If your agent learned anything, why does Run 10 cost the same as Run 1?
Summary
Critique of AI agent token consumption; proposes Return on Token Investment (ROTI) as a metric for efficiency, noting that most agents do not reduce token usage over time.
Similar Articles
Price per token is the wrong cost metric for agents
The article argues that token price is an inadequate cost metric for AI agents, proposing that effective cost should be measured by successful runs with validation, and discusses routing and failure strategies.
Subagents Account for Most Token Costs in Long Agent Runs: Fixes That Cut Usage 70 to 90 Percent in Practice
The article analyzes a 2026 paper by Bai et al. showing that subagents and context bloat cause token costs in long agent runs to be ~1000x higher than chat, and presents three practical fixes (PLAN.md, read budget, out-of-band notes) that reduce token usage by 70-90%.
The most expensive part of running AI agents isn't the tokens. It's the time figuring out why they did something.
Building AI agents reveals that the major cost is debugging—spending weeks chasing issues like upstream API changes—not just token or model inference costs.
Are we optimizing AI agent token costs way faster than we’re thinking about agent security?
The author observes that the AI agent ecosystem is heavily prioritizing token cost optimization while neglecting security, and raises important questions about agent authorization, identity boundaries, and permission inheritance as agents become more autonomous and spawn sub-processes.
How are you actually saving cost on your agent systems?
The article discusses the challenges of cost optimization and FinOps for AI agent systems, highlighting issues with unpredictable token bills, lack of granular attribution tools, and strategies like caching and hard caps.