The Gap Most Operators Are Sitting In
Most non-tech companies at this point have an AI power user problem. There's one person on the revenue or marketing team who has built up a serious Claude or ChatGPT workflow. They're faster, their output is better, and everyone knows it. Leadership sees it as a win.
It's not a win. It's a single point of failure dressed up as progress.
Emily Kramer's deep-dive into how Mintlify, LangChain, and Buffer structured team-wide AI setups is the most specific account available right now of what happens when GTM teams try to move past solo AI use—and what actually breaks. The short version: shared context collapses without deliberate architecture, and role-based permissions matter more than most teams expect.
Where the Wheels Come Off
The failure mode is consistent across the companies Kramer profiled. Teams start with a shared Claude project or a GitHub-based collaborative setup. It works fine for two people. At five or six, it falls apart. Why?
Because accumulated context doesn't transfer automatically. The prompts, the institutional knowledge baked into those prompts, the workflow logic—these live in one person's head and one person's chat history. When the power user is out, the team is flying blind. When someone new joins, there's no ramp. When leadership wants to audit what the AI is actually doing across the team, there's nothing to audit.
The teams that pushed past this moved to database-backed agent architectures with role-based permissions. That's a different infrastructure decision than "let's all use Claude." It requires someone owning the system, not just using it.
The Actionable Move This Quarter
You don't need to rebuild everything. You need to answer three questions before next quarter's planning cycle:
- Where does your team's AI context live? If the answer is "in [person's name] account," you have a dependency, not a capability.
- What breaks if your AI power user is out for two weeks? If the answer is "a lot," the single-player setup is already costing you.
- Who owns the AI system versus who uses it? These are different jobs. Most teams have no one in the first role.
For RevOps and marketing leaders specifically: the move from single-player to multiplayer AI is an operations design problem, not a software problem. The tooling exists. What's missing is someone with the mandate to build and maintain the shared layer—context, permissions, workflow standards—that makes team-scale AI actually work.
That's the gap worth closing in Q3. Not adding more tools. Operationalizing the ones you already have.