"Getting on the train": How non-technical workers are experiencing mandated AI adoption

Much of what we’ve seen about enterprise AI rollouts focuses on the executives who commissioned them and the engineers who built the deployments. The people whose experience is hardest to see are the ones for whom the mandate is brand new: workers in sales, marketing, HR, ops, and support who are being told to build AI into their day-to-day work despite having little prior exposure to it.

That's the group Deel gave Claude access to in early 2026, alongside the rest of its 7,000 employees. The rollout came with a strong internal adoption push across all teams. What the experience of that rollout has looked like for the non-technical workers inside it is a story about skepticism, friction, small wins, and identity questions.

That story is not unique to Deel. AI implementation consultant Nalini George works with non-technical workers at multiple companies going through similar mandates, and the friction points she sees repeat. "Some are skeptical [that] AI can do anything useful for their specific work," George says. "Others are already trying to use AI but hit a wall on setup, things like configuring a skill or getting training data right. Most AI tools are still built by engineers for engineers, and that usability gap can be a wall for a non-technical person."

Under mandates, the skepticism group tends to shrink. "People under a mandate have already been pushed past pure skepticism, they have to try," George says. "The bottleneck isn't willingness anymore. It's that AI tools are still built by engineers for engineers, and that gap is what stops them."

The first reaction is rarely what leadership assumes it is

For Anne-Sophie Bouygues Sonden, EMEA Communications & Global Partnerships lead at Deel, the switch from ChatGPT to Claude landed as a tool change rather than a moment of professional transformation. "I was already using ChatGPT a lot, but I guess like most people, to rephrase a message, to soften something, to draft a V1 of a press release," Bouygues Sonden says. When the importance of AI adoption was even further encouraged internally, she saw it as a sort of wake-up call. “I thought, 'wow, I need to get on the train.”

Alenka Herman, a Senior Compliance Associate at Deel, entered the mandate from the opposite direction. She was already using AI when she interviewed for the role. "I first heard about it during my job interview for my current role, and I thought, great, I'm already using some AI, so this will just be a continuation of that," Herman says. "I was glad the company didn't see it as 'cheating' or skipping steps, but instead embraced the new technology. Little did I know how cutting-edge the tools at Deel actually are, how strong the push would be, and how much we're actually expected to use them."

Both entry points, resistance and readiness, are common in non-technical rollouts. Under a mandate, both eventually converge on the same friction point.

The setup wall is real, and it's a design problem

The friction non-technical workers hit is usually about tool design rather than motivation. "People were working inside a Claude Project when what they actually needed was a Skill," George says. "They didn't know Skills existed as a separate capability. So they were trying to force a project workspace to do a job it wasn't built for, and getting stuck without knowing there was a simpler tool for it. That's a config problem that looks like a capability problem from the inside."

For Bouygues Sonden, the way through the setup wall was reframing it as a personal challenge. "There's something I love, and it makes me accomplish a million things: a challenge," she says. "If you tell me I can't jump out the window, I'm going to jump out the window. So it was a bit of a challenge, and I also function on reward. When I see that something works well, it makes me want more." She started small, building her first project on Lovable, which she attempted by asking Claude how to do it and testing as she went. Somewhere in that process, the tools stopped feeling foreign.

The first small win changes the dynamic

Starting small is the deliberate mechanism that gets non-technical workers past the setup wall, George says. "In one case, someone had a project they were already trying to do with AI, but it was too big to handle out of the gate," she says. "We broke it down to find the smallest feasible slice of that same task, one that could show a working result inside a single 30-minute coaching session. Seeing it work, in that session, was what got them to trust the process."

The bigger project follows once the pattern is established. For Bouygues Sonden, that was building an internal CRM for tracking journalist relationships. The idea came from a conversation with Deel's CEO and co-founder, Alex Bouaziz. "At one point Alex told me he thought we could build this CRM. I'd had a conversation with him in Paris where I was telling him it's a bit complicated using AI in comms," Bouygues Sonden says. "I was saying I can't contact a journalist with AI, I can't do outbound like that, it's very personal. And he said, 'I think you could build a CRM that tracks the journalists you haven't engaged with in a while.' I couldn't pretend I hadn't heard it. So I tried." The result is a working internal tool that tracks journalist engagement, outreach, follow-ups, and articles for the comms team. Bouygues Sonden built it from scratch. Her pre-rollout technical skills were, in her words, close to zero.

Herman had a similar shift with Cowork, an AI workspace tool. "The real power lies in agents, and in connecting it with your existing docs, projects, etc.," Herman says. "The ability to throw several docs and files, notes from meetings, skills, and specific instructions in and let it do the heavy lifting to create a good draft or a flow is really impressive."

The identity question

Not every friction with mandated AI adoption is technical. Beyza Gursun, a workplace psychologist who runs "The Human Side of AI Transformation" events with HR leaders, describes a pattern she keeps seeing with non-technical workers. "Autonomy is one of the strongest things that keeps people intrinsically motivated at work, and a mandate quietly takes some of it away," Gursun says. "When you hand a non-technical employee a tool that writes and phrases things better than they can, what surfaces is a hard personal question: who am I here, and what [is] my expertise even worth? You can't train your way out of that, because it sits at the level of identity."

The identity question doesn't usually get talked about, Gursun says, but it surfaces indirectly. "People who are losing their autonomy often can't name what's actually bothering them. That's why it mostly stays quiet. They rarely say out loud that this takes away their autonomy or threatens their expertise. What you see is people enjoying the work less, doing what they're told with the tool but with less of themselves in it."

Herman describes the job-security version of the same question directly. "I sometimes wonder if I'm helping write myself out of a job by teaching these tools how to do it," she says. "But I'm hoping that by using them now, I'll be at the front of the tech curve and better positioned to retrain, rethink, and adapt as my role evolves."

Bouygues Sonden's read was more pragmatic. "It's not that I was afraid for my job security. But today, it's woken me up a bit to realize that I don't want to risk my job because I didn't get on the train."

What still gets in the way

The pattern that surfaces last, Gursun says, is accountability ambiguity. It usually takes a mistake to make it visible. "People are told to use AI, but in teams without much psychological safety, nobody has talked about the harder part," she says. "When something goes wrong, whose mistake is it? How do we even discuss an error that came through the tool, who owns the call that got made, and how does the team decide things now?"

The question surfaces most clearly in review work. Gursun describes a team she heard about that used to write reports manually and now drafts them with an AI agent. "They still have to cross-check every report before it goes out, and they still own what they hand in. They say that a wrong or misleading number is much harder to spot inside a clean, well-written AI draft than in one they wrote themselves. The polish hides the mistake. The responsibility still sits with the employee, but the tool makes it harder to do that job well."

The other pattern that keeps showing up is the transition tax. The tool gets sold on efficiency, but the workers using it find efficiency arrives later than promised. "Getting fluent takes trial and error and hours these employees don't have," Gursun says. "Their day already doesn't fit into a day, and now utilizing the tool lands on top of everything else."

An HR director at a regulated company traced the overwhelm for Gursun back to something specific: the hours non-technical employees spend mapping out their own processes so a tool can be built for them. "A lot of the effort is because the process was never written down. It lived in people's heads as judgment. Putting it into words takes a lot of back and forth with the technical team before it's good enough to use."

Herman describes the same friction in a smaller form. "When I'm in a rush, I still do this sometimes," she says of tasks she's been meaning to automate. "I've delayed automating some of the small, repetitive tasks I've been meaning to tackle, simply because it's quicker to just do them directly than to spend an hour, or a few days, figuring out and testing an automation. Sometimes the results are frustrating, but it keeps getting better, so it's definitely worth investing the time."

Gursun points to organizational psychologist Amy Edmondson's research on the AI rollout at 3M as the clearest example of a company handling the accountability question well. "They treated it as a learning process, so early mistakes were seen as useful information. They rewarded people for catching AI errors. Leaders talked openly about their own AI mistakes. And they set up override protocols, so someone could say no to an AI recommendation without having to over-explain themselves."

What non-technical AI adoption actually feels like

The AI rollout inside a company is not experienced the same way at every level. The executive rollout is a strategic decision, while the engineering rollout is a technical one, and the non-technical rollout is a personal one. It lands on people whose day-to-day work wasn't built around constant tool adoption, and it asks them to rewire it anyway.

What the experience of those workers actually looks like matches George's basic pattern: skepticism gets pushed through by mandate, the setup wall stalls people who are willing, the first small win changes the dynamic, and the pattern extends into a harder project. It also matches Gursun's pattern: autonomy and identity get quietly threatened, accountability stays ambiguous, and the transition eats the time it was supposed to give back.

Both are true at the same time inside the same rollout. From the inside, non-technical AI adoption feels like a genuine capability shift that comes with a genuine tax. "I see job titles changing," Bouygues Sonden says. "There's talk of jobs disappearing, but there are also lots of new jobs being created." Getting on the train, in her framing, was the point.