Pip: Welcome to the Novia Works podcast — where we read the Microsoft Ignite announcements so you don’t have to develop a stress response to the word “preview.”
Mara: Simon has been working through what Ignite 2026 actually adds up to — not just the individual features, but whether there’s a coherent architecture underneath them. Let’s start with what he found.
Microsoft Ignite 2026: Copilot, SharePoint, and the emerging architecture
Mara: The question this post is really asking is whether Ignite 2026 is another pile of feature announcements, or whether something structurally different is taking shape in how Microsoft imagines knowledge work getting done.
Pip: And the answer, it turns out, is the latter — with caveats. The framing is this: “What Microsoft is assembling looks more like a new operating model for knowledge work than a bunch of technologies and features.”
Mara: That’s the pivot. It’s not that Copilot is incrementally better — it’s that the architecture is inverting. Historically you opened an application and called on Copilot for help. Increasingly, Microsoft’s model starts with Copilot and brings applications, data, and capabilities to it.
Pip: Which is a significant shift if it lands. The post is careful not to just hand Microsoft the trophy — Satya Nadella’s claim that Copilot is becoming the “OS for work” gets some deserved scrutiny. Outlook, Teams, and Word aren’t going anywhere, and a lot of the more transformative capabilities are still in preview or Frontier releases.
Mara: The more grounded read is that Copilot is trying to become the principal working surface rather than an assistant attached to individual applications — and whether users actually choose to work that way will be a better measure of success than Microsoft’s terminology.
Pip: The SharePoint section is where things get practically interesting, and it’s the part that tends to get buried under the Copilot headlines.
Mara: Work IQ sits underneath the new architecture and gives agents permission-aware access to organisational content. The new SharePoint Tools expose over 60 operations — lists, libraries, files, approvals, versioning — and crucially distinguish between read access and the ability to act: create, modify, or delete SharePoint resources.
Pip: So the upshot is an agent can now do more than find an outdated policy document — it can potentially update it, adjust its metadata, and trigger an approval process. That closes a real gap.
Mara: On skills and agents: the post argues we’re moving past prompt engineering as the primary customisation layer. Granular, reusable skills, connectors, and MCP tools are the direction — the Salesforce integration ships with more than 38 pre-built sales skills rather than expecting users to construct every interaction themselves.
Pip: There’s a software-engineering principle hiding in there that practitioners will recognise immediately.
Mara: Loosely coupled, reusable functions over monoliths. The same logic applies to agents. On governance, Autopilot gets its own Entra Agent ID so its activity is distinguishable from a human’s — and Agent 365 is expanding to cover agents from OpenAI, Anthropic, Salesforce, Google, and others, positioning itself as an enterprise control plane rather than just a Copilot Studio management tool.
Pip: The post flags one genuine gap in this otherwise increasingly coherent architecture: the relationship between people and agents. “Human in the loop” is the reassuring phrase, but the post argues it’s too simplistic — sometimes a person leads, sometimes an agent leads, sometimes they genuinely collaborate, and sometimes an autonomous agent should just finish the task without interrupting anyone.
Mara: The practical guidance is to prepare the environment before building agents. SharePoint content needs ownership, permissions, metadata, and lifecycle management. Redundant, outdated, and trivial content doesn’t become good organisational knowledge just because Work IQ can retrieve it intelligently — ROT will degrade your agents.
Pip: That’s a useful thing to say out loud before organisations spend six months building agents on top of a SharePoint that no one has maintained since 2019.
Mara: The post closes with five architectural layers that are now taking shape — Copilot as working surface, Work IQ for governed context, skills and connectors for reusable capabilities, agent identity and permissions for managed agency, and Agent 365 for oversight — and one that isn’t yet: managed collaboration between people and agents.
Pip: The architecture is filling in, but the human-agent relationship is still the open question.
Mara: Which may turn out to be the harder problem — and probably the more interesting one to watch. We’ll be back when there’s more to dig into.
