Make with Notion 2025: Simplicity at Scale: Inside Ramp’s AI Operating System (Ben Levick)

Ramp’s AI operating system pitch boils down to a single operational bet: teams get dramatically more productive when AI is paired with a disciplined way of “building” work—turning vague requests into precise prompts, grounding answers in single-source knowledge, and then wiring outputs into repeatable workflows. The payoff is speed plus quality at the same time, which matters because Ramp’s business goal is to save customers time and money, and because faster, better decisions let teams move faster without sacrificing accuracy.

Ramp’s context is finance and operations, where wasted effort shows up as slow approvals, messy expense work, and constant back-and-forth over where money should go. Ramp frames its mission as becoming the most productive company in the world, backed by years of interest in AI agents and a belief that agentic behavior can shift humans away from busywork toward higher-stakes judgment calls. The company cites large-scale impact—serving 45,000 customers and saving nearly 30 million hours and $10 billion in wasted time and money—while positioning Notion as a partner in that productivity loop.

The core “how” centers on making each team member faster and better. Automation handles manual tasks and delivers information instantly, but Ramp emphasizes a second lever: AI’s deep research capability can improve decision-making by surfacing what matters from large context. That combination—instant research enabling bigger decisions—creates the conditions for team velocity.

Ramp then argues that most organizations stall because they fall into three passive mindsets: “doomers” waiting for AI to replace jobs, “zoomers” expecting one-click tools to do everything without understanding what must be replaced, and “boomers” delaying adoption until the “right way” emerges. The alternative is a fourth mode: “builders.” Instead of treating AI as a black box, builders practice changing the work itself—learning to define tasks, supply the right context, and embed outputs into the tools where work happens.

To operationalize building, Ramp describes three stages: prompt, knowledge, and workflow. For prompt, Ramp gives broad access to tools—ChatGPT, Notion, Notion AI, and Perplexity—so employees must write real prompts rather than rely on hidden automation. Adoption is reinforced with onboarding boot camps, quarterly problem-solving sessions, and an internal Slack channel where people share what they tried and where they got stuck. A key lesson is that vague prompts produce vague results; Ramp recommends an “AI prompt feedback loop,” where AI asks clarifying questions, the user answers, and the user then has AI rewrite the prompt until the output is precise.

For knowledge, Ramp treats knowledge management as a first-class AI risk and advantage. It targets “A+ AI knowledge” as single-sourced, maniacally accurate, crystal clear, and actively taught—because AI systems can amplify outdated or misaligned information across many tools. Ramp uses Notion’s connectors (including Slack, GitHub, Google Drive, and Linear) and builds knowledge feedback loops: when AI answers fail or policies change, humans and agents log the issue, propose updates, and route final truth back to knowledge owners.

For workflow, Ramp focuses on scaling repeatable actions without requiring engineers for every integration. Using tools like Gum Loop, teams can connect triggers and inputs, run AI with the prompt-and-knowledge stack, and write outputs back into systems. Examples include Slack project updates, sales email drafting from research, and operations flows that branch to experts when answers are uncertain. Ramp closes by urging teams to stop waiting for a perfect moment and start building—because practice compounds, innovation spreads internally, and the work becomes more fulfilling when people automate the parts they dislike and reclaim time for judgment and creativity.