AI product management: Why insight quality is the moat

The cost of building software has collapsed. Anthropic launched Claude Design, a credible Figma competitor, built in a fraction of the time it would have taken five years ago. That kind of event is no longer rare—it's the new baseline. And it has done something most product organizations have not yet absorbed: It has dissolved the first pillar of the moat strategy that most companies were operating under.

Malte Scholz, Head of Product and co-founder of airfocus by Lucid, has been part of this conversation at the leadership level, watching many AI product managers skip this entirely. The conventional response—ship more AI features, faster—lands at what Jason Lemkin calls “60% solutions:” a chat layer on top of an existing product. It accelerates commoditization rather than defending against it. The conversation worth having is a different one, and most product leaders are not at the table where it's happening.

The silence at the top of product

Most product leaders did not design the moat their company operates under. As Malte puts it in his LinkedIn post on the moat conversation most product leaders aren’t invited to:

AI dissolved the first pillar overnight. Malte names the Claude Design moment as the wake-up call: "If they can pull this off in that short amount of time, nobody is really safe." Software-building itself is no longer a defensible competence. Whatever your output was a competitive advantage on, it isn't anymore.

Despite that, in most companies, nobody is renegotiating what comes next. The instinct is to treat the new conditions as a feature-roadmap question. A product manager gets a ticket to add a chat interface to the dashboard. A CPO presents an "AI strategy" slide to the board. The executive team approves the roadmap and moves on. Nobody asks the harder question: What actually protects this business now?

There is a deeper reason the conversation does not happen. The signal infrastructure most product organizations run on is too broken to enter it credibly. According to the Product Ops Report:

• 52% of product teams integrate customer feedback only occasionally.

• Only 18% of strategic decisions are based on real data.

• 92% of teams say the loudest voice in the room influences decisions, not signal.

When your product feedback is fragmented across Slack threads, support tickets, and someone's Notion page, you cannot walk into a moat conversation as the person who knows what customers actually need. You are locked out before it starts.

What still holds when building software is cheap

AI has driven the costs of building software and generating product ideas to near-zero. The cost it has not touched is the one most product organizations underinvest in: deciding which idea is worth building. That decision depends on something AI does not have by default: real customer signal, structured into a system AI can reason from.

In Malte's words, "Product insight quality is the new moat. In a world where AI can generate infinite ideas, the only edge left is knowing which of the signals matter."

The AI moat debate changes from "What AI feature do we ship?" into "What signal do we own that nobody else can act on?"

The five product management moats that survived AI

Five things still hold defensible value after AI collapsed the building cost:

1. Unique access to valuable customer and product usage data

2. Interconnections inside your system: the web of context that links feedback, insights, opportunities, initiatives, and decisions

3. User experience as a differentiator

4. Switching costs

5. Distribution and enterprise relationships

None of these is sufficient alone. The combination is what holds.

Each layer matters for a different reason. On data specifically, Malte is direct:

Having data is table stakes. The moat is whether the quality is good enough to feed into:

• AI that recommends actions

• Strategic decisions made on signal rather than narrative

• Workflows the customer actually runs through

Switching costs and distribution are real moats, but Malte is clear that none of the five holds in isolation. The deepest of them, and the one that compounds the fastest, is the connected context layer.

The deepest layer: the context infrastructure

This is the moat-within-the-moat, for two structural reasons:

1. It is exactly what AI needs to be useful. Without that web, AI gives you 60% answers: generic recommendations that sound plausible but ignore your specific customer history, roadmap context, and past decisions.

2. It compounds with use. A competitor can copy your features in weeks, but they can’t copy years of decision logs and feedback-to-roadmap traceability.

This is why the moat conversation, properly run, is a question of architecture.

Five strategic principles for product leaders

If the moat is the system around the AI, then the work is building that system. Here are six principles to run that work.

1. Treat your customer feedback layer like critical infrastructure

Sales committed to CRM as their critical infrastructure decades ago. Product still treats customer signal as something you collect when there is time. In the AI era, the feedback layer is not a side workflow; it is the foundation your AI runs on.

The mechanism has two parts.

1. The practice: Feedback has to be captured continuously, by someone responsible, against a quality standard. Without that, the foundation rots regardless of what tools sit on top.

2. The architecture: A captured signal does not live as a Slack message or a buried ticket. It lives as a structured insight, attached to an opportunity, linked to the initiative considering it, with the decision and rationale logged when product moves forward or doesn't. That is what gives AI enough context to be useful—and what gives a competitor nothing to copy from the outside.

A changelog can be reverse-engineered. A year of decision logs cannot.

2. Audit your AI features against the 60% test

"If you land at 60%, you're fed, because there will be companies that will do 80 or 90—and their products will be better. AI should not only allow you to ask questions to your CRM, but also tell you which customers or markets you should go after with which sales emails or which marketing campaigns."

Two questions to ask of every AI feature:

• Does it tell users what is in the data? That is the 60% version.

• Does it tell them what to do? That is the 80-90% version that wins.

Both versions can use the same LLM. The 90% version has a curated signal infrastructure underneath that lets the AI reason from real context: which customer, which segment, which usage pattern, which historical decision worked.

3. Compete on decision quality, not shipping speed

Every product team has achieved AI speed gains by now. The remaining frontier is decision quality: which problems you choose to solve, which signals you trust, and which trade-offs that implies under uncertainty.

The organizations that win over the next five years will look back on their decisions with success because they were made on better signals. Malte is explicit about the human role in this: For complex strategic work, "I take it as inspiration to either challenge something or change perspectives to then come up with the final view that I develop."

AI sits at the inspiration and challenge layer. Human judgment stays at the decision layer.

4. Pay attention to your own commoditization

Pair this with the operating posture Malte describes at airfocus: "We see the world, and we behave every day like it's day zero, and we have to prove ourselves again. We are all in on AI, and we are very aware that we have to go beyond 60% in order not to be in danger of being commoditized."

The 60% version of your product already exists in someone's lab. Anthropic showed the world what happens when a focused team with the right models attacks a category. The right response is to assume the 60% replica is coming, and ask what compounds in your favor that the replica cannot have.

5. Run the moat conversation, this week

Here are three concrete moves to make this a working session:

1. Open the conversation with a structure. Current strengths → AI-challenger lens (could a focused competitor with current models rebuild this in three months?) → exposure analysis → defense plan.

2. Use an LLM to shortcut the framing work. The session goes faster with a tool that can pressure-test the current strategy. The LLM does not replace the leadership commitment that has to follow.

3. Sharpen with a 30-day test. Design one experiment whose result would tell you whether your moat is real or a story you have been telling yourself.

This is the smallest move in this article. The commitment to act on what it surfaces, that’s where the real work starts.

What the moat conversation actually changes

When Malte looks ahead, he sees a slow but fundamental change: "Companies will not go away. They will have time, and this will happen slowly."

No incumbent collapses overnight. What happens instead is more uncomfortable: "A company like HubSpot may not be as big as it is in three years."

The companies losing the moat conversation right now will still be operating in five years. They will just be smaller and quieter than they thought they would be. The decline looks like normal market churn from the inside until it doesn't.

That’s why the work is less strategic than it sounds. It’s disciplinary. Nobody outside your company has the incentive to tell you honestly whether your moat is real—the market says so late, competitors not at all, investors after the damage. Honesty has to come from within, continuously, because the conditions keep changing. Centralized feedback and real insights loops built into the daily workflow are the form that discipline takes in a product organization.

The cost of starting the conversation next week is small. The cost of not doing it compounds quietly, on a timeline that is hard to see and impossible to reverse once it is obvious.