Software can't read the room.
And that's a product problem, not a technology problem.
My daughter turned 11 a couple of weeks ago. We took her to Olive Garden because that’s what she wanted. Eleven-year-olds don’t care about your Yelp ratings. She was ready to take what I call the “World Tour”. Aka their “Tour of Italy”.
She ordered the chicken gnocchi soup. When it was time to order, I casually said, “It’s time for the birthday girl to order,” to send a signal. She looked at the server and asked if they could put some extra gnocchi in her soup. She asked if it was a special treat for her birthday, and she had her birthday card as her receipt (just in case).
He didn’t push back.
Now, this isn’t on the menu. There is no “extra gnocchi” button in the Olive Garden POS system. Nobody wrote a user story for this. There’s probably a corporate policy somewhere about portion sizes, food costs, and consistency across 900 locations.
The waiter smiled and said, “I’ll see what kind of special magic I can work in the kitchen.” Came back with extra gnocchi in her soup. She lit up.
That moment cost Olive Garden almost nothing. It bought a family that keeps coming back.
We are all building software now, and I think most of us realize that no software would have handled that interaction the same way.
Not because the technology isn’t sophisticated enough. Because the decision itself requires something software doesn’t have: the ability to recognize that right now, in this specific moment, the rules don’t matter.
Rigidity is a feature, not a bug
We expect software to be consistent. That’s the whole point. When you open your banking app, you want it to behave exactly the same way every single time. Predictability is why we trust machines with our money, our health data, and our flight bookings.
Product managers spend entire careers making software more consistent. We write acceptance criteria to eliminate edge cases. We design systems that treat every user the same way because fairness at scale demands uniformity.
And most of the time, that’s exactly right.
But “most of the time” is doing a lot of heavy lifting in that sentence.
When consistency breaks people
Picture this. A family is driving from Michigan to Florida. Bad weather rolls in somewhere in Tennessee. Flights are grounded. Highways are a mess. They pull into a hotel at 11 pm with two exhausted kids.
The booking system says check-in is closed at 10 pm. The rate jumped by $85 due to surge pricing driven by weather demand. The cancellation policy on their original reservation three states away? Non-refundable.
A human at the front desk looks at those kids asleep in the back seat and figures it out. Waives the late fee. Finds a rate that doesn’t punish someone for a snowstorm they didn’t plan. Maybe call the other hotel to work out the cancellation.
An algorithm sees a timestamp, a price curve, and a policy. It executes flawlessly. And the family sleeps in their car.
This isn’t a technology problem. It’s a product philosophy problem.
We’ve built systems that optimize for the 98% of transactions that are routine and then abandon the 2% where human judgment is the entire product.
Follow, Break, or Bend
AI is excellent at following rules. It can also accidentally break them. What it cannot do is intentionally bend them.
That’s three different things, and we keep collapsing them into two.
Following rules is what software does best. Every transaction is processed the same way, every time. Break rules and you get chaos. Gnocchi for everyone. Late check-ins with no rate at all. Systems that can’t hold a policy because they don’t understand why the policy exists.
Bending is something else entirely. Bending requires understanding the intent behind a rule, not just the rule itself. Portion sizes exist to manage food costs and keep the kitchen running. They don’t exist to deny an 11-year-old on her birthday. A human knows the difference. A system sees the rule.
This is where product teams have actual work to do. Not replacing human judgment with AI, but designing systems that create space for judgment to happen. What if the Olive Garden POS had a “server discretion” option for small modifications? What if the hotel booking system flagged weather events and auto-adjusted cancellation policies within a range that the front desk could approve?
Those features don’t write themselves. They require someone who has sat in the dining room. Someone who has been stranded on a highway. Someone who understands that the 2% of edge cases aren’t bugs in your system. They’re the moments your product is remembered for.
That’s product management. And no language model is going to sit across from your users and feel what they feel when your system fails them at the worst possible time.
Jobs evolve. Judgment doesn’t automate.
I’ve seen the discourse. AI is coming for PM jobs. AI can write PRDs now. AI can analyze user research. AI can prioritize backlogs.
Fine. Let it.
But the job that matters most for PMs isn’t writing documents or crunching data. It’s recognizing when the system you built needs to bend. It’s designed for the birthday gnocchi moment, the snowstorm hotel moment, and the thousand other scenarios where rigid consistency actively harms the people you’re serving.
AI will get better at pattern matching. It will get better at predicting common requests. It will probably learn to suggest “would you like extra gnocchi?” based on purchase history and demographic data.
But it won’t pick up on a dad’s casual signal about a birthday. It won’t say “special magic in the kitchen” to make an 11-year-old feel like she got away with something. And that feeling is where the best product decisions come from.
The products we remember aren’t the most consistent ones. They’re the ones that surprised us by being human when we didn’t expect it.
Your job as a product person isn’t going away. It’s shifting toward the thing that's always been the hardest part: building systems smart enough to know when to stop being systems.
Until next week,
Mike Watson @ Product Party
P.S. Want to connect? Send me a message on LinkedIn, Bluesky, Threads, or Instagram.
PS: Ready to document before you schedule? I use ClickUp to keep decisions out of my calendar. Grab a free license here and see if it works for your workflow.
A very important human perspective! This often makes the difference between mediocrity and brilliance Of products.