Skip to content
Hot eventDeveloping

Airbnb uses cost-per-task to manage coding agents on Bedrock

1 reports1 reporting sourcesUpdated 21 h ago

Start here

AI digest

Airbnb has detailed how it runs AirChat, its internal coding-agent tool, through an LLM gateway on Amazon Bedrock and uses "cost per task"—the list-price cost of all tokens an agent spends to finish one task—to set model defaults and effort ceilings. Airbnb says that after eight new models shipped in about five weeks, roughly two thirds of engineers' coding-agent requests went to those new models by mid-October. Airbnb pairs pre-rollout evaluations with CloudWatch runtime signals on latency, throttling and token spend to revisit those choices. The method relies on model invocation logging, which is off by default and must be turned on in every AWS Region called; activation does not backfill earlier costs. Because the logs can hold full prompts, responses and caller identity, Airbnb advises restricting access and setting a retention period. It also notes that cache reads never count against the quota, so a request can be cheap against quota yet expensive on the bill.

Generated from reports · updated 21 h ago

Report timeline

Follow the reports to see the event from different sides.

November 19
  1. AWS Machine Learning Blog
    How Airbnb uses cost per task to operate coding agents on Amazon Bedrock

    Airbnb runs its internal coding-agent tool AirChat through an LLM gateway on Amazon Bedrock, using cost per task—the list-price cost of all tokens an agent spends to finish one task—to pick model defaults and effort ceilings. After eight new models shipped in about five weeks, roughly two thirds of engineers' coding-agent requests went to those new models by mid-October. Airbnb pairs pre-rollout evals with CloudWatch runtime signals on latency, throttling and token spend to revisit those choices.

Heat for this event

Current, comparable range
5
Peak, comparable range
10November 19 07:00
Change in 24 hours
–

The trend compares the same participants over hours observed continuously, so its range can be narrower than the current heat count. Hover or tap the chart for hourly heat; the arrow keys move through it.