Build vs Buy: Agents and the Agent Store

What it does

Decides whether to drop in a prebuilt agent from the Agent Store or build your own in Copilot Studio. The Store now holds 70-plus agents from Microsoft, partners and customers, so "we'll just build one" is no longer the automatic answer it was.

Key facts

Agent Store — the in-Copilot marketplace of ready-to-use agents (Microsoft, partner and customer-built). Third-party agents pass a Microsoft validation process for security and compliance.
Prebuilt / partner agents — deploy with no build work; examples span Jira, Monday.com, Miro, ServiceNow and Microsoft's own. You configure and grant access, you don't develop.
Build your own — Copilot Studio (or Agent Builder for simple cases) when the workflow, data or branding is specific to the client. You can also publish your own build to the Store as a distribution channel.
The real axis is differentiation: buy for commodity capability, build where the agent encodes something only this organisation does.

When to use / skip

Buy when the need is generic and someone already sells it — a ServiceNow request agent, a Jira helper, a standard HR-policy assistant. You get validated security, someone else's maintenance, and day-one value; building the same thing is wasted money.

Build when the agent has to touch your Dataverse, follow your process, speak in your brand, or ground on data no vendor knows about. Build also when no prebuilt option covers the workflow, or when the agent is itself the differentiator you're selling. Skip building a commodity agent just to own it — you'll carry the maintenance forever. And skip a prebuilt agent when it can't reach your data or bend to your process; a close-but-not-quite bought agent that everyone works around is worse than nothing.

Configuration decisions

Is this capability commodity or differentiating? Commodity leans buy; differentiating leans build.
Can a prebuilt agent actually reach your data and follow your process, or only approximately?
Who maintains it over three years — a vendor, or your team? Build means you own the upkeep.
What's the total cost including the partner agent's own licensing or consumption, not just the build hours?
Governance: does the prebuilt agent's data handling and validation meet your compliance bar?

Gotchas

"Built-in security validation" covers the agent, not your configuration. You can still overshare data through a bought agent by grounding it carelessly.
Partner agents often carry their own subscription or consumption cost on top of Copilot. "Prebuilt" is not "free".
Build estimates ignore lifecycle: prompt drift, model changes, connector updates, ALM. The build is the cheap part; keeping it working isn't.
A prebuilt agent that's 80% right invites a swarm of workarounds. Either it fits or it doesn't — half-fit is a hidden cost.
Publishing your own build to the Store adds governance and review overhead people don't scope.

Consultant notes

Default to buy for commodity, build for differentiation. Most "let's build it" requests are really commodity needs in disguise — challenge them.
Price builds on a three-year total-cost basis including maintenance, not a one-off delivery fee. That reframes a lot of build-vs-buy calls toward buy.
Where you do build, aim to build the differentiated 20% and buy or reuse the rest — don't rebuild the plumbing.
Vet partner agents on data residency and validation for the client's compliance bar before recommending; "it's in the Store" is not due diligence.

Before quoting a build, search the Agent Store first — half the time someone already made it.