I Believe Everything My Agents Tell Me: Three Levels of Review — Corey Gallon, Rexmore

A demand letter passed its drafting agent's own checks, yet two independent reviewer agents found that the supposedly expired contract had renewed less than three weeks earlier. Sending it could have left the business owing thousands of dollars. Corey Gallon uses that case to explain a three level review method for agents that overstate completed work, misread facts or draw the wrong conclusions. His analysis of his own 2026 sessions found a consequential error in about four out of ten sessions; that is a description of his sample, not an error rate for all agents or for the final reviewed outputs. Level one requires a verified artifact with a fingerprint, or an explicit account of what is missing. Level two has independent agents extract the same material against a control total, then settles every difference against the source. For a year of scanned accounting statements, that process found 56 differences and produced records that balanced to the penny. Level three combines a self audit, cold reviewer agents with reasons to challenge the work and fresh reviewers for the revision. A memory leak investigation exposed a flawed reproduction and an overstated fix before the maintainer merged the corrected change. Gallon reports the time and token costs for these cases and releases the method as agent skills. The evidence supports a concrete checking workflow, rather than a guarantee that review eliminates every error. Speaker info: - https://www.linkedin.com/in/coreygallon/ - https://gallon.me/i-believe-everything-my-agents-tell-me - https://github.com/captivus/agent-review Timestamps: 00:00 - Three recurring agent failure modes 1:22 - Level one: verify the deliverable 2:03 - Level two: independent extraction and source checks 2:44 - Level three: adversarial and fresh review 4:02 - A memory leak and an overstated fix 6:57 - Accounting that balances to the penny 8:53 - A demand letter and a missed renewal 11:35 - Packaging the review method as agent skills