The AI Coding Tool Bait-and-Switch: Why Flat Rates Are Dead

If you ask an AI chatbot about AI coding subscriptions, it will give you a generic comparison of features. That entirely misses the urgent financial crisis hitting engineering teams right now. The flat-rate subscription model for AI coding assistants is dead, and the vendors are executing a coordinated bait-and-switch that will make your Q1 2026 procurement budget obsolete.

This week, Cursor became the latest domino to fall, but the strategy was written months ago by Anthropic and GitHub. The real danger isn't just the rising cost of tokens—it's the proprietary lock-in mechanisms these vendors have built into your developers' IDEs.

---

Cursor's "Auto" Bait and Switch

The reality is that flat-rate AI subscriptions were always a customer acquisition tactic, not a sustainable business model. Cursor's sudden pricing update, leaked via subscriber emails with no public blog post or changelog, proves this.

Starting August 24, Cursor's "Auto" feature—which dynamically routes prompts to the best model—will no longer be a flat rate. It will be metered based on the specific model each request hits. If Auto routes your query to a heavy-duty reasoning model, your per-request cost spikes. By Cursor's own admission in the email, "most requests will draw at a higher rate than they do today." This is a silent contract renegotiation forced onto users who built their workflows around a predictable flat fee.

The Anthropic and GitHub Copilot Blueprint

Data indicates that the industry has universally converged on metered billing to mask the true cost of heavy agentic workflows. Cursor is just following the playbook.

Anthropic doesn't disguise it with dynamic routing; they simply cut you off. Claude Code users hit weekly rate limits that force enterprise clients onto raw API pricing from the very first token. GitHub Copilot took an even harder line on June 1, 2026, retiring annual plans entirely and shifting to strict token billing drawn on input, output, and cached tokens at API rates. The era of "unlimited AI" for $20 a month is officially over.

The $0.25 Tollbooth: Why BYOK Doesn't Save You

The most insidious aspect of the current market isn't the API costs, but vendors charging a toll on infrastructure you already own. Enterprise teams thought they could beat the metered pricing trap by using Bring Your Own Key (BYOK).

If you plug your own Anthropic or OpenAI API key into Cursor (Teams or Enterprise tiers), you bypass their model costs, but Cursor still charges you $0.25 per million tokens. You are paying a premium to use your own key inside their editor. This proves these companies are no longer selling AI models; they are selling access to your own developers' workflow and taxing the data pipeline.

The VS Code Fork Trap

Procurement teams fail to realize that migrating AI coding tools isn't an API swap; it is a developer workflow disruption. Tools like Cursor, Kiro, and Windsurf aren't standalone apps—they are VS Code forks.

Your custom rules, skills, MCP configurations, and prompt libraries don't live in your repository in a portable format. They live in the vendor's team marketplace. If you decide to switch providers tomorrow because of a price hike, you aren't just changing an API key. You are forcing every developer to abandon their configured IDE environment, migrate their custom syntax, and learn a new fork's quirks. This friction is exactly what vendors are counting on to keep you paying the toll.

The Pass-Through Alternative

Arriving at metered pricing is an economic reality, but arriving without vendor leverage is a procurement failure. The solution isn't to find another flat-rate plan—it's to eliminate the middleman's markup.

Gateways like Kilo represent the counter-movement. They offer pass-through billing at exact provider rates across 500+ models, with zero markup. If Anthropic raises Opus rates tomorrow, Kilo users pay the difference, but they aren't paying a $0.25/million tax to a wrapper, and they aren't locked into a specific editor. You point your agent at a different model instantly, without changing your IDE.