The keynote demo will be the easy part
Ryan Cunningham, Microsoft's corporate vice president for Copilot Studio and Power Platform, published the PPCC 2026 preview on September 3, 2026. The Power Platform Community Conference runs October 27 to 29 at the MGM Grand in Las Vegas, with more than 200 sessions and 24 hands-on workshops, and the post asks attendees to bring one process that matters to their organisation. Its closing advice is four short sentences. Start with one meaningful process. Learn from it. Improve it. Scale what works.
Our read is that the agent on stage will work. Demo agents always work. What decides whether the same agent works in your tenant in February is whether anyone in your organisation can say, without checking, what the process is supposed to do and who is allowed to change it. The post calls the approach process-first. We'd call it owner-first, because a process nobody owns cannot be changed on purpose.
Process-first only means something if someone can say no
The post frames the shift as improving the entire process rather than focusing on one siloed tool, and it uses a procurement example to show the design question. Where is automation enough, where does agent reasoning add value, and where do people need to remain in control. Higher-risk exceptions, in the post's wording, go to the right person for review.
Every one of those is a judgement about one particular process, and the platform team cannot make it. Whether an agent may raise a purchase order under a threshold on its own, or may only draft one, is a call for the person who takes the call from the auditor when a supplier turns out to be the requester's cousin. In the reviews we sit in, that person exists and is almost never on the project. The brief says procurement, or finance. A department cannot say no to a prompt change. A person can.
Our sibling piece on where a human approval step adds real process value follows the same procurement example to the approval itself. The owner question comes first. You cannot decide which exceptions count as higher-risk until someone with the authority to be wrong writes the rule down.
The NFL example is a process that already had one
The customer example in the post is the National Football League, which Microsoft says modernised game-day operations with an AI-powered Game Ops Dashboard built on Copilot Studio and Power Platform. The post says the solution standardised incident reporting and supported operational learning, for what it calls a more consistent, data-driven approach to stadium operations.
Read what standardised incident reporting requires. Somebody at the league wanted every stadium to write up an incident the same way, and had enough standing to make every venue do it. That want existed before any agent did, and the agent serves it. The post doesn't say who that person is or what changed in their game day once the dashboard went live, and that's the question we'd ask if the story gets a session in Las Vegas. A dashboard that standardises reporting without a named person behind the standard would have been ignored by the second week of the season.
The governance sessions assume the owner already exists
The sessions the post recommends include Secure by Design on preparing agents for production and scale, an Under the Hood session on Dataverse, Copilot and Work IQ, one on workflows and agents together, and a full-day workshop on building intelligent business processes with apps, automations, and agents that covers application lifecycle management. All of that is the platform half of the answer. Every control in it protects a decision somebody else has already made.
A pipeline moves a solution from development to production. It does not tell you who read the prompt change that travelled inside it. A data policy says which connectors an environment may use. It does not say whether the accounts payable process should let an agent place a hold on a payment run. When the owner's chair is empty, the platform team fills the gap by guessing, and the guess is nearly always read-only, which is how a sponsor ends up asking why the pilot does so little.
We covered the flow-side version of this gap in where automation owners lose the thread. Agents make it wider, because more of the work now happens while nobody is looking at a screen. Our other piece on the durable process record in Dataverse covers where the owner's answer gets written down once it exists.
Write the owner's name down before you book a workshop
Take the one process the post asks you to bring and add a single line to it. The name of the person who would sign the change request if the agent's behaviour changed. Not the team, and not the steering group. One person, with a phone number. If that line fills in, that's the process to carry into the full-day workshop, and you'll get more from the room because you know what the agent has to replace and who will tell you when it has replaced it badly. Our guide on running a vendor demo on your terms covers how to bring your own messy data and a failure path into that kind of session.
If the line stays blank, spend the workshop afternoon on the agent readiness check instead, and spend the weeks before October 27 finding the owner. Start with whoever last complained about the process in writing, because that complaint is usually the only record of what the process was meant to do. The keynote can hand you the agent. Nobody in Las Vegas can hand you the person who is accountable for the process, and the pilot cannot run without them.
