Why Most "AI Powered" CRMs Aren't Actually Agentic (And How to Tell)

"AI powered" appears on nearly every vendor's homepage now. Almost none of them mean the same thing by it. Here's the practical checklist we use before believing any of them: the same checklist behind every score on this site.

The bar is lower than the marketing suggests

Somewhere around 2024, "AI powered" became a checkbox rather than a claim. A chat widget that drafts an email. A field that auto suggests a next step. These are useful, genuinely, but they're assistive, not agentic. The distinction matters because it determines whether the system can actually act on your behalf inside a workflow, or whether a human is still the one clicking "send" every time.

Four questions that actually separate the two

Can it write, not just read? Assistive AI reads your data and suggests. Agentic AI can update a record, trigger an email, or change a deal stage on its own: with permissions and an audit trail, not just a recommendation panel.
Is the action logged as a first class event? If an agent updates a field, does the system record who (or what) did it and why, the same way it would for a human user? If not, you have no way to debug or trust agent behaviour at scale.
Does the API expose scoped write access? A read only API or one requiring full account access isn't built for safe agent operation. Look for granular, scoped permissions designed for machine actors, not just human OAuth logins: increasingly, look specifically for a documented MCP server.
Does the pricing model assume a human is doing the work? Seat based pricing is a tell. If a vendor charges per human seat but markets "AI agents doing the work," ask what happens to your bill as agents (not people) do more of the volume.

Why this is a commercial problem, not just a tech one

None of this is really about software features. A CRM that can't support agent initiated actions is a symptom of a company's whole commercial model still being built for the pre agentic era: org structures that wall sales apart from service, comp plans that pay for a moment of signature rather than value delivered over time, pricing that charges for access instead of outcomes. The tool is downstream of all of that.

That's the premise behind the Agentic Commercial Model: a framework for scoring not just your tech stack, but the nine layers of a commercial organisation against how ready each one actually is for this shift. This site applies the same four question test above to every vendor across all nine categories in the full stack.