The Architect
Plan before acting
The one who has already solved the problem before the first line is written. The Architect works in the densest part of the process: the part where ten viable approaches exist and nine must be discarded. They compress. They cross-cut the design space, pressure-testing constraints against each other until what remains is structure so clean that execution becomes inevitable. This is the hardest cognitive work in any project, and the least visible, because by the time anyone else sees the output, every decision has already been made.
"Beautiful plan. Ship date was yesterday."
Claude in plan mode, challenged on its assumptions, pushed to justify its decomposition, argued with until every ambiguity resolves. That is the Architect’s native territory. The result is a spec so precisely resolved that implementation is mechanical. One shot. No rework. The AI performs because the Architect eliminated every point where it would have had to guess. The discipline to resolve when every tool is urging you to execute is the Architect’s rarest contribution, and the one least visible to anyone who wasn’t in the room when the decisions were made.
Shadow: Procrastineering
Engineering as procrastination. The plan is on version twelve. Each version is genuinely better than the last. Each version also cost another hour that execution didn’t get. The Architect is doing real cognitive work; that’s what makes procrastineering invisible from the inside. It feels productive. The sprint board knows the difference.
Transition States
Procrastineering → Context Gilding
The Architect whose plan is done but who can’t start executing discovers a new reason to delay: the context isn’t right yet. The plan requires perfect inputs. Perfect inputs require more curation. The planning paralysis has found a logistics alibi.
"The plan needs better context. The context needs a plan for how to load it. You’ve built an infinite loop and called it preparation."
← Developing from Proof Spiraling
The Sentinel who starts planning their verification discovers they enjoy the planning more than the checking. They’re still not shipping, but now they have a beautiful system design for how they’ll verify. Two shadows feeding each other.
"You’ve replaced checking everything with planning how to check everything. Progress."
← Developing from Rube Goldberging
The Watchmaker whose automation keeps breaking decides the problem is inadequate planning. They stop building and start designing. The design for the automation framework becomes the new project. The original task that needed automating is now three abstraction layers below the current conversation.
"You were automating the wrong thing, so now you’re planning how to automate the right thing. At no point in this story does anyone do the thing."