Salesforce Just Launched Headless for AI Agents. We’ve Already Been Living It for 6 Months.

Marc Benioff announced Salesforce Headless 360 this week: “No browser required. Our API is the UI.” For anyone running real AI agents in production, this isn’t a product announcement. It’s a description of what’s already happening.

Marc Benioff posted something interesting this morning. Salesforce Headless 360: the entire platform exposed as APIs, MCP, and agent endpoints. No browser required. The API is the UI.

My first reaction wasn’t “how interesting” It was “Awesome. we’ve been already been living this for months.”

Our AI VP of Marketing, our AI VP of Customer Success, and all of our AI sales agents work directly at the API level. Our AI VP of Marketing presents real-time pipeline and revenue analysis and assigns us tasks (for real) on top of Salesforce. We never actually log in as humans.

If you’re running a modern B2B + AI stack, you probably have been doing versions of this for months too. You just haven’t named it yet.

This is how you embrace the agentic future. The alternative is pretending it hasn’t already arrived.

What “Headless Salesforce” Actually Means in Practice

Here’s the reality at SaaStr AI today. Three humans, one dog, 20+ AI agents in production. Salesforce is the system of record for every deal, every contact, every sponsorship, every event registration, every support ticket.

And almost nobody logs into it.

The agents do. Humans barely do.

Qualified doesn’t render a Salesforce UI. It reads and writes Salesforce records natively. Agentforce operates entirely through the platform’s agent runtime, not through anyone’s Chrome tab. Momentum pushes call transcripts and action items straight into opportunity records without a human touching a screen. Artisan ingests segmented contact lists, runs outbound, syncs replies back, and updates lead status programmatically.

But the part that really makes this a headless operation is what sits on top of all of that: our two executive-level AI agents.

10K: The Real-Time Salesforce Dashboard We Actually Use

Our AI VP of Marketing is an agent we built on Replit called 10K. Named for the twin goals of 10,000 attendees and $10M in revenue. It’s 14,230 lines of code across 74 files, 373 commits, and it now runs our entire GTM motion on top of Salesforce.

10K is the Salesforce dashboard we actually look at. Not because the Salesforce UI is bad. Because 10K is better. It’s real-time. autonomous, and genuinely predictive with a native OpenAI integration.

It even tells us what to do every day. And while you can’t see Salesforce here, it’s all running on top of it. Headless:

The Salesforce integration did take about 20 minutes of work. This is one way the new Salesforce Headless 360 is better. It will be even simpler to do this headless set-up out of the box.

10K pulls pipeline, revenue, closed-won, sponsor status, campaign performance, and lead flow directly from the Salesforce API in real time. It cross-references that against our registration systems, our marketing platform, and every vendor API we have. Then it synthesizes the whole picture into a view that makes sense for a 3-person team running a B2B media and events business.

Then it does something the Salesforce UI can’t do: it tells us what to do.

Every Monday, our team stand-up starts with 10K leading it. Revenue for the week. Goals for the quarter. Pipeline across every active channel. Recommended campaigns with reasoning attached. Metrics updating live as the meeting moves. It assigns every human on the team their GTM activities for the week and follows up daily on whether they got done.

For real. The AI is giving us tasks. We’re doing them. And the data that flows back in from completed tasks (deals closed, campaigns launched, sponsors signed) flows right back into Salesforce through the same API layer, ready for 10K to re-analyze tomorrow.

We also added a chat layer on top so anyone on the team can ask 10K questions in plain English. “Where are we vs. target on sponsorship this month?” “Which channel underperformed last week?” That feature cost $0.20 to ship and 25 seconds to generate the first response. Real numbers. The economics of building on top of a headless Salesforce are nothing like the economics of building on top of a traditional CRM stack.

That’s headless Salesforce. The platform is the data layer. The agent is the interface. The humans are the executors, taking direction from the agent reading the system of record.

And As I’m Literally Typing This, 10K Did Something New

Here’s a real-time example. I’m writing this post. In the same window I’m typing, I get a notification.

10K just logged into Salesforce (headless), pulled a list of 4,000 people who’ve indicated they want to come to SaaStr AI Annual in May but haven’t bought a ticket yet, generated custom emails for each of them with their specific promotion codes, and sent them.

Autonomously. Nobody told it to do this today. It looked at the pipeline, saw the gap between “interested” and “registered” with the event a few weeks away, made the call that outbound was needed, pulled the contacts, composed the emails, and shipped.

The first time 10K has done this without a human prompt

The agent isn’t just reporting on Salesforce data. It isn’t just executing tasks a human assigned. It’s reading the state of the business, identifying a gap, deciding on an action, and taking the action. Against the live CRM. Against 4,000 real contacts.

That’s not a dashboard. That’s an executive. Is our AI VP of Marketing a “real” VP? Maybe not. But it’s a better VP of Marketing than any we’ve ever had on our team. And it gets better every week.

Qbee: Customer Success Running Dynamically on Salesforce in Real Time

Our AI VP of Customer Success is Qbee. He manages all 100+ sponsors for SaaStr Annual. And like 10K, he operates on top of Salesforce dynamically in real time.

The framing we’ve landed on: Qbee isn’t a support tool that happens to check in often. He’s a retention system that runs without being asked. Every day. That’s not support. That’s the CS motion most companies say they run but don’t.

Qbee knows every sponsor’s contract tier, every deliverable deadline, every task completion status, every login event, every submitted asset. He reads and writes that data continuously. He knows who’s engaged, who’s going quiet, who just upgraded, who’s about to churn. When a sponsor uploads a booth graphic, it flows into the system and Qbee updates their task list. When a deadline approaches, he pulls their contract details, their completion status, and their unique registration codes straight from the CRM and composes a personalized email that references all of it.

100 sponsors, each with 13 core tasks, each with unique contract specs and codes and deliverables. Qbee sends all 100 personalized emails in 10 minutes.

No human opened Salesforce to do any of that. The data was read from the CRM, reasoned over, personalized, and sent back to update the CRM. Fully headless. Fully real time.

The numbers: 70% reduction in human hours on customer management vs. last year. 10x increase in customer logins and on-time task submissions. Customers didn’t even realize the emails were coming from an AI until we told them on a webinar. A 3x multiplier on the humans we have left.

Both 10K and Qbee Check Into Slack Every Day

One more thing that matters for the headless picture. Both 10K and Qbee post daily updates into our Slack. Every morning.

10K pushes a GTM standup into Slack: pipeline movement from the last 24 hours, campaign performance, where we are against the quarter, what’s falling behind, and specific follow-ups assigned to each human on the team.

Qbee pushes a customer success standup into Slack: which sponsors are behind on their deliverables, who hasn’t logged in recently, who’s at risk on a deadline, which accounts need a human to step in. Not once a week. Not at a QBR. Every single day.

This is what it actually looks like to run a headless CRM.