You change the model, the prompt or how often the agent's context is cleared, rerun, and every test is green. That tells you less than it seems. In 36 runs of a coding agent under six context-clearing policies, 4,086 tests passed with zero failures while cost differed by 33.5% between the most and least expensive of the six condition averages. The agent wrote its own tests in every run, from 97 to 127 per run, so the runs had no shared checks to compare. This episode explains why a test suite can be blind to the very difference you are testing, why the tempting "longer sessions did less work" reading did not hold up, and the proposal that came out of a W3C Agent Conformance group discussion: every result set should say whether any check changed state (some, none or incomparable) and which ones did. Because a record that cannot say incomparable will say none, and a reader will hear equivalent. Paper on Qeios: https://doi.org/10.32388/0BV3Z8

Full transcript of this episode: arsentev.ai/podcast

An AI content engine is a system where AI writes the scripts, AI avatars appear on camera, a synthetic voice talks and software edits and publishes. This episode walks through a working one: built by one person with no technical background, 1,497 videos across 67 channels in 101 days, 52 cents per 1,000 views. What a human studio or influencers would cost for the same output, why organic reach grows with a delay, a forecast for 100 AI presenters, and four steps for a business. Technical report with every calculation: https://doi.org/10.5281/zenodo.22802228

Full transcript of this episode: arsentev.ai/podcast

Correction, 15 September 2026: the audio and the transcript below use the prices of report versions 1.0 and 1.1, which were wrong. At the correct prices, a run costs $2.17 when the context is cleared after every task, $1.62 at every third task and $1.78 when it is never cleared (not $2.68, $2.14 and $2.47). Clearing after every task costs 33.5% more than the best policy (not 25%). Never clearing costs 9.7% more (not 16%), and that difference is not statistically significant (p = 0.046; 0.14 after adjustment for multiple comparisons). A cache write costs 20 times a cache read, not twelve and a half. Corrected report: https://doi.org/10.5281/zenodo.22759217

Thirty-six runs, six session policies, and an answer that matches neither of the two popular rules

Full transcript of this episode: arsentev.ai/podcast

A long episode for anyone who has never opened the black window full of text and is sure it isn't for them

Full transcript of this episode: arsentev.ai/podcast

How to cut your token spend in Claude Code, or any other AI, several times over. Why the model rereads the whole conversation on every reply, what the context actually is, and why a bloated context makes the model dumber. One command, one habit of writing to a file, and one rule: one task, one chat, closed.

Full transcript of this episode: arsentev.ai/podcast