ANDREJ KARPATHY JUST SAID SOMETHING THAT SHOULD MAKE EVERY AI DEVELOPER UNCOMFORTABLE. "90% of your AI coding bill is paying for context you didn't need to send." You are not paying for intelligence. You are paying for waste. Here are the 10 things senior AI engineers stopped doing that cut their bills by 60 to 80%: Auto-loading 50 files for a 30-line fix. $1.20 per turn on tokens you never read. 80% input waste every single session. Running Opus on lint, format, and rename tasks. $0.60 for what Haiku nails at $0.02. 30x overpay on cleanup work. Tool call loops that re-send the full repo on every retry. 5x context cost per agentic flow. Fixing this one thing cuts 30 to 50% of your total bill. Defaulting to Sonnet in 2026. Kimi 2.6 matches it on most coding tasks at one sixth the cost. You are leaving 60 to 70% on the table every session. Streaming responses on stable-prefix workflows. Kills your prompt cache. You pay 10x for tokens that should have cost cents. Just in case file includes. 80,000-token prompts that should be 3,000. Context bloat is the silent budget killer. Per-session knowledge rebuilding. 10 minutes writing a SKILL.md once versus paying agents to re-figure out your environment every run. $4 versus $0.30 per execution. Single-model setups. Premium tier on every task is the most expensive mistake in AI coding right now. Asking 10 small questions one at a time. 10 separate input prefix charges versus one batched call. 70 to 90% savings on routine workflows. Paying for Claude Pro plus ChatGPT Plus plus Cursor Pro simultaneously. You seriously use one. The other two are habit not utility. What actually compounds: Context discipline. Grep before fetching. Always. Prompt caching on every stable prefix. Multi-model routing. Kimi 2.6 as default. Opus for the 10% that needs it. Graduated skills via SKILL.md files. The routing mindset. Right model for the right task. In 12 months the gap between developers shipping on $200 a month and $4,000 a month will not be skill. It will be how well they route. Bookmark this. Study it. Follow @cyrilXBT for every AI engineering insight that compounds your output and cuts your costs.(CyrilXBT)