AGENT SIZES
Size determines scope, timeline, price and risk — from one spreadsheet task to an entire legacy application.
P
One task currently living in a spreadsheet, script or someone's head. It proves execution in weeks, not in a slide deck.
Boundary: Runs reversible backstage work autonomously. No customer, money or irreversible output.
M
One end-to-end process crossing the systems you already have, including systems without a mature API.
Boundary: Autonomous backstage work; external, financial and irreversible actions require approval.
G
One legacy application goes offline after the agent runs in Parallel and reaches the agreed match criteria.
Boundary: Parallel first. Shadow outputs are compared; cutover is assisted and approved by you.
XG
Multiple agents orchestrated under the same House Rules, with a named owner and SLA. This is an operation, not a project.
Boundary: Each agent keeps its boundary. Orchestration never weakens approval for irreversible work.
No. Size G requires Parallel: the agent shadow-runs, outputs are compared and you approve cutover only after the agreed threshold.
By five criteria: systems touched, data access, monthly volume, approval boundaries and whether output is external or irreversible.
We do not recommend or sell it that way. A size-P or size-M agent must already be live.
A size-P agent typically goes live in one to two weeks. Access, information and approval dependencies pause the contractual clock.