Essays on measuring AI-assisted engineering and on the cost of supervising it, originally published at charlessieg.com and republished here because they are the argument the leverage engagement rests on.
Leverage is a small input controlling a large output. Applied to AI-assisted engineering, it is the ratio of the senior-engineer hours a task would have taken to the minutes the agent took, and it is the number behind everything Vantalect measures.
Originally at charlessieg.com
The first leverage article drew hostile mail. This one takes the criticism seriously: what the human estimates mean, why the METR study does not apply, the supervisory factor, and the compounding reasons one operator's numbers run higher than most.
Originally at charlessieg.com
Supervising AI agents is a job of hundreds of accept-or-reject decisions a day, and decision quality degrades by afternoon. What the cognitive-science literature says, why agentic coding is uniquely demanding, and the routines that compensate, including teaching the agent to adjust to your fatigue.
Originally at charlessieg.com
Whether you need a compliance audit, a cloud migration plan, or an AI pipeline, we'll scope it and give you a clear path forward.
Remote engagements nationwide. Serving clients since 2020.