OpenAI Is Not Your Competitor. Your Roadmap Is.

Every time OpenAI ships a release, my inbox fills with founders in crisis mode. "They just built what we were building." "Our roadmap is dead." "What do we do?"

Here is my answer: if a single model release can kill your roadmap, your roadmap was the problem.

I have shipped over a hundred products across commercial and government contracts at Virgent AI, and AltonTech. I have watched platform cycles eat companies whole. I have also watched companies survive them. The difference is never the feature set. It is the foundation underneath.

When Products Become Features

This is the core question TechCrunch Disrupt 2026 is putting on stage this October, with Michel Tricot of Airbyte, Rob Toews of Radical Ventures, and Linda Tong of Webflow. The session is called "What Happens When OpenAI Ships Your Roadmap." The fact that this is a headlining conversation at the biggest startup event of the year tells you everything about where the industry is right now.

The question has shifted from "Can we build it?" to "Can we still own it?"

That is the right question. Most founders are asking the wrong one.

Features Are Not Companies

I co-founded Magick ML. We partnered with Google. We failed spectacularly. One of the things I learned in that wreckage is that proximity to a platform giant is not a moat. It is a liability you have not accounted for yet.

Founders who build on top of foundation models and call it a product are making a bet. The bet is that OpenAI, Anthropic, and Google will leave that specific capability gap open long enough for you to build a business around it. That is a shit bet. These labs ship fast and they ship broad.

The data backs this up. 80% of AI wrapper startups are projected to fail by end of 2026. OpenAI's product releases alone have already cannibalized 200+ funded companies that built their entire pitch around a prompt. The startups surviving, at 55-71% rates in vertical, workflow-embedded, and infrastructure categories, all share one trait: a moat the foundation labs cannot ship in a release note.

That number should scare you. And then it should focus you.

What Actually Survives Platform Shifts

I have been through enough platform cycles to know what holds. Here is what I have seen work, across projects, across clients, across my own companies:

Workflow lock-in. When your product is embedded in how a team actually does their job, switching costs are real. Not theoretical. Real. The user has to retrain, re-integrate, re-trust. That friction is your moat. Build for it intentionally.

Data moats. Proprietary data that improves your outputs in ways a generic model cannot replicate. But here is the nuance: data is only a moat if it meaningfully improves outputs, decisions, personalization, or reliability. Sitting on a database is not a moat. A feedback loop that makes your product smarter every time a user touches it is.

Trust and domain credibility. In government work especially, this is everything. I have won contracts not because we had the best tech, but because we had the track record, the clearances, the relationships, and the institutional knowledge. A foundation model cannot ship that in a quarterly update.

Distribution and network effects. Who you already have access to, and how deeply they are connected to each other through your product, compounds over time in ways that raw capability does not.

The Shallow Roadmap Problem

Here is the honest diagnosis. Most founders whose roadmaps get eaten by OpenAI were building features, not companies. They saw a gap in what the models could do, wrapped it in a UI, raised a seed round on the demo, and called it a product.

That is a fine experiment. That is a terrible company.

The question is not what happens when OpenAI ships your roadmap. The question is why your roadmap was that shallow to begin with.

At Kindred Ventures' takeaway from Disrupt 2025, the theme was clear: the next generation of category-defining companies will pair AI-native design with deep domain expertise. Not AI on top of existing systems. Rebuilding the systems themselves. That is the unlock. That is what creates lasting value.

I am building Cadderly right now. Coordination agent. Intent recognition, MCP integration, A2A agent coordination. The reason it is defensible is not because it uses a particular model. It is because the coordination logic, the workflow integrations, and the trust architecture are things that take real time and real domain knowledge to build. A model release does not erase that.

What To Do About It

If you are an AI founder reading this, here is the audit I would run today:

• Strip out the AI. What is left? If the answer is nothing, you have a feature, not a company.
• Map your switching costs. If a user wanted to leave tomorrow, what would it cost them? If the answer is "not much," that is your problem statement.
• Identify your feedback loops. Is your product getting smarter from usage? Is that intelligence proprietary? If not, build toward it.
• Find your domain depth. What do you know about this vertical, this workflow, this customer that OpenAI does not and cannot learn from a release cycle? That knowledge is your real asset.
• Audit your roadmap. How much of it is "build what the model cannot yet do" versus "deepen what only we can own"? Shift the ratio.

The Panic Is Telling You Something

Every time I see a founder panic at a model release, I see the same thing: they built on the assumption of a capability gap that was always temporary. The panic is useful information. It is telling you that your defensibility was never real.

The founders who are not panicking are the ones who built on workflow, data, and trust. They see a new model release and think: great, my product just got more powerful. They are not threatened by the platform. They are leveraging it.

That is the company worth building.

If you are at TechCrunch Disrupt in San Francisco this October 13-15, the session with Michel Tricot, Rob Toews, and Linda Tong is worth your time. These are operators and investors who have lived this. Go listen.

And if you want to think through your own defensibility before you get there, reach out. I have done this enough times to have opinions worth sharing.