Ramp’s internal AI usage numbers for March are in, and they’re ridiculous. Usage by our team is up 6,300% year over year and 400% since three months ago. 99.5% of the team is active on AI tools and 84% use powerful coding agents weekly. Over 1,500 apps shipped on our internal app hosting platform in six weeks, contributed by 800+ different builders. On the production codebase side, non-engineers now account for 12% of all human-initiated PRs – thousands per month – using our home built coding agent, Ramp Inspect.

Our AI usage charts are starting to look suspiciously like our company logo. The personal AI assistant helping me edit this article finds this very funny.

It’s not enough. We just kicked off an internal AI hackathon for our non-engineering teams, with 700 participants – sellers, CX, legal, marketing, finance – coached by 100 of our most AI pilled engineering and product teammates. They’re all building in Glass, our home-built application that brings all the features of a highly configured Claude Code to our whole team with no need to open the terminal. Glass comes complete with hundreds of domain-specific skills served by Dojo, Ramp’s internal skill marketplace. I used three writing skills to help tune this article. The hackathon winners will receive a laptop. They will also be incredibly good at using AI.

If these results are interesting to you, I want to talk about how we got here. The interesting part (especially if you are on the hook for similar results) actually isn’t the numbers or the tools. It’s that we didn’t have a plan when we started. We had a culture that turned out to be the right raw material, and we just kept doing things with the people and technology in front of us and watched it compound. The most important lesson is the simplest one – just get started.

The starting conditions

Ramp has always been an impatient company. Allergic to inefficiency. Curious about new tools. The kind of place where people try things without asking permission, which occasionally creates chaos but mostly creates velocity.

In early 2025, I formalized an internal AI effort. I was head of ops and picked up “internal AI” as a leadership area, which is to say I was the person who wanted it enough to own it. I had a PM moonlighting part-time and was lucky to land some enthusiastic data and ops partners. No dedicated team. No real budget.

At our January 2025 company kickoff, I got up on stage and told the whole company we would become the most productive company in the world. I believed we could do it given Ramp’s culture. I had no idea how at that point.

For most of last year, nobody pushed back on the vision, but few people actively pushed forward outside of our engineering org. The investment followed the results, not the other way around. That’s worth being honest about, because most companies wait for an executive mandate and formal funding before starting. We had conceptual support and a culture where people could lend their time freely, and it turned out to be more than enough.

It’s a learning curve, not a light switch

A year ago, most of us used AI the way everyone did. ChatGPT in a tab, useful AI knowledge searches in Notion. Fine, but not transformational.

I started mapping what AI proficiency actually looks like at the individual level, and what I observed was that ability to drive productive output leapt up when you cleared certain thresholds of comfort. Almost nobody outside of some special engineers was operating at the upper levels before 2025, and even if they did the technical hurdles of the models were punishing. But in late 2025 and this year, we’ve been able to accelerate massively up the curve, because we spent last year building a strong Level 1 foundation first.

Moving people up the curve – especially from L1 to L2 – means building tools and training that meet people where they are, and reshaping the talent strategy (hiring, onboarding, training, performance, culture) to ensure higher AI acumen by default. We started by shifting the whole company to Notion AI connected to all of our other workplace tools, providing a low technical bar where everyone could participate and get meaningful benefit. As the tools matured, we raised expectations. AI proficiency moved into hiring screens, onboarding, and how we talk about performance. Not as an end in itself, but as a tracked and stated expectation that getting good at using these tools is essential to doing any job at Ramp well.

Each cultural step up only worked because the tools had kept up enough to support it. You can’t tell someone “use AI to transform your work” when the tools available to them are a completely disconnected ChatGPT app and vibes.

Embrace creative destruction

Ramp is exhilarating to work at this year, but uncomfortable, and this is why.

Many of the tools we shipped in January 2026 are already obsolete, replaced by better versions, often from the same builders. We’ve had to get our team comfortable with a shelf life of weeks, not months. Every update to base LLM models, every improvement to Claude Code or Codex harnesses, every new batch of Claude skills we release, reshapes what’s possible. If your internal tools from three months ago still feel state-of-the-art, you’re not moving boldly enough.

Using AI to democratize our data tells this story well. When our best option was Notion AI, we would pipe important data into Notion databases just to run our agent over it. Then we launched Ramp Research, a powerful slack based Snowflake research tool. Then as the Claude Code and Ramp Inspect power curve rose, we began to encode Snowflake research into skills those coding agents could use directly. Now we’re focused on how to make this data research interactive and self improving via new innovations. Each generation opened doors the previous one couldn’t and changed ambitions overnight. Each former generation was quietly sunset.

Builders at Ramp expect to rebuild. Trainers expect to retrain. Managers expect to re-scope. Users expect to migrate when a new demo drops. The tools we’re running right now? We genuinely hope they’re obsolete by June.

From the outside, this looks chaotic. From the inside, it’s the opposite. People aren’t attached to their tools. They’re attached to their problems. When a better way to solve the problem shows up, they grab it.

Build from the center, drive from the spokes

We got the organizational design wrong before we got it right.

My initial instinct was to centralize: my small team would build tools for the whole company. The demand outstripped our capacity almost immediately. Then, out of necessity, we swung decentralized, letting every team build their own things. There was a lot of redundant re-learning of the same lessons.

The answer was ultimately to do both. A small collection of central teams builds and improves the platforms, the connectors, the plumbing across LLMs, data, knowledge, and workflows. That team also manages the training, the enablement, the change management. Functional teams then build on top of those platforms and give the feedback that helps drive the central team’s roadmap in a powerful feedback loop.

As a result, the end products built are highly attuned to the needs of functions. A risk analyst automated 16 hours per month of manual financial modeling. A sales ops lead replaced a spreadsheet-based comp model across three orgs in 48 hours. An L&D lead built a training simulator in 15 minutes. Someone in finance built a contract reviewer that saves 45 minutes per contract – and Ramp has a lot of contracts.

None of them are engineers.

These people didn’t file a ticket. They found their own pain, prototyped a fix, and pulled engineering in when it was time to go to production (when that was even necessary). The value was there to be captured by those who started building, hit a wall, told us what they needed, and got us to build the infrastructure to unblock them. The spokes drove the center as much as the center drove the spokes.

Give people a stage, not just a mandate

The strategy, to the extent there was one, was to light as many small fires as possible and see which ones grew. Early on I started a Slack channel called #ramp-uses-ai just to…see what happened? I randomly started AI office hours every Friday – we often get 40-50+ people popping in with questions these days. We built AI onboarding into the new-hire experience – and then rebuilt it four different times over the last year as our ambition grew. Each of these was designed to find the people who were already curious and give them a reason to go deeper.

The early converts mattered more than anything. There was always one person on any given team we would see – the ambitious sales ops lead, the frustrated product operator, the eager data scientist. They got curious, got sucked in, and became contagion for their teams. As they leveled up, we made them a visible part of the overall effort, with company All Hands spotlights, giving them resources to build team-level tools, and pairing them up when a collaboration was warranted.

Our CEO and senior team are vocal and genuine in asserting that AI is a priority. That matters. But mandates decay, and culture is what remains. The #ramp-uses-ai channel now has over 1,000 people in it. It has spun off 40+ team-specific channels (#risk-uses-ai, #csms-use-ai, #growth-marketing-uses-ai) that collectively generate 20,000 messages per month. Every day, someone posts a new app, a new workflow, a Loom walkthrough. People react, ask questions, request copies for their teams. Our platform team monitors the channels and swarms on blockers.

All this public building creates a competitive dynamic that everyone feels. Nobody wants to be the team that isn’t building anything. When a CSM sees a risk analyst ship something that saves 16 hours a month, they don’t think “good for Risk.” They think “what can I build?”

That loop (build, share, inspire, build more) does more than any mandate or memo.

Why it worked

We didn’t start with a better strategy than most companies trying to do this, but we might have had better starting conditions. A culture that rewards speed and initiative. People who try things without waiting for permission. A leadership team that wants to back bold, winning AI ideas because we know that ultimately it’s good for our customers.

In lieu of a master plan, we just started. We kept building (or buying) tools, kept raising the bar, kept investing in data and AI infrastructure, kept creating venues for people to show off. Each of those tracks compounded separately, and as they continued reinforcing each other, the curve went vertical.

We’re hiring the people to build with us. AI Operators on my team, platform engineers, and builders across every function. If you want to work somewhere that non-engineers ship production code, internal tools have a shelf life of weeks, and nobody thinks that’s weird — come build with us at Ramp.