Most AI transformations end at the moment the hardest work begins.
Employees get access to the tools. They complete workshops. They learn to prompt, automate, or build. Some even produce something useful.
Leaders can point to usage, completion rates, and demos as proof that adoption is underway. Then employees return to the same meetings, handoffs, deadlines, information silos, and definitions of finished work.
And, make no mistake, training creates new capabilities within an organization. But it does not automatically redesign the organization around those capabilities.
Zapier reached this transition point quickly. The company helps businesses connect their apps and automate repetitive work, and its roughly 900 employees work fully remotely. In February 2026, Chief Marketing and AI Transformation Officer Dan Slagen launched what he called a Code Red for the marketing team: a five-week push to move employees beyond chatbots and into AI builder tools.
Slagen gave the team three weeks to get up and running on Cursor, a popular coding agent that lets people create software with AI, or an alternative such as Claude Code or Codex. He recruited five internal coaches who could offer confidential help. Then the team spent two weeks building useful tools for its real work.
The program worked. Marketers produced dashboards, automated systems, and complete applications. Zapier could see who was just getting started, who was building advanced workflows, and who was ready to create software.
But the success of Code Red exposed the next, and potentially bigger, issue. Zapier was also developing an AI marketing brain designed to give people and AI agents access to the same current company context. Making that system useful would require more than another round of technical training. It would require the team to change how it worked.
"At the moment, I don't feel like learning the next tool or the next thing is where I need the team to go," Slagen says. "I feel like it's changing the way they work to enable this marketing brain.”
Zapier's experience points to a phase of AI adoption leaders can easily miss. Tool training can create new capabilities. But it cannot redesign information flows, assign ownership, change quality standards, or make room for people to learn a new production system.
The hardest work begins when leaders stop asking whether employees can use AI and start deciding how the organization must change because they can.
Training asks whether employees can use the technology. Transformation asks whether the work now happens differently because they can.
An employee may know how to build a dashboard while the team still spends hours assembling the same weekly report. A marketer may create an AI workflow while approvals still move through the same slow chain. A manager may use an AI assistant while the knowledge it needs remains scattered across messages, meetings, and documents.
Usage, course completion, and successful builds are signs that new capability exists, which is a critical and admirable first step to AI transformation. But they are not, on their own, proof that the capability has changed a recurring process, decision, role, or result.
Early adoption sometimes needs a forcing function. A deadline, mandate, or public commitment can make action difficult to postpone.
Slagen is unusually candid about what made Zapier's initial rollout work.
"The Cursor Code Red worked probably more out of fear than inspiration," Slagen says. "I don't love that that was the case, but that's probably what the case was."
The urgency made the skills gap impossible to ignore and gave employees a reason to act. But fear can push someone through onboarding or a first build. It cannot make a new system indispensable.
For Zapier's AI marketing brain, Slagen wanted a different response. He wanted employees to experience enough value that they could not imagine working without the system and believed it helped them move faster.
This creates a shift in the leader's job. Early adoption may depend on urgency, deadlines, and participation. Sustained adoption depends on whether the new way of working is genuinely better than what came before.
When AI moves from a personal tool to a shared company system, the technical build is only one part of the job. The system needs current information and maintenance. Employees need to know when to use it. Leaders need to decide who is accountable when adoption stalls or results disappoint.
Zapier's brain could only use company knowledge stored somewhere it could access. That put pressure on employees to move useful context from private calls and messages into shared channels.
But the answer could not simply be "make everything public." Slagen's effort to communicate more openly surfaced legitimate boundaries around confidentiality and privacy. The leadership challenge was deciding which knowledge needed to become accessible without pretending every exchange belonged in the open.
The brain also needed ongoing maintenance, so Zapier assigned a full-time marketer to own that work. Other responsibilities remained less clear. Who owns adoption? Who monitors whether employees use the system? Who is accountable for its success, the person who built it or each team manager?
"That stuff's all been solvable," Slagen says of the technical challenges. "It's much more been the organizational stuff."
The harder work is deciding what people must do differently and who is responsible for making that change stick. These are not training questions. They are management decisions created by the technology.
Even employees who want to change face an immediate conflict: The existing work does not stop. Campaigns still have to launch, goals still have to be hit, and deliverables still have to reach the people waiting for them.
If leaders add AI experimentation on top of the same deadlines and quality expectations, employees have a rational reason to return to the process they already know. It is faster and safer in the short term, even if the new system could eventually be better.
Slagen changed that tradeoff.
"If I need to give them a little bit more time to get the deliverable, because I know they're going to be working in a new way, that's okay," Slagen says. "If the deliverable comes in a little bit more raw or it's not perfect, but they were able to build it with a new system, that's okay."
This is not permission to lower standards permanently. A team cannot learn a new production system while pretending the first attempt will match the speed and polish of a process employees have used for years.
Work redesign needs a transition period. Leaders have to decide where extra time, rougher first versions, or temporary inefficiency are acceptable so the new process has a chance to become the better one.
The line between training and transformation will look different in every company. Zapier had clear evidence that its marketing team was ready to cross it. Employees had access, had learned builder tools, and had produced real systems with them.
At that point, another workshop would not answer the harder questions:
These questions move AI out of the training calendar and into management, team design, process ownership, and daily operations.
Leaders should keep teaching as AI capabilities change. But they also need to recognize when lack of skill is no longer the main constraint.
Did you enjoy this transformation story? Go deeper on how AI is reshaping work and business with The Artificial Intelligence Show. Each week, we break down what matters in AI and what leaders should do about it.