Ramp runs on Notion: How they built an AI operating system for work

Even after Ramp consolidated its work in Notion, the day-to-day friction did not magically disappear. Meetings still dragged on, key context still lived in tools that were not yet connected, and the new system only worked if people built new habits of delegating tasks.

So leadership made the call: assume the AI game has changed, tear up the old timelines and org charts, and stop treating “reasonable” as fixed. The philosophy was not to replace people, but to multiply their impact.

And Ramp put someone in charge of making that real: Ben Levick, Ramp’s Head of Ops and Internal AI.

“Everybody has a responsibility right now to bring their teams up the curve of AI usage and to get them more comfortable being AI native,” said Ben.

Ramp started treating AI as teammates: always-available helpers that take tasks off people’s plates and expand what each team can deliver. They focused first on simple, high-leverage use cases: meetings, where the most important context is created and the most expensive misalignment begins.

Notion’s AI meeting notes changed how Ramp ran meetings. Conversation context stayed close to the work it impacted, decisions flowed into the right places, and follow-ups were easier to track without someone playing human stenographer or project manager.

Then Ramp widened the funnel from “capturing” to “finding and verifying.” AI search got stronger as connectors improved and scoping got sharper, pulling in what still lived in Slack, GitHub, and Ramp’s own systems into concise, accurate answers.

“Our AI doesn’t just search keywords,” says Cameron Leavenworth, Manager of Corporate IT. “It understands our workspace’s actual structure and relationships.”

The final step was giving everyone a personal Notion Agent to take on time-consuming manual tasks. Like a great assistant that has mastered Notion, it helped people draft first passes, analyze and update databases, route requests, and build workflows.

These were not AI novelties. They became reliable accelerants to everyday work.

In mid-2025, Ramp started experimenting with autonomous Notion Agents, and Ben’s systems mind took off. “The goal now,” he says, “is to build your way out of your workflows. When a single agent can complete your tasks thousands of times—that’s the productivity magic.”

Today, the work does not start from scratch each week. People set up agents once, and those agents keep running across shared, durable workflows the whole company can count on.

There are more than 300 Notion Agents getting things done every day. Here are some of the most impactful:

1. The Product Q&A Oracle—a Slack-connected agent that answers up-to-the-moment questions about Ramp’s products.

2. Sales Feedback Categorizer—maps inbound sales feedback to the roadmap and closes the loop when features ship.

3. Referral Bonus Roy—identifies customers who should receive a referral bonus and routes the work to completion.

4. Enablement Eddie—helps the GTM org find the right customer-facing assets and answer nuanced questions fast.

5. Customer Advocacy Miner—surfaces high-satisfaction customers and compelling use cases across many sources.

6. AI Compass—guides Ramplings to the right AI tool for a given job.

7. RCARoundup—turns incident threads into structured summaries in a Notion database.

8. The Underwriter—a policy-grounded Q&A agent for underwriting.

9. AI Inspector—summarizes AI news into a daily digest.

10. Creative Chris—routes creative requests and nudges work forward.

11. Custom Agent Carrie—helps teammates build better agents based on what Ramp has learned.

Ben describes the change in simple terms: “Agents get created in three minutes between meetings,” then “hours of manual operational work disappear.”

As Ben sees it, the world needs more builders.

His north star is simple: turn everyone into a builder who makes fewer low-stakes decisions, more high-stakes judgments. “It’s important to think about productivity as a combination of speed and efficiency, but also quality and output,” he says.

On the other side of the AI adoption curve, Ramp has found that people not only move faster. They also feel better about the work.

As Ben puts it: “There are two dopamine hits along the way. One is when you crack the problem and automate work you don’t like doing. The other is when, on the other side, things go a little more quiet, you have more time in your day.”

So what can you take from Ramp’s path?

• Do not wait for a perfect button. Use AI enough to learn where it breaks.

• Consolidate the work so AI can navigate the same context and workflows as your team.

• Start small. Nail the quality of each unit of AI work, then automate.

• Keep humans focused on judgment calls.

If you give people the right tools and the right foundation, more of the company becomes builders, and the work gets better—not just faster.