Engineering leadership · Book notes
The Art of Action says brief the intent, even to an agent
Stephen Bungay argues that adding detail to a plan widens the gap it was meant to close, and I think the same reflex shows up when you delegate engineering work to software agents.
When a plan fails to translate into the right action, our instinct is to add more detail. In The Art of Action (2011), Stephen Bungay identifies 3 gaps that explain this instinct: the knowledge gap (between plans and outcomes), the alignment gap (between plans and actions), and the effects gap (between actions and outcomes). We try to close the knowledge gap by adding more data, the alignment gap by issuing more detailed instructions, and the effects gap by introducing tighter controls. But, he says, adding more detail usually makes the gap you’re trying to close even bigger.
Bungay’s solution to this is to use a technique from the 19th-century Prussian army called mission command, or directed opportunism. You limit your direction to intent.. what must happen and why.. and each level comes up with its own route and reports back to the person who briefed them (a backbrief). If reality doesn’t match the plan, people are free to adjust their actions as long as they’re consistent with the intent. But you have to keep the brief short: “What has not been made simple cannot be made clear and what is not clear will not get done.”
I see the same reflex when I delegate work (writing code or changing infrastructure) to software agents. When the agent gets the task wrong, the natural reaction is to make the prompt longer, add more rules, or grow the list of things it should never do. That’s the alignment-gap reflex, just in a new setting. But as with human teams, making the list of instructions longer rarely fixes an agent that misunderstood the task.. it’s the same over-specification Bungay warns about. Telling the agent what needs to happen and why usually works better. Constantin Gonzalez makes a similar connection in his essay on delegating to models, where he describes prompting as delegation and lists commander’s intent as one of several frameworks that point in the same direction. His advice: tell the model what success looks like and why it matters.
The backbrief is where I land as the most useful part of Bungay’s approach to carry over to agents. Asking the agent to explain its plan before it makes any change is the one point where you see how your intent was read. A misread intent is cheapest to catch at the plan stage, before the effects gap opens on a real change that would need undoing.
I’m less sure about the freedom to adjust. If an officer adapts to the situation within the intent, he shares training and doctrine with the briefer and is accountable for his decision. The agent fills in the gaps with a confident guess. Does intent work without shared doctrine, or must the backbrief carry it alone?