← Back to feed

Disney is dropping GitHub Copilot and multiple other AI coding tools from its US engineering operation in August, replacing them with OpenAI Codex. Tech leadership communicated the decision to select staff this week. The move is one of the highest-profile enterprise Copilot exits since Microsoft restructured Copilot pricing to token-based AI credits in June.

The same week, Cursor — the IDE that displaced Copilot across much of the startup market — quietly removed cost information from its usage page and CSV export, triggering a Hacker News thread from enterprise buyers who lost visibility into actual token spend.

Both events point to the same pressure: enterprise AI coding is consolidating around fewer tools, with different winners at different segments.

The Copilot-to-Codex Switch

Disney’s decision to standardize on Codex is a concrete data point in a market where benchmark performance matters less than procurement trust and total cost predictability.

GitHub Copilot’s June pricing restructuring — abandoning flat-rate subscriptions for credit-metered usage across all tiers — changed the enterprise calculus. Copilot is now billed by consumption, the same model as every other token-based AI tool. The differentiation it had as a flat-fee IDE add-on is gone.

Codex, by contrast, operates on a defined $200/month Pro tier with a consumption ceiling, plus pay-as-you-go above it. For enterprise procurement teams managing engineering headcount costs, a defined ceiling is easier to budget than open-ended credit consumption.

On capability metrics, Codex built on GPT-5.6 Sol is competitive. The Codex CLI + GPT-5.5 combination sits at 82.2% accuracy on Terminal-Bench 2.1 — the same harness where Cursor Composer 2.5 scores at the $0.07/task level and Claude Code runs on Anthropic’s Fable 5. GPT-5.6 Sol xHigh Codex-Harness entered Arena’s Code leaderboard at #2 in July.

Disney operating across streaming infrastructure, theme park systems, and media production pipelines means its engineering footprint is diverse enough that this isn’t a single team’s preference — it’s a deliberate enterprise consolidation.

Cursor’s Transparency Problem

Cursor hit $2B ARR earlier this year and raised at a $50B valuation. It won the startup IDE market decisively, with Claude Code as its main remaining competitor at the frontier tier.

But Cursor is closing an acquisition by SpaceX — in the same transaction context that absorbed xAI’s Grok into SpaceX AI. TechCrunch reported the company making its India push “weeks before its expected acquisition closes.” The deal structure is all-stock, reportedly valuing Cursor at approximately $60B.

Removing cost data from the usage page in the final weeks before an acquisition close has a straightforward interpretation: Cursor is optimizing growth metrics for the deal period. Enterprise users who relied on per-seat cost tracking for internal chargebacks now have no automated path to that data.

The Hacker News thread drew immediate pushback from engineering teams who described the removal as a trust breach. For enterprise buyers evaluating whether to standardize on Cursor long-term, opacity on costs is a structural concern, not a minor UX issue.

What Enterprise Buyers Now See

The enterprise AI coding market in August 2026 has effectively stratified:

Agentic premium tier: Claude Code (Anthropic), Codex (OpenAI/SpaceX), Cursor (Anysphere/SpaceX). These compete on benchmark performance and agent loop depth. Fable 5 drives Claude Code; GPT-5.6 Sol drives Codex; Cursor uses both via model picker.

IDE-embedded tier: GitHub Copilot (Microsoft, credit-metered), JetBrains AI (Mellum2 17B backbone), VS Code Copilot. These compete on integration depth and accessibility for teams that don’t run agentic workflows.

Open-weight/self-hosted tier: Kimi K2.7-Code (Moonshot AI, Apache 2.0), Inkling-Small (Thinking Machines, Apache 2.0), GLM-5.2 (Z.ai). These compete on cost and deployment control — Kimi K2.7-Code is the first open-weight model in GitHub Copilot’s model picker.

Disney moving from Copilot (embedded tier) to Codex (agentic premium) reflects the same shift visible across enterprise procurement: as AI coding tools add agent loops and multi-step task completion, the flat IDE-plugin model loses ground to dedicated coding agent platforms.

Key Numbers

• Disney: Dropping Copilot and other AI coding tools in August, moving to Codex
• Codex CLI + GPT-5.5: 82.2% on Terminal-Bench 2.1
• Codex Pro pricing: $200/month with consumption ceiling
• Cursor valuation at last raise: $50B; SpaceX acquisition reported at ~$60B all-stock
• GitHub Copilot pricing: switched to credit-metered June 1 across all plans
• Kimi K2.7-Code: first open-weight model in GitHub Copilot’s model picker, $4/M output