Journal

Everything we have published.

Field notes, guides, our own primary research and the benchmark editions behind it. Filter to what you actually want to read.

The answers you discard are still on the bill

A router that fans one request out to several models bills every arm. What that does to cost per run when only one candidate is kept.

4 September 2026

Topics

Blog2 September 20265 min
A cost figure you computed is not a receipt
Tutorial2 September 202615 min
Per-agent cost in a LangGraph supervisor loop
Blog22 August 20265 min
The experiment that measures a saving has its own bill
Blog21 August 20265 min
Cost per run depends on what counts as one run
Blog18 August 20265 min
A model has no single price per token
Blog17 August 20265 min
What a run costs depends on how the capacity was bought
Blog16 August 20265 min
A budget check cannot price the call it allows
Tutorial12 August 202614 min
Per-role cost in a LangChain coding agent
Blog10 August 20265 min
The identity a budget is keyed to decides what it stops
Blog9 August 20265 min
A daily cost rollup cannot answer per-run questions
Blog8 August 20265 min
A configured cost control is not an enforced one
Blog7 August 20265 min
Agent cost includes the calls you never wrote
Blog6 August 20265 min
Run quotas and concurrency caps do not bound spend
Tutorial6 August 202615 min
Cost per role in a LangGraph AI scientist that fans out
Blog5 August 20265 min
Tracking tokens is not the same as tracking cost
Tutorial5 August 202612 min
Cost per agent in a LangChain incident-brief pipeline
Blog4 August 20266 min
A cheaper model is not automatically a cheaper run
Blog3 August 20265 min
Finding which sub-agent burned the tokens in one run
Guide1 August 2026
How to evaluate an LLM cost tool
Guide1 August 2026
Find the agents already in your codebase
Guide1 August 2026
Gateway vs observability vs governance
Evidence1 August 2026
Every question we have been asked
Benchmark31 July 2026
We scanned 133 public agent repos. Most had a cost defect.
Evidence31 July 2026
The four cost defects
Benchmark31 July 2026
All benchmark editions
Blog31 July 20265 min
Provider spend limits are alerts, not limits
Blog31 July 20264 min
What a runaway agent actually costs

Everything we have published.

The answers you discard are still on the bill

A cost figure you computed is not a receipt

Per-agent cost in a LangGraph supervisor loop

The experiment that measures a saving has its own bill

Cost per run depends on what counts as one run

A model has no single price per token

What a run costs depends on how the capacity was bought

A budget check cannot price the call it allows

Per-role cost in a LangChain coding agent

The identity a budget is keyed to decides what it stops

A daily cost rollup cannot answer per-run questions

A configured cost control is not an enforced one

Agent cost includes the calls you never wrote

Run quotas and concurrency caps do not bound spend

Cost per role in a LangGraph AI scientist that fans out

Tracking tokens is not the same as tracking cost

Cost per agent in a LangChain incident-brief pipeline

A cheaper model is not automatically a cheaper run

Finding which sub-agent burned the tokens in one run

How to evaluate an LLM cost tool

Find the agents already in your codebase

Gateway vs observability vs governance

Every question we have been asked

We scanned 133 public agent repos. Most had a cost defect.

The four cost defects

All benchmark editions

Provider spend limits are alerts, not limits

What a runaway agent actually costs