A look at where low-code and agentic AI are heading together - and what it means for makers, IT, and business leaders. For years, Power Platform (Power Apps, Power Automate, Power Pages, Power BI, Dataverse) and Copilot Studio lived on somewhat separate tracks: one for building apps and workflows, the other for building conversational bots and agents. Through 2026, Microsoft has been steadily erasing that line. The platforms, the roadmap process, and the underlying AI models are converging into one connected system for building, running, and governing both apps and agents.
Here's what's actually changing, based on Microsoft's own announcements, with the caveat up front that roadmap items are, by definition, forward-looking and subject to change. Anything below without a specific August 2026 citation reflects earlier, more general reporting and should be treated as less certain than the August-dated items.
1. One roadmap instead of several - now with specifics
The clearest sign of convergence: on August 25, 2026, Microsoft confirmed that starting in September 2026, new Dynamics 365, Power Platform, and Dataverse roadmap items will be published through the AI at Work roadmap instead of the twice-yearly release-wave process. A few concrete details worth knowing:
- Existing roadmap content with a public preview or GA date of June 1, 2026 or later will transition to the new system between September and November.
- Release Planner is scheduled to retire by November 15, 2026 (older release plans will remain available for historical reference).
- Microsoft explicitly confirmed there will be no Power Platform 2026 release wave 2 announcement or release plan - this cycle is being skipped entirely in favor of the new model.
- The new roadmap supports filtering, CSV export, RSS subscriptions, feature IDs, and - notably - a Release Communications MCP Server, letting organizations query roadmap data programmatically from their own AI tools rather than only reading it as a webpage.
Why it matters: This isn't just a website reorganization. It changes how IT, architecture, and change-management teams should operate - a twice-yearly readiness review no longer matches how Microsoft is shipping. Microsoft's own August guidance points toward a recurring monthly or quarterly roadmap-review rhythm instead.
2. Copilot Studio is becoming the agent layer that spans everything
At the European Power Platform Conference (EPPC) 2026, Microsoft product leadership described Copilot Studio and Microsoft 365 Copilot as capabilities that add value on top of existing Power Platform investments, rather than replacing them. The framing from Microsoft CVP Ryan Cunningham was explicit: new AI tools are meant to enhance apps, flows, and Dataverse assets that organizations have already built, not force a rebuild.
August 2026 gave that story a concrete, production-ready building block: on August 3, Microsoft made the GitHub Copilot harness in Copilot Studio generally available. Copilot Studio now offers makers three harness choices for building an agent:
- Copilot Chat - the lightweight, conversational default
- Standard - the existing Copilot Studio authoring experience
- GitHub Copilot harness - built for complex, long-horizon, multi-step business processes; it can plan, run agentic reasoning loops, invoke skills and workflows, connect to tools and other agents, and use frontier reasoning models
This is the clearest evidence yet of the "three-tier" agent-building strategy Microsoft previewed earlier in the year (a lighter authoring mode, a full Copilot Studio experience, and deeper integration with more code-first, Azure AI Foundry–style capability) actually shipping. Agents built this way can draw on Power Platform connectors, query Dataverse, and trigger downstream systems from natural-language instructions - letting an agent own a multi-step business process end-to-end.
3. AI is becoming the default interface, not an add-on
Power Platform's release notes describe a shift toward AI as "the primary interface for building, analyzing, and administering" - not a bolted-on assistant. August's feature update backs this up with several concrete, dated capabilities:
-
1
Power Pages Security agent (public preview, Aug 6)
An AI assistant inside the Power Pages Security workspace that can inspect a site's actual configuration, explain findings, and recommend remediations for authentication, authorization, and application-security settings using natural language. Configuration changes still require maker approval, so it's positioned as a copilot for security review, not an autonomous change-maker.
-
2
Power Automate plugin for AI coding agents (Aug 6)
Available in the Power Platform skills marketplace for GitHub Copilot CLI and Claude Code, letting developers create, edit, run, and troubleshoot cloud flows conversationally from a terminal.
-
3
Microsoft Entra Agent ID for Dataverse (public preview, Aug 6)
Dataverse can now associate a dedicated agent user with an agent's Entra identity, giving that non-human actor explicit least-privilege security roles and attributing its data changes to it individually, rather than to a shared service account.
-
4
Data grid modern control (generally available, Aug 6)
The modern, Fluent UI–based canvas-app grid reached GA, with improved search, sorting, column types, and formula-driven content. Microsoft is reinforcing "modern controls" as its strategic UI direction more broadly: modern forms, cards, and icons all received improvements in August, and older Fluent UI "hero" controls (used in custom pages and Teams apps) now have a documented migration path - though classic controls remain supported for now.
4. Convergence with the Microsoft 365 Copilot / "Work IQ" stack
On the Microsoft 365 side, Microsoft has been building out Work IQ, the intelligence/context layer behind Copilot and agents - the component responsible for helping Copilot understand a given user, their role, and their company context. Work IQ underpins several 2026 developments that intersect directly with Power Platform and Copilot Studio, and August brought some of the most concrete convergence evidence yet:
-
A unified Work IQ REST endpoint (Aug 25)
For invoking agents and workflows, replacing multiple legacy Copilot Chat endpoints - a genuine architectural convergence point rather than just a marketing one, since it's now a single integration surface for agent/workflow invocation. -
A unified work/web chat experience with a Work IQ control (Aug 25)
Letting users govern whether their work context participates in a given conversation without leaving it. -
Copilot Notebooks expanded grounding (Aug 25)
Outlook email and Teams meeting transcripts/notes/chats can now be attached as Notebook knowledge sources, alongside existing file-based grounding. PowerPoint can also draw on email content when generating a presentation, and private Viva Engage community/event content can participate in Copilot grounding while respecting existing user permissions. -
Python-backed editing through Copilot in Excel (Aug 25)
Copilot can execute Python for statistics, simulations, and data transformation, writing results back into the workbook. -
Researcher model selection (Aug 25)
Users can choose the specific Researcher model/mode when invoking it from Copilot Chat, reinforcing that Microsoft's AI stack is model-agnostic by design rather than tied to one vendor. This builds on the broader model-choice story from earlier in the year, where Claude models became available in the Microsoft 365 Copilot model selector and inside Copilot Cowork. -
Multi-tenant agent management in the Microsoft 365 admin center (public preview)
Gives IT a consolidated inventory of agents with add/install/block operations and risk/activity insights - though Microsoft specifies that an Agent 365 license is required for the end-user activity and risky-agent insight views specifically.
5. The licensing model is converging too - and this is where it gets consequential
Perhaps the least glamorous but most operationally important part of this story is that Copilot Credits are becoming the single common currency for AI consumption across Power Platform, Copilot Studio, and AI Builder. AI Builder is winding down its own legacy credit system: new customers can no longer buy AI Builder capacity add-ons except for renewal or true-up on existing contracts, trials have been discontinued, and both agents and agent flows now consume Copilot Credits directly, with a firm deadline attached. The 5,000 AI Builder credits bundled with the per-user attended-RPA entitlement disappear on November 1, 2026, so anyone relying on that seeded capacity should start modeling their Copilot Credit needs now rather than later.
Licensing is loosening in one other place at the same time: Power Automate Flow groups, introduced August 6, let a single Process license (250,000 actions a day) cover up to 25 solution-aware cloud flows instead of just one, a genuine cost lever for organizations running many moderate-volume flows. What hasn't changed is the separation between AI Builder and Power Automate premium licensing. A flow using an AI Builder action still needs the underlying Power Automate premium entitlement on top of whatever Copilot Credits it consumes, and that's worth building into any total-cost estimate from the start.
Why it matters: As agent-building shifts toward harnesses and models that consume credits per-task (see the GitHub Copilot harness above), FinOps and licensing management are becoming inseparable from solution architecture - not a separate conversation that happens after a project ships.
6. What this convergence means in practice
Putting the pieces together, a few practical implications stand out. There will be fewer places to check for roadmap news, but more of it, more often: with release wave 2 explicitly skipped for 2026 and Release Planner retiring by November 15, the twice-yearly readiness cycle is gone, and Microsoft's own architectural guidance points toward monthly or quarterly roadmap reviews instead, with the new Release Communications MCP Server built specifically to make that kind of continuous, programmatic tracking feasible. Governance is becoming agent-aware, and identity is the foundation of that shift. Entra Agent ID for Dataverse and multi-tenant agent management in the Microsoft 365 admin center both point the same direction: agents are being treated as individually identifiable, least-privilege security principals rather than generic service accounts.
None of this comes at the expense of what's already built. Existing Power Platform investments are being preserved, not replaced, and Microsoft's messaging has been consistent on this point; classic Power Apps controls remaining supported alongside the modern-control push is a concrete example of that in practice. What has genuinely shifted is where cost sits in the process. Cost and licensing decisions now belong inside solution design, not after it: harness choice in Copilot Studio, AI Builder's shift to Copilot Credits, and Flow groups all change how an agent or automation's true cost gets calculated, and this is arguably the most concrete, actionable convergence signal to come out of August.
