What Microsoft announced at Ignite
Microsoft ran Ignite in San Francisco and online from 17 to 20 November 2026. Two of its announcements change how organisations build and pay for automation in Copilot Studio.
The first is a change in how workflows get built. The Copilot Managed Runtime team announced on 17 November a public preview of development tools that let makers build Copilot Studio agents and workflows on their own machines. Makers edit them in the developer tool they prefer and publish back to Copilot Studio through a command-line tool. Microsoft says the project files can sit in an existing version control platform such as GitHub, giving a team one source of truth. A Power Platform recap of the event also lists the feature as a public preview.
The second is about cost. Microsoft's Ignite Copilot post describes Agent 365 Cost Management as one place where IT can see and control AI spend. Leaders can set spending policies, limits and alerts there. The Copilot Studio admin documentation explains how the policies work. A resource-based policy can cover individual agents or whole environments, including the apps and workflows deployed in them, with a monthly limit and an optional per-agent limit. Microsoft's own example is a 10,000-credit policy that lets no single agent use more than 1,000 credits. Independent coverage from Beri reports that the spending controls are in preview. It also says a capped agent loses access until credits reset on the first of the next month.
Why the pairing matters
Plenty of New Zealand organisations built their first Power Platform automations quickly. A capable staff member builds flows in the browser, under their own account, often in the default environment. Those flows run fine until their author leaves or someone edits a step without telling anyone. A downstream system may also change, and nobody can say what the flow looked like a week earlier. Copilot Studio workflows now carry more weight than the simple flows of a few years ago. They can call AI models, make decisions on messy input and write into finance, HR and customer systems.
Storing a workflow definition in a repository gives every change an author, a timestamp and a record of exactly what changed. A second person can review an edit before anyone publishes it, and a bad release can be rolled back by republishing the last good version. Software teams have worked this way for decades. Automation built by business users has mostly gone without those controls because the tools didn't support them.
The spending caps answer a separate risk. Workflows with AI steps consume Copilot Credits on every run, so the monthly bill follows volume and complexity, and seat counts don't limit it. A flow that keeps retrying on bad data, or an invoice-processing flow in a heavy month, can spend far more than anyone budgeted. Until now, Copilot Studio credits were managed through capacity settings in the Power Platform admin centre. Those settings had no simple way to say "this workflow may spend this much and no more".
The fine print admins should read
Microsoft's documentation has several details that change how these policies should be set up. Apps and workflows don't appear in the agent picker, so a single workflow can't have its own policy. The only way to cap a workflow is a policy on the environment where the workflow runs. Resource-based policies also skip AI Builder features, which power the prompt nodes inside agent flows. Microsoft advises keeping resources that use AI Builder out of these policies for now.
Scope needs care as well. Choosing "All environments" moves credit management for every environment in the tenant into the Microsoft 365 admin centre and switches off the Power Platform capacity allocations for those environments. Microsoft recommends starting with specific environments. A new policy can take up to 12 minutes to apply. Where policies overlap, their limits aren't added together: the policy with the highest per-agent or per-user limit applies.
The hard stop deserves the most thought. If a cap halts a drafting assistant, staff are mildly inconvenienced. If a cap halts the workflow that posts supplier invoices into your ERP in the last week of the month, someone has to process the backlog by hand. Size each cap against what happens to the business if the workflow stops. Set alert thresholds early enough for a named person to act before the limit is reached.
Microsoft's announcement of the developer tools doesn't list prerequisites, supported platforms or limits. As a public preview, the tooling may still change before general availability. We'd use the tools on new or low-risk workflows before moving anything the business depends on.
What it means for NZ businesses
Environment design now matters for two reasons. Organisations already use environments to keep development work apart from production. Under the new policies, the environment is also the smallest unit that can cap workflow spending. A firm that keeps its accounts payable automation, its HR onboarding flows and its staff experiments in one environment can't give them separate budgets or separate approval rules. Grouping workflows by business impact solves both problems at once.
Git-based workflows also bring automation into normal change management. A change to a production workflow can go through the same review and approval as any other production change, with a record an auditor can follow. That record will matter more as workflows make decisions with AI. Boards and regulators are starting to ask who approved an automated decision and when. We cover that design work in our process and automation practice.
Practical next steps
- List every Copilot Studio agent and workflow in your tenant with its owner, environment, AI steps and any AI Builder prompt nodes.
- Regroup workflows into environments by business impact, keeping finance-critical flows apart from experiments and low-risk assistants.
- Create one resource-based spending policy for a single non-production environment, set alerts at around half and three-quarters of the limit, and watch usage for a full month before widening the scope.
- Keep flows that depend on AI Builder out of the new policies until Microsoft extends coverage to them.
- For each critical workflow, name the person who responds to a cap alert and write down the manual fallback if the workflow stops.
- Try the command-line tools on one new workflow, store it in a private repository and require a second reviewer before every publish.
These controls are easiest to put in place while an organisation has a dozen workflows rather than two hundred. If you're working out how much autonomy to give AI-driven automation and who signs off on it, our AI consultancy team can help set those rules.
Sources
- Microsoft Ignite 2026 Book of News
- Copilot Managed Runtime is now generally available, with additional new preview features (Microsoft Tech Community)
- Copilot at Microsoft Ignite: expanding what's possible with intelligence built for work (Microsoft)
- Manage Copilot Studio costs in the Microsoft 365 admin center (Microsoft Learn)
- Microsoft Adds Hard Spending Caps for Copilot Studio Agents (Beri)
- Ignite 2026 Day 1: What Changes for Power Platform Makers and Admins (Benschmark)
Want to know what this means for your organisation? Our team in Auckland can show you your top three opportunities in a free 30 minute assessment.
Get my free assessment How we help