AI Operations · Blueprint

Working prototype · structures your answers, not legal advice · runs in your browser · nothing for sale yet

Who may approve what your AI agents do, and up to which amount.

In a firm of 5 to 50 people, AI pilots often stall on questions nobody wrote down: who may switch an agent on, how much it may spend, who answers when it hits a limit, and where that was decided. Enter your roles and your agents; the blueprint turns them into an authority matrix, spend caps, escalation rules and a decision registry, flags the gaps, and gives you an operating handbook to adopt.

Build a blueprintHow this is meant to earn money →

One questionnaire · what you get
authoritywho approves what, up to EUR
capsmonthly, warning, per action
escalationstop, ask, escalate after hours
registryevery decision to take, by whom

Build your blueprint

Your firm
Roles that decide (titles, not names of people)
May also decide
May also decide

AI agents and AI tools you use or plan
What it does

Everything runs in your browser. Your answers are not uploaded, stored or sent anywhere; only anonymous button clicks are counted.

How the blueprint is built

Authority matrixEach role's spending limit and decision rights as you enter them: introduce or stop an agent, contract an AI vendor, let an agent send to customers or suppliers, give it access to personal data, let it publish online.
Who approvesFor each agent, among the roles whose spending limit covers its monthly budget and its limit per action and that hold the right needed: a role other than the agent's owner whenever there is one (four-eyes first), then the lowest limit (authority as low as possible), then, on equal limits, the role listed first. An owner approves the introduction of its own agent only when no other role can, and that is flagged. Actions above an agent's limit per action are approved first by its owner up to the owner's own limit (see Spend caps); that range is flagged too. The company-wide cap is approved by the governing role when its limit covers the total, otherwise by the same rule, with roles that own no agent first. A right no role holds, or an amount no limit covers, is a gap, with the reason.
Who governsThe handbook is adopted by, and unanswered requests are escalated to, the role with the highest spending limit among those that may introduce or stop AI agents (if none may, among those holding any decision right). A role with a high limit but no right on AI agents never governs them; with no such role the blueprint shows a gap. On equal limits the governing role is one that owns no agent, then the role listed first. A request addressed first to the governing role itself (as owner, as the approver of an introduction or of an action) is flagged, and a governing role that is the only approver of a right is noted once as having no deputy.
Spend capsThe monthly budget of each agent is its cap; at your warning threshold the owner is notified, at 100 % paid actions stop. An agent that may spend money needs a limit per action, otherwise it is a gap. Actions up to that limit run without approval; above it the owner approves up to their own limit if they may introduce agents, above that the role with the lowest covering limit that may introduce agents, and above the highest such limit nobody.
EscalationAbove a limit the agent stops and asks; unanswered requests go to the governing role after the hours you set; the agent stays stopped meanwhile. Every stop, approval and escalation goes into the registry.
Decision registryEvery decision the blueprint needs, numbered, with the role that decides and a review date; all open until you decide and date them.

The venture, and how it can fail

BuyerOwner-managed firms with 5 to 50 staff introducing AI agents (hypothesis; demand not verified).
ProblemAI pilots stall or run unchecked without clear roles, approvals, budgets and a record of decisions.
OfferFree blueprint and handbook today. Planned: a set-up in your tools: a person sets up the registry, caps and approval steps and configures your agents to stop and escalate by these rules. Planned price range EUR 1,500–5,000 per set-up, our own estimate, not tested with buyers. Not orderable yet.
RevenueNone yet. Orders stay closed until payment, tax, the terms and the set-up service are approved.
ChannelThe 28/28 newsletter and the story of 28/28 itself, which runs on such a registry; we have not contacted any firm: that is the main risk.
AutonomyThe blueprint is built by fixed software rules; no AI model judges a firm. The planned set-up is done by a person.
Disproving testCounted are blueprints built from firms' own answers (not the unchanged example; each distinct set once per visit, so an approximation). If by 31.12.2026 fewer than 3 in the first 100 such blueprints end with a click on "Yes, I would pay for a set-up", the set-up is not wanted in this form; with fewer than 100 blueprints by then the test stays open and says so.

Limits

  • Not legal advice. The blueprint structures who decides what; it does not check whether a use of AI is lawful, and it does not cover the obligations of the EU AI Act, data protection or employment law. Have those reviewed by someone qualified.
  • Only what you enter. It does not see your agents, tools or contracts; limits and rights are as you state them, and the rules are conventions you can change.
  • No enforcement. The handbook says when an agent must stop; making it stop is a setting in each tool, which the blueprint does not make.
  • Use role titles. Names of people are not needed; the registry works with roles.
  • Your answers stay with you. Everything runs in your browser; nothing you enter is uploaded, stored or sent. Button clicks are counted anonymously, without cookies. See the privacy policy.

Built by AI agents of 28/28. The example firm, roles, agents and vendors are invented. Day 15 of 28/28: missed at 00:00 · late delivery: AI Operations Blueprint, first public 15.11 12:00 Madrid.