May 15, 2026
Why We Told Our Client Not to Use AI
There's a version of this story where we walk in, assess the situation, and sell the AI implementation. It would have been an easier conversation. The client came in with a solution vs the problem.
We said no.
They thought they had an idea intake problem. What they actually had was a product strategy process that couldn't defend itself. Features were getting added to the roadmap with no tradeoff analysis. The board had no visibility into where the product was heading. The PM team was losing trust in each other because decisions felt arbitrary and changes came without explanation. This may sound familiar. We see this a lot.
Slapping automation on top of that wouldn't have cleaned it up. It would have made it worse faster.
The Harder Path Is Usually the Right One
We convinced them to put the AI conversation on hold and let us do a deep dive, intake to feature delivery, everything in between.
What came out of it wasn't a dramatic overhaul. A few targeted changes: tighter intake controls, bi-weekly grooming sessions for the PM team, a gating process that required PMs to prove value before anything got promoted to the roadmap, and a structured path for customer-facing teams to flag injection requests without blowing up approved work.
Not a lot on paper. The difference was real.
Ninety days later, they walked into a C-Level planning session with a roadmap they could actually defend. Product had a clear answer to "why this, why now." The board could see where things were heading. The PM team, who had been quietly losing faith in the process, started working through decisions together instead of just receiving them.
AI wouldn't have gotten them there. That part required doing the work.
What Happens When You Skip It
I have a second example that went the other way.
Different client, same pressure to move fast. They had insights coming in from everywhere: Zendesk, Gong, Salesforce, Jira. The data existed. The capture wasn't structured at all. So they fed it into AI and asked it to synthesize a roadmap. This may also sound familiar.
The output hallucinated. It overindexed on loud requests and poorly written tickets. No user value metrics were applied. The critical thinking that should have shaped the roadmap, the "why does this matter to the customer" work, got skipped entirely because the assumption was that AI would handle it.
The roadmap wasn't usable. They went back to the old way and acknowledged they had a garbage-in problem. No changes yet. Just the recognition that the shortcut hadn't been one.
That's the part that doesn't come up when people are excited about the efficiency gain. Time lost. Decisions delayed. A team that tried to do less work and ended up doing more of the wrong kind.
The Human Cost of Getting This Wrong
Process change is hard. When we went back to the first client and asked their team to take on new rituals, monthly gating sessions, structured grooming, formalized injection requests, we were asking people who were already stretched to add more to their plates.
That's not an easy sell. We knew that going in.
But trust is slow to rebuild once it's gone. The team in that first engagement had already felt what it's like to work inside a process that doesn't hold. Decisions made without them. Roadmaps that changed without explanation. That kind of thing compounds. It doesn't just affect the process. It affects whether people bring their judgment to work or just execute.
The second client went the other way. Let AI absorb the tedious work, skip the heavy lifting. Nobody had to change anything. The human cost landed anyway, in the wasted effort, the unusable output, and the team that now has to figure out what to fix before they can move forward.
The bill shows up. It's just rarely the one you expected.