Let the Platform Own the Plumbing
Microsoft Foundry just absorbed the agent scaffolding every team keeps rebuilding, and put it on Azure-hosted models. The lesson is a permission slip: stop rewriting commodity plumbing and spend your engineers on what only you can build.
- #foundry
- #agents
- #azure
Microsoft Foundry just added five capabilities to Claude models hosted on Azure: structured outputs, web search, web fetch, an MCP connector, and tool search. On its own that reads like a feature list. The framing in the announcement is the interesting part, because it names something every team building on models eventually discovers: you ship a strong feature, then spend the next quarter rebuilding the same four pieces of scaffolding.
You know the four. A retry loop that re-prompts because the model returned JSON with a trailing comma. A search-and-scrape service with its own crawler, cache, robots handling, and citation plumbing. A hand-rolled MCP client so the model can reach your ticketing system and three internal APIs. A tool router, because once you wire up three hundred tools the model starts grabbing the wrong one. None of that is your product. All of it is table stakes you rebuilt because the platform did not offer it yet.
This is the quiet, healthy pattern of a maturing platform: the commodity 80 percent gets absorbed, and your differentiation moves up the stack to the part only you can build. Structured outputs constrain generation against a schema so malformed output becomes impossible, not merely rare. Web search returns answers with citations and no crawler to run. The connector and tool search handle reach and selection. The plumbing becomes someone else’s problem to keep running.
The part I would underline for enterprise teams is smaller and more consequential than any single feature: these run hosted on Azure, so you no longer choose between agentic capability and keeping prompts and completions inside your own network, identity, and billing boundary. For a regulated-industry customer, that boundary is often the whole decision.
So the takeaway is a permission slip: stop rebuilding the plumbing. If a capability is undifferentiated scaffolding that every team writes and none of them wants to own, let the platform carry it, and spend your scarce engineers on the domain logic, the evaluations, and the guardrails that make the product yours. The teams that win the agent era will be disciplined about what not to build.