Microsoft
Microsoft's own models, Copilot and enterprise agent platform.. Every brief item about Microsoft, newest first, each with its primary source.
Where it stands
This topic starts at Ignite on 11-17, when Microsoft shipped MAI Code 2 and Code 2 Flash and MAI Licensed Weights, with over 50 million agents in Agent 365. On 11-18 Foundry Agent Service put the managed Copilot harness in public preview and made voice agents, computer use and browser automation GA; on 11-19 custom agents gained their own Microsoft 365 accounts with an accountable manager. On 11-20 Foundry Managed Compute went GA: 150+ open models on dedicated GPUs, choosing NVIDIA A100, H100, H200 or AMD MI300X, with deployment templates setting context length, runtime and GPU configuration, plus US and EU Data Zones.
Microsoft is building at the model, runtime and governance layers at once, and has now added a managed layer for open models: its own MAI models, OpenAI models and open models can all be deployed and governed in one Foundry, and the reason to pay is that these layers connect.
Worth watching next: where Managed Compute's hourly pricing breaks even against per-token APIs; how agent accounts are authorized and billed; when the Copilot harness goes GA.
My advice: teams building collaborative agents in Microsoft 365 should prefer dedicated agent identities; steady open-model workloads with data-residency needs can try Managed Compute once daily token volume is worked out; use the harness preview for internal pilots.
Darius · Updated Nov 21, 2026
Timeline 5 items
-
Nov 21 · Sat · 1 item
- Industry devblogs.microsoft.com ↗Microsoft Foundry Managed Compute is generally available: deploy 150+ open models on dedicated GPUs, choosing NVIDIA A100, H100, H200 or AMD MI300X
Take · Renting dedicated GPUs by the hour suits steady traffic and data-residency requirements; for spiky traffic, per-token managed APIs are usually cheaper. I'd work out daily token volume first and migrate only past the break-even point, and try an MI300X template while at it: a second GPU supplier is leverage in itself.
From the Nov 21 brief · item 06 →
-
-
Nov 20 · Fri · 1 item
- Industry devblogs.microsoft.com ↗Microsoft Foundry lets custom agents have their own Microsoft 365 accounts: their own email, calendar, OneDrive and Teams identity, with a manager accountable for them
Take · The most practical item I read this week: the hard part of agents is not capability but who answers when something goes wrong. Give an agent its own account and an accountable manager, and audit, permissions and offboarding reuse the processes you already have for staff. If you're building an enterprise product, separate agent identity from acting on behalf of a user in your data model now, before a customer's security team asks.
From the Nov 20 brief · item 05 →
-
-
Nov 19 · Thu · 1 item
- Tools devblogs.microsoft.com ↗Foundry Agent Service at Ignite: the managed Copilot harness enters public preview, while voice agents, computer use and browser automation become generally available
Take · Alongside Anthropic's self-hosted Managed Agents on 11-12 and AgentCore's new runtime, all three now sell the same thing: you write what the agent knows, and the platform owns the loop, sandbox and egress controls. My advice: start with an egress allowlist, which should be a default on any platform; the harness is still in preview, so do not lock production work to it yet.
From the Nov 19 brief · item 07 →
-
-
Nov 18 · Wed · 2 items
- Model release blogs.microsoft.com ↗Microsoft Ignite: MAI Code 2 and Code 2 Flash coding models, MAI Licensed Weights for trusted customers, and over 50 million agents in Agent 365
Take · MAI Code 2 is in Copilot's model picker, so Copilot teams can compare it on the same tasks today without wiring up another API. I would put Code 2 Flash on PandaClaws' routine back-end fixes for a week and judge it on rework rate and cost per task. The weight licensing is far from a team our size, but for large customers with data residency needs it is a new option worth noting.
From the Nov 18 brief · item 02 → - Safety github.blog ↗GitHub and Microsoft scan six open source projects with AI: thousands of possible issues become 99 private reports, and 55 were fixed by November 15
Take · Thousands down to 99 is the lesson in itself: the value of AI scanning is in filtering and verifying, not volume. When we wire up AI security scanning, I would require every result to come with reproduction steps and a patch, or not come at all. FastMCP is on the list, so teams building MCP servers on it should check recent release notes and make sure they have upgraded.
From the Nov 18 brief · item 05 →
-