Moats Died When Model Releases Got Weekly

A founder I respect said something last week that stuck with me: any secret you keep today is worthless tomorrow. He was talking about AI startups, but the line lands harder for digital marketers and agencies, who keep getting sold the idea that they need a proprietary AI stack to compete.

They don’t. The moat is gone. And that’s actually good news.

The proprietary AI play is dead for marketers

Two years ago, every agency pitch I saw included some version of “our custom AI model trained on X.” Most of it was a thin wrapper on GPT-3.5 with a system prompt. Some of it was real fine-tuning that became obsolete the moment GPT-4 dropped, then Claude 3, then Gemini 2, then GPT-5.

The pace makes proprietary tech a depreciating asset. If you spent six months building a custom content scoring model in 2024, Claude 3.5 Sonnet probably matched it out of the box by mid-year. If you spent Q1 2025 building a custom research agent, the native deep research features in ChatGPT, Perplexity, and Gemini caught up by summer.

The pattern is clear. Anything you build on top of a frontier model gets absorbed into the frontier model within 6 to 12 months. Sometimes faster.

So what’s left to compete on? Speed of adoption and speed of execution. That’s it.

What a zero-moat strategy actually looks like

If you accept that no tool advantage lasts more than two quarters, your operating model has to change.

A few things I’ve been testing internally:

Model-agnostic prompt libraries. Every workflow I document gets written against an interface, not a model. When Anthropic ships a new Sonnet, I can swap it into my content brief generator in about 20 minutes of testing. When Google’s pricing on Gemini Flash drops, the same swap happens. The workflow is the asset. The model is a commodity input.

Weekly model evaluation as part of the job. Not a research project. Not a quarterly initiative. Every Monday, somebody on the team spends an hour running our top five workflows against whatever shipped that week. If something beats our current default by a meaningful margin on a real task, we switch.

Kill-it-fast culture. Half the AI experiments I run die within two weeks. That used to feel like failure. Now it feels like the system working. If you’re not killing experiments, you’re not running enough of them.

Execution speed is the real competitive surface

The agencies and in-house teams pulling ahead right now aren’t the ones with the smartest AI strategy decks. They’re the ones who ship.

I watched a competitor launch a programmatic SEO play in early 2025 that took our team about three weeks to match. By the time we caught up, they had moved on to using video models for paid social creative iteration. Three weeks later, they were testing voice agents for outbound. Each individual move was something any of us could have copied. The compounding came from doing it 12 times in a year while everyone else did it twice.

This is the part people miss about AI advantage. It’s not the technology. It’s the metabolic rate of the team using it.

What this means for hiring and team design

If execution speed is the moat, then team composition matters more than tooling.

The roles I’d hire for today look different than the ones I’d have hired for in 2023. Less “AI specialist,” more “operator who reads release notes and ships by Friday.” Less “prompt engineer” as a title, more “marketer who runs evals on their own work.” Less reverence for credentials, more attention to whether someone has actually shipped 10 things in the last quarter.

The feedback loop with engineering or external partners has to be tight to the point of being uncomfortable. If a new model drops on Tuesday and your team is still in a Thursday standup deciding whether to test it, you’ve already lost the week.

The catch most marketers will miss

Here’s the part that gets glossed over. Going zero-moat sounds like permission to stop investing in anything durable, which is wrong.

The durable assets in this environment are not technology. They’re the things that compound regardless of which model is on top: proprietary data you collect from your own customers, distribution channels you own outright (email lists, audiences, owned media), and the institutional muscle of a team that can absorb a new tool every week without burning out. Those are the actual moats. They just don’t look like AI moats, which is why most people don’t notice they’re building them.

If I were running an agency or in-house marketing team today, I’d stop chasing the question “what’s our AI advantage” and start asking “how fast can we go from a model release to a production change.” Measure that number. Compress it monthly. The team that gets it under 72 hours wins the next two years, and they win it without owning any technology at all.