When the browser disappears, who is checking the agent?

Last month, Salesforce announced Headless 360. The whole platform (CRM, Agentforce, Data Cloud, Slack, Marketing Cloud) is now exposed as APIs, MCP tools, and CLI commands. Marc Benioff put it directly: "Our API is the UI." In plainer language, agents can build, schedule, and send campaigns without anyone opening a browser.

For CRM and lifecycle teams already deploying Agentforce, Composer, or any of the new generation of campaign-building agents, this is the moment the supervision gap stops being theoretical.

The verification layer no one wrote down

When a marketer logs into Marketing Cloud, builds a campaign, previews it, and clicks send, a lot of informal checking happens. They notice the broken merge tag because it renders as {{ first_name }} in the preview. They catch the wrong segment because the audience count looks off. They see the missing hero image because the screen tells them.

None of that is in any runbook. It just happens, because a human is in the loop.

Headless 360 removes that loop by design. The agent doesn't open the browser. It calls the API. The send goes out, and the only audit trail is whatever the platform writes to its own logs.

This is the gap we wrote about in our earlier post on agentic CRM supervision, but more acute. Composer and Agentforce still let humans see the campaign as it's built. Headless 360 removes the surface entirely.

Salesforce knows. That's why they shipped observability tools.

Headless 360 launched with a meaningful suite of platform-native controls.

Session Tracing logs what an agent did across a workflow. Testing Center lets teams run agent behaviour tests pre-deployment. Custom Scoring Evals score agent output against criteria the team defines. A/B Testing extends to agent variants.

All useful. All structurally limited in the same way.

These tools are Salesforce reporting on what Salesforce did. They tell you the agent invoked the send API, that the API returned a 200, and that the platform considered the campaign delivered. They don't tell you whether the email landed in the customer's inbox, rendered correctly, or arrived at all.

If you've worked in CRM long enough, you know the distance between "sent" and "received correctly" is where most quiet failures live.

The verification surface that actually matters

That's not a Telltide opinion. It's a structural fact about how email delivery works. The ESP can log a send event. The MTA can log handoff to the recipient's server. The recipient's server can log acceptance. None of those events confirm that a real inbox in front of a real person displayed the intended email at the intended time with the intended content.

The only confirmation that lives outside the platform is the inbox itself.

In a pre-agentic world, this was a tolerable blind spot. The volume was lower, the changes were rarer, and a human looked at the campaign before send. In a post-Headless 360 world, where agents are building and sending continuously and no one is looking at the rendered output, that blind spot gets very loud very fast.

What this means for CRM and lifecycle teams

If you're running Agentforce, or considering it, three questions are worth asking your team this quarter.

First, what is the chain of trust between "agent says it sent" and "customer received correctly"? If the answer is "we trust the platform logs," that is the gap.

Second, what is the failure mode if an agent introduces a regression like a wrong subject line, a broken merge tag, or a misrouted segment, and the only check is the agent reviewing its own work? How long until someone notices, and what is the cost in the meantime?

Third, what is the independent observation layer? Not platform logs. Not agent traces. Something that sits where your customer sits.

These questions don't have great answers in the Headless 360 world unless you build them in deliberately. Salesforce gives you visibility into what the agent did. Customer-side verification gives you confirmation that what the agent did actually worked.

Where Telltide sits

When CRM platforms become agent execution layers with no browser required, verification has to come from somewhere the agent isn't. Telltide watches what your customers actually receive, regardless of which ESP, which platform, or which agent built the campaign. Independent monitoring for the headless era.