not by app.

You do not govern by policing every build. You route each app, flow, and agent into one of three risk zones, and the zone decides the environment, the controls, and who owns it. This is Microsoft's recommended risk-based model: match governance to risk instead of treating everything the same.

Route the work first, then assign the people, environments, security, ALM, and support. A solution does not start in Red because it uses Power Platform. It starts in the zone that matches its risk, and moves up only when it must.

Before anyone enters Green, every maker clears three gates: manager sign-off, security training, and a data-classification quiz at 100 percent.

Building your program, decision 3

Commit to service levels.

Publishing these numbers is what turns your CoE from a black box into a service people trust. If Maria does not know whether a review takes two days or two months, she builds around you instead of through you.

Give makers one front door

💡 The design rule: every step you add to intake, remove one somewhere else. Boston Scientific built this front door as a Power App, moved compliance to the start, and saved days to weeks on starting a new project. Run approvals with a human in the loop at first, then let the routine ones move to Power Automate.

Example targets. Adjust to your team's capacity, but commit to them in writing.

Building your program, decision 4

Certification is a status, not a trophy.

Most organizations treat a certification like a diploma on the wall, earned once and kept forever. Wrong mental model. It grants real platform access, so it has to behave like an access control: auditable, reversible, time bound. Stop thinking diploma. Start thinking door, a door you can close.

Building your program, decision 5

Govern how agents go live.

The decision: define one extra gate an agent must clear before it is published to real users. Why it matters: an app fails with a broken formula you can see, but an agent's risk shows up the moment it goes live. It can reach people, ground on real data, take real actions, and draw on a shared credit pool, and it fails quietly by hallucinating or oversharing. So before any agent is published, it clears four checks.

The outcome: no agent reaches users on trust alone. Every published agent has a named owner, an approved knowledge boundary, a tested safety profile, and a funded capacity tier, matched to how far it can reach.

Bringing makers along

The easiest on-ramp is a buildathon.

More proof from the field

Two programs worth borrowing from.

Both start with purpose, surround makers with support, and end with solutions other teams can reuse.

💡 The pattern: a purpose to build for, support alongside, and a showcase or catalog so every win becomes something the next maker can reuse.

As presented in Power HUG sessions by Takeda (#18) and Boston Scientific (#20). Figures are as stated by the presenters.

Maria's story, continued

Seventeen days from idea to impact.

Same motivated employee, same app. But now it is governed, visible, and supported instead of hidden and risky. Maria is illustrative, but this exact arc is what a working program produces, in under three weeks of part time effort.

When it scales

When an idea takes off, do not start over.

The work is preserved, not discarded. The business intent, the user learning, the working prototype, and the maker's domain expertise all become inputs to the next stage, instead of being thrown away and rebuilt from zero.

When it scales

Keep what works, add engineering where it counts.

Scaling rarely means scrapping the maker's solution for a brand-new app. Hybrid delivery keeps the parts that already work and adds professional engineering only where it creates real value.

Makers prove the value. Specialists strengthen the solution. Professional engineering enables enterprise scale. The original maker stays on the journey, because that person usually understands the problem best. Not every solution needs this. Escalate only when consequence or technical demands justify it.

When it scales

Monitor everything. Review what materially changes.

Monitor everything automatically. Review only what materially changes. Use the controls your approved platforms already provide. There is no custom portfolio tool to build.

Knowing it is working

The six signals of a healthy program.

A healthy program is rising maker and agent count with flat or falling risk indicators. Rising count with rising incidents is the warning sign that a guardrail quietly became a gate, or the reverse.

It starts with you

The first move is one meeting, not a program.