Every developer on the team has the same Cursor license. One of them ships twice as much code. The problem isn't talent - it's configuration.
A recent analysis from Dylan Etkin, CEO of Skills.new (and former Jira lead at Atlassian), pulls together examples from Confluent, Rapid7, Atlassian, Rippling, and Google that all point to the same pattern: organizations buy AI coding tool licenses in bulk, a handful of developers figure out how to use them well, and the rest of the team never catches up.
The gap between power users and everyone else isn't about prompting tricks. It's about the invisible infrastructure that top performers build around their tools - custom rule files like .cursorrules or CLAUDE.md, curated MCP server setups, domain-specific prompts tuned to their codebase. These configurations live in individual dotfiles and never leave one person's machine.
The Numbers Behind the Gap
Confluent found a "power law distribution of effectiveness" across their AI tool users. Rapid7 saw a core group adopt AI deeply while the broader org barely touched it. Rippling's solution was extreme: they built a custom Go service to sync AI configurations across 800+ repositories. Most companies don't have the engineering bandwidth for that.
Google's senior director acknowledged that knowledge transfer for AI tooling remains "brutally hard" even with their resources. If Google can't solve this with internal tooling, a 50-person startup has no chance doing it ad hoc.
Why Current Sharing Methods Fall Short
Three approaches exist today, and all have problems.
Git-based sharing - checking rule files into repos - fragments configurations across dozens of repositories with no synchronization. A team's best Cursor rules sit in one repo while another team reinvents them poorly.
Vendor marketplaces like Cursor's rule directory or GitHub Copilot Spaces keep configurations locked inside one tool's walls. The developer using Claude Code can't benefit from configurations optimized for Cursor, and vice versa.
Manual documentation decays immediately. By the time someone writes a wiki page about their AI workflow, the configurations have already changed.
The missing piece across all three: no governance layer. There's no approval workflow, no security review, no audit trail for the AI configurations that increasingly shape how code gets written.
What This Means for Teams Buying AI Tools
Etkin's company is pitching a specific product to solve this (a centralized registry for AI tool configurations), and that pitch should be evaluated on its own merits. But the underlying observation holds regardless of whether you buy their solution.
The real cost of AI coding tools isn't the per-seat license. It's the organizational learning debt that accumulates when each developer figures things out independently. A $20/month Cursor Pro seat delivers wildly different value depending on whether the developer has spent 40 hours tuning their setup or is still using defaults.
Practical takeaways for teams right now: designate your best AI tool users as internal coaches, not just individual contributors. Share rule files in a central repo even if the synchronization is manual. Run regular sessions where power users screen-share their actual AI workflows, including the configuration files that make them work.
The AI tool vendors will eventually solve the distribution problem - it's too obvious a feature gap to ignore. Until then, the teams that treat AI configuration as shared infrastructure rather than personal preference will pull further ahead.