Three separate issues hit Cursor users this month. And the timing couldn't be worse.

First, the bug reports. Cursor has been reverting code changes — users make edits, the AI "helps," and suddenly their work is overwritten with older versions. This isn't a one-off glitch. The root causes point to three distinct issues: context window overflow dropping earlier changes, aggressive "diff-matching" that replaces instead of appending, and a tab completion bug that fires when it shouldn't.

Second, the pricing erosion. Remember when Pro users got 500 "fast" requests per month? It's now 225. Anthropic raised API prices, Cursor passed the cost through, and users got half the runway they expected.

Third, the trust problem. When your coding tool might silently undo your work, you start second-guessing every suggestion. That's not a bug — that's a relationship issue.

Cursor isn't alone in feeling the squeeze. Anthropic's API pricing changes rippled through the entire ecosystem. But while Cursor absorbed the hit and cut user allocations, Windsurf went the other direction — they added GPT-5.4 with promotional pricing and made GPT-5-Codex free for paid users.

The AI IDE market just split into two camps: tools that optimize for price stability and tools that optimize for feature velocity. Cursor chose neither. They raised prices and let quality slip.

If you're staying on Cursor, practice defensive coding: git commit early, git commit often. Review AI suggestions before accepting. And set up a pre-commit hook that warns you about large diffs.

If you're shopping alternatives, Windsurf at $15/month (vs Cursor's $20) is the obvious comparison. Claude Code's 1M token context window is now generally available on Max plans if you want to skip the IDE middleman entirely.

Anthropic made a quiet but significant change: the 1M token context window for Claude Opus 4.6 and Sonnet 4.6 is now GA at standard pricing. No more 200k+ surcharges. If you're building agents that need to reason over entire codebases, the math just got a lot friendlier.

Why it matters: This is Anthropic competing on value, not just capability. OpenAI still charges premium rates for extended context. Google's Gemini 1M window has quality tradeoffs. Anthropic is betting that "same price, bigger brain" wins the agent developer market.

Replit's Agent 3 burns through credits six times faster than its predecessor. The output quality is higher — legitimately better code generation — but the economics are brutal for indie devs.

At the same time, Replit just raised $400M at a $9B valuation, up from $3B six months ago. They're not optimizing for hobbyists anymore. This is enterprise positioning.

The takeaway: If you're on Replit's free tier, Agent 3 will drain your credits fast. Budget accordingly or downgrade to Agent 2 for prototyping.

OpenAI is changing how they bill containers — and the deadline is March 31st. If you're running persistent compute environments (like AI agents with always-on processes), your bill structure is about to shift.

The change itself isn't dramatic, but the timing is tight. Two weeks to audit and adjust.

Action item: Check your OpenAI billing dashboard and review any "container" or "persistent compute" line items before month-end.

Claude Code v2.1.76 shipped push-to-talk voice control. Hold spacebar, speak, release to send. It's not "always listening" — it's a controlled, opt-in interaction model.

Combined with the /loop command for recurring tasks and the 1M context window on Max plans, Claude Code is quietly becoming the power user's choice for agentic coding.

Cognition (the Devin people) acquired Windsurf in December, and Wave 12 shows what they're doing with it. The big add: Devin's DeepWiki for codebase documentation and planning mode that automatically triggers when tasks get complex.

Arena Mode is the sleeper feature. Run two AI models side-by-side on the same task, vote on which performs better, and build your own "which model works for me" leaderboard. If you've ever wondered whether Claude or GPT handles your specific codebase better, this is how you find out.

One concern: Windsurf's future is uncertain. Will it stay a standalone IDE or merge into Devin? Will pricing change? The Cognition acquisition was complex (Google got a separate licensing deal, OpenAI was in the bidding). If you're building workflows around Windsurf, factor in some platform risk.

Lots of changes happening in rapid fire. Subscribe to stay up to date on the changes happening everyday.

— Trish, Drift Intel

StackDrift · Unsubscribe