A 2023 benchmark found 73% of U.S. marketers had used generative AI tools. However, only 6% of B2B teams qualified as AI high performers, according to a separate B2B workflow integration study. This reveals a 67-point gap between adoption and actual value. The tools aren't the bottleneck; the operating model is.

Emily Kramer, co-founder of MKT1, emphasizes this point. In the first installment of a three-part newsletter series published in July 2025, she introduced "Multiplayer Claude": a framework where the entire marketing team uses the same AI system with shared skills and context. The goal is clear: whatever insights the most AI-native teammate discovers should benefit the whole team.

This concept seems obvious, yet it rarely happens.

Why Individual AI Fluency Doesn't Scale

Most B2B SaaS marketing teams are stuck in a pattern where one or two individuals excel at prompting. They create personal workflows, save custom instructions, and may automate reporting tasks. Unfortunately, these workflows remain in their heads or local environments. When they leave or change roles, that knowledge departs with them.

Gaetano Romeo, a marketing leader managing a distributed team across five markets, described this failure mode in the comments on Kramer's post: "Everyone's individually good with Claude, but nobody's outputs compound because there's no shared system to ship into." The 19% of B2B teams that integrated AI into daily workflows in 2023 likely addressed this issue. The remaining 81% did not.

Elvex, an AI infrastructure company, frames the gap similarly: the next productivity leap comes from transforming one person's AI discoveries into shared infrastructure. This requires current company knowledge, repeatable workflows, distribution of improvements, and, crucially, ownership, testing, and maintenance.

Kramer's Four-Step Setup

Kramer's framework consists of four deceptively simple steps.

1. Assign an owner. This person builds and maintains the system, reviews contributions, and helps the team adopt it. Daniel Kravtsov bluntly stated: "The owner role carried more weight than the tooling." Without a designated individual responsible for the shared system, the "team brain" concept fails when the creator of the initial workflow departs.

2. Move the team beyond chat. Encourage collaboration in AI environments (Kramer specifically references Claude's Cowork and Code modes), rather than relying solely on single-user prompting.

3. Set up a shared repository and plugin. Team members should automate their work and upload it to a shared location, ensuring everyone benefits from updates. Harinder Singh described implementing this for a 14-person agency: an MCP gateway sits in front of GitHub repos, translating files into live AI context. Marketers can paste a connection token to load all team playbooks without directly accessing Git.

4. Give the AI real context. Start with strategy documents, then connect it to your CRM, call recordings, CMS, and analytics via MCPs (Model Context Protocol). This step often stalls teams. As Romeo noted, "Connecting the tools is easy; deciding what belongs in shared context is the hard part."

The Governance Problem Nobody Wants to Talk About

Speed without coordination creates chaos. Birdeye has noted that multiple AI agents generating copy and creative in parallel can lead to brand inconsistency if they lack shared context and conflict-resolution rules. Gabriel Mangabeira, testing a multiplayer setup, highlighted a key issue: when two people edit the same shared skill, which version prevails? The repository can become disorganized quickly.

Mangabeira's team discovered that a "context owner" is as crucial as a code owner, and these roles should be distinct. In an agency, the account manager or relationship manager can serve as the guardian of context, while a separate role manages the skill library. This creates two operational responsibilities that didn't exist six months ago.

The Humans with AI team advocates for an agent-led but human-approved model: agents prepare governed work packets, and humans approve their release. Samet Ozkale suggests dual-layer permissions and visible agent presence as effective multiplayer patterns. This governance overhead is real work. Without it, teams risk sacrificing brand consistency for speed, a trade-off that can have negative long-term effects.

Where This Lands for Ops Teams

A SaaS marketing leader survey found that the most valued AI use cases were research and competitive intelligence (46%), reporting and performance measurement (44%), and creative/design production (40%). Each of these is cross-functional and benefits from shared context. They deteriorate when multiple individuals prompt the same AI tool with differing assumptions about positioning, ICP, or competitive landscape.

The operational implication is clear: multiplayer AI isn't merely a product decision; it's an infrastructure decision. Someone must own the context layer, skill library, permissions model, and maintenance cadence. That someone is likely you.

Kramer acknowledged the reality: "It's gonna get messy at first, and that's okay." However, the 67-point gap between AI adoption and performance suggests many teams have tolerated messiness for too long. The pressing question is not whether to build shared AI infrastructure, but whether the person responsible for it has been named yet.