What This Is

EVE Online, the space-MMORPG launched in 2003, finally kicked off its Python 3 migration this week—uncovering a legacy debt of 2.4 million lines of code and roughly 20,000 behavioral differences requiring manual review. Its server stack runs on Stackless Python 2.7 (a Python interpreter modified for high concurrency), last upgraded back in 2010. The workflow runs futurize (a script that auto-converts Python 2 syntax to Python 3) across the codebase first, then humans audit the semantic diffs. Notably, CCP Games already replaced Stackless with its own open-source scheduler in last year's new title EVE Frontier—meaning the technical path is proven, and the EVE Online main server is taking the "slow and steady" route.

Industry View

This is the most vivid specimen of "legacy system modernization" we've seen. We've noticed the discourse splitting into two camps:

The bullish camp argues that 20,000 diffs are exactly the kind of bulk semantic-reconciliation work AI coding assistants (Cursor, Claude Code, Copilot) excel at—orders of magnitude faster than humans. Zoom out further: the world still runs countless mission-critical systems on Python 2, Java 8, and COBOL. This is the next main battleground for AI coding tools.

The skeptical camp counters: the very fact that "script-batch conversion + manual review" is still a step exposes the problem—automation handles syntax, but domain logic still needs humans. In Python 2, 1 / 2 = 0; in Python 3, 1 / 2 = 0.5. Drop that difference into a game's numerical system and you may have 23 years of latent bugs. AI doesn't make this painless. Going deeper: EVE Online choosing "publish the plan, then move slowly" already counts as responsible. Many enterprises won't even admit their core systems have this problem, let alone fix it.

Impact on Regular People

  • Enterprise IT: Legacy system refactoring will shift from "maintenance budget" to "AI-assisted refactoring" line items—reshaping enterprise software spend over the next year or two.
  • Individual careers: People who understand both business and code are worth more than pure feature builders—AI can fix syntax, not business semantics.
  • Consumer market: Players won't notice short-term change; if CCP uses this to rebuild the underlying scheduler, longer-term new features may ship faster.