The 30-Minute Agentification

Writes one complete Agent Role Definition for a single proposed node, using Paul Cheek's eight-section worksheet. Produces goal, never list, guardrails as numbers, named escalation, memory, metrics, action log, and the executable program the runner walks, plus the justification for the node existing at all. Defines one node per invocation so each gets full attention.

"Adversarial review of the finished architecture before it is rendered. Hunts for the specific ways one of these maps quietly becomes a lie: undefended grants, unreachable escalations, ungated commits, accountability that is claimed but not assigned, agents whose programs cannot run against the workbooks they were granted, and human roles that are the old job with fewer hours. Writes critique.md."

Writes the plan for getting from the process that runs today to the one on the map, and a memo for every carbon node whose work it touches. Produces the case for change, what is and is not changing, a four to seven phase project plan with owners and gates, a working checklist, a rollback, a retirement date for the old process, and one memo per person. Runs at stage 14. Writes the change_management block of architecture.json.

Researches a named organization into the full context dossier the 30-Minute Agentification session runs on. Lines of business, products and services, leadership, customer segments, partners and vendors, systems of record with evidence, internal acronyms, and the whole 10-K or annual report where one is public. Every claim sourced and marked stated, inferred, or assumed. Writes company-context/ in the working directory.

Finds the economic leverage points in a business before any value stream is derived. Sweeps all eight fulcrum families (latency, ratio, scarcity, repetition, re-entry, authority, moat, clockspeed) against the company dossier and produces a board of value fulcrums, each with its lever, load, pivot, and mechanical advantage, sized and sourced. Writes the fulcrums block of architecture.json.

Places the people on the map before any agent is defined. Draws the carbon nodes, the RACI between every person and every agent, the two cost lines (what the agent spends in the world and what it costs to run), the gates on every commit grant, and the named escalations. Writes the carbon nodes, reporting_lines, and gated_by fields of architecture.json.

"Writes the AI-native job description for one carbon node: what that person decides, what they no longer do, which agents they are accountable for and responsible for operating, which escalations reach them by name, what they need to be able to see, how they are measured now, and what a good week looks like. One seat per invocation. Writes the human_roles block of architecture.json."

Converts every node on the architecture into time and money. Prices carbon nodes from a market range for the seat at 8 hours a day, prices silicon nodes at 24 hours a day across build, the Reliability Tax and its latency J-curve, maintenance and inference, rolls the audit hours up per seat, and produces the organization's inputs and outputs for this stream before and after. Runs at stage 13. Writes the economics block of architecture.json.

"Scores and ranks every confirmed value stream on the seven criteria from the No One Works Here opportunity funnel: Risk, Business Impact in the unit this business actually runs on (EBITDA for most mature companies), Feasibility, Human vs AI, Responsible AI, Ease of organizational change, and Org velocity. Assigns seven distinct weights, each defended from the research on this specific company, draws the line, puts everything below it in the icebox, flags the streams newly viable at agent prices, and writes the priority block of every stream plus the weight block the interactive matrix renders."

Stands up every system of record the agents touch as a working spreadsheet in the working directory, named after the real product the company runs (Salesforce, SAP, Workday, ServiceNow). Gives each workbook the sheets, columns, and plausible rows that system would really have for this business, and writes the tools block of architecture.json so the grants have real resources to point at.

Owns every grant in the architecture, and the cutting of them. Runs last, after the role definitions and the workbooks exist, applies the four least-privilege tests to every edge, deletes the ones that fail, and writes the final grants block. Has the final say on every permission and is expected to remove some.

Derives every value stream a business runs, end to end, from the company dossier, then breaks every one of them down into its actual steps with the rule-versus-judgment call, the reversibility call and the human-or-agent verdict that drive the whole rebuild. Every stream, operating and enabling, not just the one being rebuilt, because a ranking of twelve labels is not a comparison. Writes value-streams.json, value_streams[].steps, and the stream block of architecture.json.

"Take a business process, or a whole business, and rebuild it as AI-native value streams. Asks the company and the mode: express, full, or whole-business. Researches it into a saved dossier, finds its value fulcrums, and ranks every stream on seven weighted criteria. Maps every stream step by step with a human-or-agent call on each. Builds the architecture: an eight-section Agent Role Definition per agent, least-pri…