The productivity divide
Enterprise AI adoption is splitting into two camps, and the gap explains conflicting productivity data. Power users are adopting Claude Code, MCPs, and CLI agents—often without engineering backgrounds. Finance teams especially are moving Excel workflows to Python environments, unlocking capabilities impossible in spreadsheets.
The second group remains limited to basic ChatGPT-style interfaces. The culprit: Microsoft Copilot's enterprise dominance through Office 365 bundling.
Microsoft's Copilot problem
Copilot holds massive enterprise market share but delivers poor results compared to alternatives. Code execution fails with larger files due to aggressive resource limits. The agent features lag far behind CLI coding tools—including Microsoft's own GitHub Copilot.
Telling signal: Microsoft is reportedly deploying Claude Code to internal teams despite owning OpenAI stakes and having free Copilot access. That's a clear verdict on their own product.
For enterprises where Copilot is the only approved AI tool, employees face a choice: accept limitations, risk job security by going around IT policy, or spend months navigating procurement.
Why enterprises are stuck
Three factors combine to block enterprise AI adoption:
Locked-down environments: Most corporate IT prohibits local script execution. Even VBA may be limited by Group Policy.
Legacy systems: Core workflows lack APIs that agents could connect to. No integration points means no automation.
Siloed or outsourced engineering: No internal team exists to build safely sandboxed agent infrastructure. Outsourced teams lack process knowledge to automate effectively.
Security concerns are legitimate—you don't want uncontrolled agents accessing production databases. But the result is enterprises can't deploy the tools driving productivity elsewhere.
The real productivity gap
Smaller companies without this baggage are outpacing enterprise competitors. Example: a non-technical executive recently converted a 30-sheet Excel financial model to Python using Claude Code—essentially one-shot. Once in Python, Claude Code enables Monte Carlo simulations, external data integration, and dashboard building. Tasks that took days in Excel now take hours.
The pattern: real productivity gains come from employees who understand their processes building AI-assisted workflows—not top-down strategies or outsourced development teams.
History suggests this matters. The tools that stick are the ones employees choose, not the ones IT mandates. Enterprises risk writing off AI entirely based on poor Copilot experiences while competitors pull ahead.
The question: how long before the productivity gap becomes impossible to ignore?