An account manager. A finance lead. Two people from partner marketing. None of them engineers. Several of them had never built a piece of software before in their lives. By the time they finished demoing what they had built in five days, I was convinced that most enterprises are measuring adoption of artificial intelligence wrong, and that the people closest to the work know exactly where customer friction and revenue leakage hide.
Last month, we spent five days running an internal sprint called “AI passion week.” It was open to any employee, any function. We partnered with Gumloop to give the entire company access to their agentic workflow automation platform, wrapped the effort in real support, and made building AI workflows and agents the priority for the entire week.
What our employees built changed how leadership teams should be measuring AI adoption. The gap between what I saw that Friday and what most enterprise dashboards would have shown that same day is wider than most boards realize. It is the gap that will separate companies that turn AI into operating leverage from those that only turn it into another software habit.
THE BUILD GAP
There is a gap most enterprises are not measuring, and most leaders are not investing against. It is the gap between using AI and building with AI.
Using AI matters. It saves time and builds fluency. Your employees are summarizing meetings, drafting emails, and generating drafts of documents—all work that is real, useful, and worth doing. But it is quickly becoming a baseline skill, not a durable advantage.
Building with AI is something else entirely. It is wiring intelligence into the how work gets done, connecting it to customer relationship management, the data warehouse, the campaign workflows, and the support queue. An employee asking a large language model, or LLM, to summarize a meeting is using AI. An employee shipping an agent that ingests every customer call, flags at-risk accounts, writes a renewal brief by 7 a.m. Monday, and gives the account team a better shot at saving the relationship is building with it. They are each operating in different categories of capability. Most companies treat them as if they are doing the same thing.
Most enterprise dashboards do not measure this build gap. Most enterprise training programs do not close it. Most enterprise AI budgets are sized for habits, with very little allocated to capability. The longer it goes unmeasured, the wider it gets. The fix is not complicated: Start measuring it by who builds something that changes how work gets done.
WHAT BUILDING LOOKS LIKE IN PRACTICE
In five days, our employees built 379 working agents. The 379 number gets attention. But it is not the lesson.
What is interesting is what those agents spent the week doing. They made 64,453 tool calls into our production systems, writing to real customer records, posting in real Slack channels, and pulling from real call transcripts. They executed 958,875 lines of code in the background to complete the work. Submissions came from six different functions across the company, most of them non-technical. By Friday, a sales pipeline risk reviewer was already running in production. All of this happened inside governed systems, with clear boundaries around data access and production use.
This is the part I want every CEO reading this to sit with: The best ideas did not come from leadership planning sessions. They came from our employees—the people other companies still send to AI fluency training. While ours are shipping, their employees are still credentialing. Fluency does not become capability until people build against real work.
Two people on our marketing operations team built an agent that compresses display ad campaign creation from eight hours to one, which changes how often the team can ship and test new campaigns. That was one of dozens of agents built. A solo employee compressed a week of keyword-spotting setup into 20 minutes. A small sales team built a competitive intelligence agent that measures how quickly and persistently other companies follow up on inbound leads, then turns the results into a sharper sales narrative.
What made it possible was surprisingly simple. Most of it came down to leadership choices.
WHAT IT TAKES TO ACHIEVE
First, we made access real, for every employee, regardless of function. Using Gumloop as a build platform, they could wire AI into real systems and ship working agents. That decision had a line item attached to it.
Next, our employees needed dedicated building time. We protected the calendar by clearing internal meetings during build blocks.
To buffer against frustration, we surrounded the week with real support. Gumloop’s team worked alongside ours in a shared Slack channel and ran nine live build sessions. We implemented internal AI office hours. Over five days, our employees opened 61 support threads against roadblocks. We stressed early on that the work didn’t need to be production ready: Showing up to demo your build was what mattered most.
Leadership choices—and what it takes to achieve them—were the primary lessons. The platform license takes money. Protecting the calendar takes management capital. Internal people available to unblock other internal people takes salary hours. Public permission to demo rough work takes something harder to budget: ego, trust, and culture.
The build gap will hold most enterprises stuck in place longer than their boards think. Companies that close the build gap will have leaders who treat capability building as something with a real price tag worth paying. Enterprise access to a real build platform. Five days of protected time. A vendor partner reachable in a Slack channel. Internal people available to unblock other internal people. Permission to demo work in progress.
Companies that pay those costs will learn faster, build faster, and turn AI from a habit into a durable operating advantage.
Gregg Johnson is the CEO of Invoca.