ACI-055: SAAC to MAAC: The Step Up

Thesis

Single-agent agentic coding (SAAC) measures how autonomous one agent is; multi-agent agentic coding (MAAC) measures how many agents one human runs at once. Stepping up from one to the other changes the human's job from doing the work to arbitrating it, and its entry condition is SAAC fluency. A fleet, once running, is a standing way of working with a running cost you can measure, so budget that cost before you start and keep an off switch.

Story

From its first commit in September 2025 (Lamplight@17aa789ec, 2025-09-15) until the board in May, I built Lamplight, a narrative-engine project, one agent at a time: scope a steel thread, delegate it to one Claude Code session, verify the result, commit. Then the work outgrew one agent. On 18 May I bootstrapped a whiteboard, a "live coordination channel for concurrent Claude Code sessions", split by streams of work (Lamplight@fce7c0b7d, 2026-05-18). The lanes came first and the roles were named later, as ACI-056: Composing a Fleet traces. By mid-June the fleet had five nodes: me as hypervisor (hv), Control Claude (cc) for the engine, Interface Claude (ic) for the user-facing apps, Validation Claude (vc) for independent verification and Author Claude (ac) for content (Lamplight@4d0014e2a, 2026-06-15).