Don't Do AI to People

John Matoso joined me for the first Coffee Brief. He owns SubTonique. We have been talking for over a year. This one was live and unscripted, and we ran long.

I asked him what happened to change management. The question we opened with, from Kishore Kumar, was how you build trust, capability, and ownership alongside the technology. A lot of companies skipped that. They put a tool in people’s hands and told them to figure it out.

What I took from John

John will not start with the tool. If you frame the change as a technology project, he thinks the conversation is already wrong. A lot of what he still hears is about the tool. I see the same thing in how this gets marketed. The feature is the easy story. People want to know what changes in the work.

He wants the first question to be about value. What is value to you, in this workplace, in the work you do. He is wary of automation that copies a process as it stands. If the process was already flawed, faster just spreads the flaw. His move is to bring in the people who do the work before anyone redraws it. Ask what they dislike about the day, and what they would rather be doing. Then the tool can enable that work, which is a different conversation from a mandate dropped on the team.

He named the trap he is seeing now: forever pilot cycles. Pilot after pilot, and it never becomes the way the work is done. A line he brought from the week before we recorded stuck with me. Don’t do AI to people. Do not dump it on them and expect them to pick it up. Bring them in and listen. He thinks that extra time is worth it, because the people who do the work know where a redesign would help. A lot of processes, he said, are leftovers from some earlier system. He gave the example of a CRM rolled out ten years ago, with the whole flow built around it.

He called AI literacy essential. A rollout handled badly can hit the organization’s reputation. Older change work had time to see the impact coming. Someone can pick this up tomorrow.

The case he wished he could play for the room was an agentic rollout at Microsoft that still had to be change managed. The company shipping the models needed help adopting them, and that work started as a value conversation. He said he would share the case, and a desirability matrix, afterward. He described the matrix as more than “is it feasible.” It also has to be wanted by the client and by the employee, especially when the employee is the one naming what to automate away. Those links are not in yet.

Where I pushed back

I agree that people belong in the room. I do not agree that the first move should be reinventing the work.

People do not trust this yet, and adoption stalls on that fear. The way I build the trust is a quick win on a workflow they already run. Same job, same steps, done faster. On the episode I used thirty minutes or an hour back in a day as the kind of proof a person can feel. Once they have seen it work on their own work, then we can talk about making the process better. Open with a redesign and you are asking them to believe the tool can invent a better process before it has earned the right to touch the current one.

John thinks that path is how you end up piloting forever, and that the people closest to the work already know what should change. He may be right that a flawed process, automated, is a faster flawed process. I still want the first proof to be small and familiar, on the workflow they already have.

There is a practical problem under the curiosity argument too. John pointed at how I actually learned this: tinkering, trial and error, with license to explore. He thinks a workplace should start from that same curiosity, inside real security limits, with room to try things so the learning compounds. A rule that says thou shalt use it this way will not do it.

That runs into the calendar. There is rarely time set aside for a working professional to learn and try things. Developers often have to poke at something until it works. Operators, marketers, and the people running the day are handed a procedure and a full slate of work. Tinkering does not happen in those hours.

The line I left with

The more I use AI, the more I crave actual human interaction. Coding and research were the examples we used. Those are parts I can trust a tool to bring back, so I can decide where to take the work. That frees an hour for a real conversation about what we are trying to make. It only works after the trust is there.

John would start that trust by asking what is important to the person in the job, on an ordinary day, and designing around the answer. I would start by taking one piece of their existing work off their plate. We did not settle it. We ran out of time. That is the argument I want to keep having with him.

Key takeaways

Ask what the work is worth to the person doing it. John keeps the first question on value, in that workplace, before anyone picks a tool.

A faster flawed process is still flawed. If you automate a process that was already wrong, you spread the mistake. He would bring the people who do the work in before the redesign.

Quick wins are how I earn trust. I would prove the tool on a workflow people already have, so they can feel the time back, and only then talk about reinventing it.

Forever pilots are the failure John is watching. Pilot after pilot teaches the team that this never becomes real work. “Don’t do AI to people” was his shorthand for dumping a tool on them.

Teams are not given time to tinker. Curiosity is how I learned to use these tools. It does not show up for a busy operator unless someone protects the hours.

More AI made me want people in the room. Once I trust the tool with a piece of the work, the conversation I want is a human one.

Timestamps

Time	Topic
01:38	John Matoso’s background: 25 years in change, launch of SubTonique alongside ChatGPT
04:24	Why the human element is essential to AI implementation
06:14	Marketing features vs. outcomes: people care about impact, not specs
07:38	Trust, capability, and ownership as the real adoption metrics
09:02	Reframing around personal and workplace value
09:59	Automation vs. reinvention: accelerating flaws vs. creating new value
14:02	Bringing employees into process redesign through desirability
16:48	The trap of forever pilot cycles
17:51	”Don’t do AI to people”: co-design over imposition
19:43	AI literacy as a prerequisite, not an afterthought
22:23	Building trust through demonstrated value
26:38	Curiosity and tinkering as the original learning path
28:39	Scaling curiosity: giving teams license to explore safely
29:04	The time deficit for business professionals’ learning
33:19	Augmentation over replacement: keeping headcount while expanding capability
34:33	Desirability matrix: feasible, viable, and wanted by the people doing the work
36:42	Series announcement and next steps

About the guest

John Matoso is a change consultant. Counting communications, training, business consulting, and change work, he has been at it for close to 25 years, including a long stretch as an internal consultant. He launched SubTonique about three years ago, around the time ChatGPT came out.

If you want help choosing that first workflow and earning the trust to go further, that is the work I do as a fractional CAIO. Here is how that works.