I discovered Cursor was burning $5 per request due to excessive cache reads. After switching to Aider for specific workflows, I cut that to $0.17. That's a 97% reduction that scales to thousands of dollars monthly.
Hi, my name is Tom Smykowski, I'm a staff full-stack engineer. I build and scale SaaS platforms to millions of users, working end-to-end from system architecture to frontend to mobile. On this blog I share practical insights on AI-assisted development and cost optimization strategies.
The problem started when I noticed my AI coding bills climbing faster than my actual output. After some investigation, I found that Cursor was reading cache excessively on certain operations. I reported the bug to the Cursor team, and they're investigating. But even without bugs, there's a fundamental issue with how modern AI coding tools handle context.
Cursor, Claude Code, Windsurf, and similar tools are built for broad appeal. They want to serve professional engineers while also attracting the general public. This means they optimize for convenience, not cost efficiency. You get minimal control over context management, cache behavior, and token usage.
The Hidden Cost of "Knowing Everything"
Engineering work often requires broad project knowledge. The AI needs to understand file structures, dependencies, and relationships. Tools that "know everything" about your project seem ideal for this. But that knowledge comes with a cost.
When you're doing exploratory development or complex refactoring, paying for full context makes sense. But not every task requires the AI to understand your entire codebase. Consider what I call "AI scripts" - workflows that combine deterministic and non-deterministic processes. These are perfect candidates for optimization.