Ramp, the corporate finance platform, achieved something most companies can only dream about: 99% AI tool adoption across the entire company. Every employee using AI tools. And then, as Seb Goddijn, one of the engineers who built their response, wrote in a recent post: "we noticed something concerning."
Most people were stuck.
Not because the models were bad. Not because people lacked ambition. But because the harness was not there. Terminal windows, npm installs, and MCP configurations were too much for most people to navigate. The few who pushed through had wildly different setups, with no way to share what they had learned. They had created urgency without building infrastructure.
This is the pattern I keep seeing. Leadership mandates AI tools. Adoption numbers go up. And then, somewhere between the announcement and the actual change in how work gets done, most of the potential gets left on the table. The models are not the problem. The harness is.
What they built
Ramp built Glass, an internal AI productivity suite designed around three principles. They are worth restating carefully because each one pushes back against the conventional wisdom about how to introduce AI to non-technical employees.
Principle one: do not limit anyone's upside. The standard approach for non-technical users is to simplify, to put the product on rails, offer fewer options, make it dummy-proof. Goddijn's team disagreed explicitly. Power users at Ramp thrive on multi-window workflows, deep integrations, scheduled automations, persistent memory, and reusable skills. The goal was not to remove complexity. It was to make it invisible while preserving full capability.
Principle two: one person's breakthrough should become everyone's baseline. The biggest failure mode was not that people could not figure things out. It was that everyone had to figure things out alone. A workflow discovered by one person did not help anyone else. Glass was designed to compound wins into organisational capability.
Principle three: the product is the enablement. No amount of training workshops can match a targeted nudge while you are already doing the work. The product itself should show you what good looks like, in the moment.
The skill marketplace that compounds
The most distinctive feature of Glass is Dojo, a skills marketplace where employees share reusable AI workflows with the whole company. Skills are markdown files that teach an agent exactly how to perform a specific task. They are Git-backed, versioned, and reviewed like code.
Over 350 skills have been shared company-wide. When a CX engineer builds a Zendesk investigation workflow that pulls ticket history, checks account health, and suggests resolution paths, they package it as a skill. Through Dojo, the entire support team levels up overnight. A new account manager does not have to browse 350 options: a built-in AI guide called the Sensei looks at which tools they have connected, what role they are in, and what they have been working on, and surfaces the five most relevant skills on day one.
This is the flywheel that matters. Every skill shared raises the floor for everyone. The organisation learns, not just the individual.
It works while you do not
Glass goes beyond being a better chat interface. It turns a laptop into a server. Employees can schedule automations that run daily, weekly, or on custom cron schedules and post results directly to Slack. Slack-native assistants listen and respond in channels using a full Glass setup: integrations, memory, and skills intact. For long-running tasks, a headless mode lets you kick off a job, approve permission requests from your phone, and come back to finished work.
The interface itself is a split-pane workspace, not a single conversation thread. Multiple sessions side by side. Markdown, HTML, CSVs, and code rendered inline as tabs. The layout persists between sessions, so when you return the next morning, your workspace is exactly as you left it.
Why they did not just buy this
Goddijn addresses this directly, and his reasoning is worth understanding in full. There are three reasons they built Glass in-house.
First: internal productivity is a moat. The companies that make every employee effective with AI will move faster, serve customers better, and compound advantages their competitors cannot match. You do not hand your moat to a vendor.
Second: speed. When you own the tool, you see exactly where people get stuck and can ship fixes the same day. Ramp has a Slack channel where users report issues and most are resolved within hours. That is not possible while waiting on a vendor's roadmap.
Third: it directly informs the external product. Ramp builds AI-native products for finance teams. Solving the harness problem internally gives them conviction about what works before shipping it to customers. Glass gives them reps on the hardest AI product problems without those reps happening at their customers' expense.
The lesson that changes how you think about rollouts
The single most important thing the Ramp team learned was this: the people who got the most value were not the ones who attended training sessions. They were the ones who installed a skill on day one and immediately got a result.
Every feature in Glass is, as Goddijn puts it, secretly a lesson. Skills show you what great AI output looks like before you know how to ask for it yourself. Memory shows you that context is the difference between a generic answer and a useful one. Self-healing integrations show you that errors are not your fault, the system has your back.
None of this was designed as education. But when you hand someone a tool that just works, they learn by doing. And they learn fast.
This is the question every organisation should now be asking: not "how do we get people to use AI tools?" but "what is the harness that makes the tools actually usable?"