signalDEV Community2026-09-28
Your cost dashboard can't tell you which agent ran up the bill
Cloud cost monitors lag by up to 24 hours, and provider invoices cannot decompose spending into specific agent runs, retries, or tool calls. OpenTelemetry's GenAI conventions define no cost attribute, and token counters are only a proxy because cache reads cost 0.1x and writes 1.25x the uncached rate. A reported month-long run of about 100 agent instances billed $1,305,088.81 across 603 billion tokens, leaving the cost source unexplained.
- for who
- Engineers and teams building AI agents who need per-call cost attribution.
- why now
- With the OpenTelemetry GenAI cost attribute proposal still open, per-span pricing is the timely fix.
- what changes
- They can move from invoice-level alarms to runtime alerts tied to specific agent runs, enabling real-time cost control.
- to do
- Price each span individually, roll up costs from child inference spans, and set runtime alarms instead of waiting for invoices.
key points
- Provider invoices show total spend but cannot attribute to agent runs, retries, or tool calls
- OpenTelemetry spec has token counters but no cost attributes; proposal PR #443 open since Aug 2026
- Cache pricing varies: read 0.1x, write 1.25x, so token counts are not costs
#agent cost tracking#opentelemetry#token pricing#cost attribution
score
score 8 out of 10. 0-10: how dense the facts are, multiplied by how much you can do with them after reading. 8+ means the topic's evidence bar is met: benchmarks and availability for a new model, amount and investors for a funding round, revenue figures for a solo-money story. Below 5 an item does not enter the digest. A press release scores 3 or less, a reprint loses 2, anything older than 14 days loses 1, a headline that misleads loses 3.
read the source