https://getagentburn.com/blog/ The independent ledger of what AI agents actually cost: tokens per task, cost per solved task and framework overhead, with per-field provenance for every number. en Mon, 02 Nov 2026 00:00:00 +0000 AgentBurn scripts/gen-feed.mjs https://getagentburn.com/blog/agent-framework-token-overhead/ https://getagentburn.com/blog/agent-framework-token-overhead/ The same tool-use task costs 379 tokens per run in AutoGen and 4,382 in CrewAI — an 11.6x orchestration tax measured on matched runs. What causes it, when it matters at fleet scale, and how to measure your own stack. Mon, 02 Nov 2026 00:00:00 +0000 Agent frameworks https://getagentburn.com/blog/claude-code-vs-codex-cost/ https://getagentburn.com/blog/claude-code-vs-codex-cost/ Two coding agents, the same benchmark family, a 10x token spread. What the published rows actually compare — and which differences come from the harness rather than the model. Mon, 02 Nov 2026 00:00:00 +0000 Agent economics https://getagentburn.com/blog/monthly-agent-bill-formula/ https://getagentburn.com/blog/monthly-agent-bill-formula/ Runs per day × tokens per task × blended price, plus retries and success rate. A worked estimate across three agent profiles, and the places where the model is weakest. Mon, 02 Nov 2026 00:00:00 +0000 Agent economics https://getagentburn.com/blog/agent-retry-overhead/ https://getagentburn.com/blog/agent-retry-overhead/ Debug-and-retry loops are the largest single token line in the Claude Code workflow split. How to measure retry overhead and what it does to cost per solved task. Mon, 02 Nov 2026 00:00:00 +0000 Agent economics https://getagentburn.com/blog/prompt-caching-agent-cost/ https://getagentburn.com/blog/prompt-caching-agent-cost/ Cached input is priced at a fraction of fresh input, but only if the prefix is stable. A sensitivity model across the 22 models in the dataset. Mon, 02 Nov 2026 00:00:00 +0000 Agent economics https://getagentburn.com/blog/agent-pricing-units/ https://getagentburn.com/blog/agent-pricing-units/ Agentforce sells 20 credits per action; Copilot Studio meters 2/5/10 credits; coding agents quote per task. How to normalize the units before comparing vendors. Mon, 02 Nov 2026 00:00:00 +0000 Agent economics https://getagentburn.com/blog/agent-roi-break-even/ https://getagentburn.com/blog/agent-roi-break-even/ Cost per solved task only matters next to what the human alternative costs. A break-even formula with worked examples from the calculator human baseline. Mon, 02 Nov 2026 00:00:00 +0000 Agent economics https://getagentburn.com/blog/read-coding-agent-leaderboard/ https://getagentburn.com/blog/read-coding-agent-leaderboard/ SWE-bench Verified, Terminal-Bench 3.0 and the AA Coding Agent Index measure different things on different task sets. A checklist for reading the rows. Mon, 02 Nov 2026 00:00:00 +0000 Agent benchmarks https://getagentburn.com/blog/when-to-instrument-agent-telemetry/ https://getagentburn.com/blog/when-to-instrument-agent-telemetry/ Published benchmarks describe products, not your stack. What telemetry adds, what it costs to instrument, and the minimum viable measurement set. Mon, 02 Nov 2026 00:00:00 +0000 Measurement https://getagentburn.com/blog/llms-txt-ai-crawler-policy/ https://getagentburn.com/blog/llms-txt-ai-crawler-policy/ What we publish for AI crawlers — llms.txt, llms-full.txt, ai.txt, robots content signals — and what actually gets read. A practical checklist for tool sites. Mon, 02 Nov 2026 00:00:00 +0000 GEO and AI search https://getagentburn.com/blog/agent-cost-data-gaps/ https://getagentburn.com/blog/agent-cost-data-gaps/ Only three projects in the dataset publish tokens per task. The provenance tiers that keep an agent cost database honest, and what to do when a vendor publishes nothing. Mon, 02 Nov 2026 00:00:00 +0000 Measurement https://getagentburn.com/blog/cost-per-solved-task/ https://getagentburn.com/blog/cost-per-solved-task/ Cost per task flatters agents that fail. Divide by success rate and the benchmark leaderboard changes order. Worked examples from SWE-bench Verified rows and the formula AgentBurn uses. Sat, 31 Oct 2026 00:00:00 +0000 Agent economics https://getagentburn.com/blog/ai-agent-cost-per-task/ https://getagentburn.com/blog/ai-agent-cost-per-task/ Agent prices are quoted per seat, per credit or per million tokens. Tokens and cost per task are the only units that survive contact with production — here is what the published numbers actually show. Thu, 29 Oct 2026 00:00:00 +0000 Agent economics