A logistics manager tells me:

"The receiving screen is too heavy. My receiving clerks enter everything at the end of the day. By then, they've forgotten half the discrepancies."

A few weeks later, another logistics manager, at another customer:

"We made the batch number, the expiry date and the discrepancy reason mandatory. Result: they tick 'quantity compliant' everywhere so they can move on to the next truck."

Then financial control speaks up:

"Without discrepancies entered line by line, reconciliation between the order, the receipt and the invoice falls apart. We pay invoices we shouldn't pay, and supplier disputes are exploding."

Three pieces of feedback. All serious. All true.

And when it comes time to sum them up for the team, I write, almost in spite of myself, the sentence everyone writes:

"We need to simplify receiving while making the data reliable."

The sentence is correct.

It is also perfectly useless.

Smoothing over is the worst thing you can do

Smoothing over means rewriting contradictory feedback until the contradiction disappears.

You take what the dock says. You take what accounting says. You look for a phrasing that holds both. You end up with a balanced, reasonable, consensual sentence.

And you have lost what mattered.

Because the two pieces of feedback are not saying the same thing with different nuances. They describe two realities that pull the decision in opposite directions:

• the data-entry constraint at the dock degrades the information actually collected;
• the lack of structure degrades invoice reconciliation and the relationship with suppliers.

Both are true at the same time.

That, precisely, is the product problem.

The smoothed summary makes it disappear. It produces knowledge that looks coherent but no longer matches what customers said. The team reads "simplify while making the data reliable", everyone projects whatever they want onto it, and the decision gets made on a fiction.

The problem, meanwhile, does not disappear.

It comes back in production. As a simplified screen that accounting rejects. Or as a reliable screen that the dock works around.

Why everything pushes us to smooth things over

If smoothing over is a mistake, why do we all do it?

Because everything pushes us to.

Meeting minutes call for a conclusion. A slide calls for a message. A ticket calls for an action. Nobody wants to bring to a steering committee a problem that boils down to "both sides are right and they don't fit together".

A contradiction is uncomfortable.

So we make it disappear.

And with AI, the temptation grows even stronger.

An AI smooths things over naturally. Give it several pieces of feedback on the same topic and it looks for what they have in common. It merges whatever looks alike. It produces a well-written, nuanced compromise that gives the impression of having taken everything into account.

Which is exactly what we ask it for, by the way. "Give me a summary."

It does.

And the tension vanishes into an elegant sentence.

A space built to keep tensions

For several months now, I have been working on my product topics in what I call a Context.

It is a durable workspace, separate from the conversation with the AI, that keeps what a piece of work needs to hold on to over time: the material received, the ideas drawn from it, the decisions made and the progress status. It can fit in a few readable files that the AI rereads at every session.

Its basic rule fits in one sentence: it keeps what the sources actually say, including when they contradict each other.

This is not a constraint added after the fact.

It is the whole point.

In practice, this rule translates into a few very simple moves.

Each piece of feedback stays intact

The first move is to keep each piece of feedback exactly as it was received.

The interview with the first logistics manager goes into the Context as a source. So does the second. So does the note from financial control.

Nobody rewrites them. Nobody summarizes them on their behalf.

That sounds obvious. It isn't.

In most organizations, a piece of customer feedback lives mainly through what was said about it: a line in meeting minutes, a reworded ticket, a sentence on a slide. Three months later, the source is gone and only the interpretation remains.

Keeping the source intact means keeping the possibility of going back to what was really said, before anyone interpreted it.

It is the first protection against smoothing.

Each idea keeps its place

The second move is more subtle.

From each source, you extract an idea. One per card, phrased to stand on its own.

The first interview produces this one:

Overly constraining data entry at the dock can lead receiving clerks to postpone receiving and lose discrepancies.

Then comes the second interview.

It does not describe exactly the same behavior. The first dock postpones data entry. The second fills it in, but only for show. And yet both say the same thing underneath: a constraint meant to obtain more information ends up reducing its actual quality.

So the second interview strengthens the first idea. It grows:

Overly constraining data entry at the dock can lead receiving clerks to postpone or work around receiving, at the expense of the information collected.

Two sources now support it.

Then comes the note from financial control.

This is where everything is decided.

The temptation would be to fold it into the same idea, to qualify that idea, to turn it into a compromise. That would be a mistake.

Financial control does not bring one more piece of evidence. It carries a different proposition. So it gets its own card:

Receipts structured line by line are necessary to reconcile orders, receipts and invoices.

This is the boundary that decides everything:

• when a source says the same thing differently, the existing idea grows;
• when a source says something else, it deserves its own idea.

Strengthening, yes.

Merging to silence it, no.

The tension becomes an idea in its own right

The Context now contains two distinct ideas, pulling in opposite directions.

The third move is to name what putting them side by side reveals.

You write one more idea, which replaces neither of the first two:

The quality of a receipt depends on a tension between the effort asked of the dock and the structure accounting needs.

This is not a diplomatic average.

"Simplify while making the data reliable" tried to make the conflict disappear. This sentence, on the contrary, puts it at the center. It says where the real product problem lies: it is neither about simplifying nor about making the data reliable, but about finding which information must be entered at the dock, and which can be entered elsewhere or later.

That is a far more useful question.

And it did not exist as long as we were smoothing things over.

The contradiction stays open

The fourth move is to record the contradiction in the state of the work.

The Context permanently holds a state: what seems settled, what remains uncertain, the next action. The contradiction appears there explicitly:

What seems settled Data-entry constraints at the dock cause postponement and workarounds. Enough structure is still needed to reconcile invoices.

What remains open We don't know which discrepancies absolutely must be entered at the dock, and which can be entered later.

Next action Prepare interviews that tell the causes apart instead of presenting a solution.

Every time the case is picked back up, this contradiction reappears.

Nobody can forget it for convenience. Nobody can decide in committee as if it did not exist.

It stays open as long as the sources do not allow it to be settled.

The AI becomes a guardrail

This is probably the effect I did not expect.

A few days later, I pick the topic back up and suggest to the AI:

"Maybe we could make the discrepancy reason optional."

In an isolated conversation, the AI would probably have defended the idea. It would have offered a lighter interface, listed the benefits, perhaps suggested a nice clean screen.

In the Context, it runs into the two ideas and the tension that connects them.

It answers roughly this:

"This proposal would reduce part of the friction described by the two logistics managers. However, it runs against the reconciliation need expressed by financial control. The sources show two real constraints; they do not yet let us choose between removing the field, deferred entry, a shorter list of reasons, or adapting by supplier."

The AI does not decide for me.

It simply prevents my new intuition from erasing a reality already established.

This response does not come from general knowledge of product management. It is grounded in the work in progress. The AI knows which question we are trying to understand, which ideas rest on which sources, and which contradiction remains open.

The same AI that smoothed things over in a conversation becomes, in a Context, a guardrail against smoothing.

The first deliverable does not settle anything

The product reflex, at this stage, would be to design a new receiving screen.

The Context does not suggest that.

The first deliverable is more modest: an interview guide for going to see the dock and accounts payable. It should help understand:

• when receiving is actually entered;
• which discrepancies really matter for invoice reconciliation;
• which ones can be completed later, and by whom;
• what forms of workaround already exist.

The deliverable does not settle the contradiction.

It prepares the material that may allow us to do so.

The tension, instead of being hidden, steers the research.

Humans stay in charge of the boundaries

It would be wrong to claim that all of this works on its own.

The AI will not always choose the right boundary. It may want to merge two ideas because their topics look alike. It may also create two ideas where there was only one.

The Context does not guarantee the right choice.

It makes the error visible and correctable. You can reject a merge, ask for a separate idea, check the relationship created between two cards. And that correction sticks, because it is written down.

Judgment stays human.

The Context simply gives it something to hold on to.

An open contradiction is information

Customer feedback that is contradictory is not a flaw in the material.

It is often the most valuable kind.

It shows the exact spot where the product has to make a choice. It signals that two functions, two uses or two real constraints are meeting. It says where the real design work is hiding.

Smoothing it over means throwing that information away at the very moment it becomes useful.

An open contradiction is information.

A smoothed summary is a mistake you can't see yet.

---

A note on the examples

The examples in this article draw on more than thirty years of practice in IT and team management, and on many exchanges with peers. I have modified, generalized or recombined them: none describes a real customer, product or organization, nor reflects the data or positions of my employer. The opinions expressed are mine alone.