The delegation boundary: how far you let the machine go

My companion piece argued that businesses sit at different points on the agentic spectrum, measured by how much of their clientele will be a machine. That tells you how many of your buyers are agents. It says nothing about how much those agents are trusted to do, and that’s a separate question with a very different answer.

A person might let an agent research every laptop on the market and still insist on choosing the one themselves. Another person, same market, same agent, says “just buy the sensible one.” Those are two different worlds for your brand, and the difference between them is the delegation boundary.

What the delegation boundary measures

The delegation boundary is the point inside a single decision where the human stops deciding and the agent takes over. It runs from “tell me what you found” at one end to “handle it” at the other, and it moves per person and per purchase, not per market. The same buyer sits low on it for a house and high on it for printer ink.

It matters because the deeper the boundary, the less human judgement stands between your brand and the outcome. At the shallow end a human still reads everything and forgives your gaps. At the deep end the agent decides on the evidence in front of it, and if your brand is unclear or uncorroborated, you’re cut with no appeal and no conversation.

The ladder: six rungs across three modes

The boundary isn’t loose. It falls on one of six rungs, and the six sit inside the three modes of delegation: Search, Assistive and Agential. Search holds one rung, Assistive two, Agential three, because the gradation that matters is all in the deep end. Each rung has two faces, what the agent does and what the human keeps.

RungModeThe agentThe human
1 · InformSearchgathers and presents, ranks nothingActive: evaluates and chooses everything
2 · ShortlistAssistivecuts the field to a qualified set, favours noneGuided: picks from the set
3 · RecommendAssistivenames one option and argues for itDeciding: one binary yes or no on that option
4 · ConfirmAgentialis ready to act and holds at the thresholdSupervised: confirms or rejects each act in real time
5 · Bounded MandateAgentialacts alone up to a set limit, then reverts to ConfirmMandating: sets the envelope, not each act
6 · Standing DelegationAgentialacts continuously across many transactionsPassive: keeps only the outcome and the right to revoke

A ladder to argue with, not a standard to cite.

The top half, rungs one to three, carries no mandate: the agent advises and the human decides. The bottom half, rungs four to six, carries a mandate: the agent acts, and the three rungs are three widths of the same thing, one act, then one transaction up to a limit, then a whole project. Nothing down there is truly unbounded, not even standing delegation. The mandate just gets wider.

The line that matters most is the one between rung three and rung four, the crossing from Recommend to Confirm. That is where Assistive becomes Agential, where the machine stops advising and starts acting, and where your brand stops being read and starts being executed against. It is the same crossing the customer journey makes when it moves from recommendation to action.

The boundary moves both ways. It ratchets up as a buyer learns to trust an agent, and it drops back when they decide they have handed over too much. Some of that reversal is automatic: a bounded mandate that meets a transaction over its limit falls back to Confirm and asks first. The ladder is a dial, not a one-way gate.

Why the boundary decides what your brand has to be

At the top of the ladder your brand is talking to a human through an agent, and the old rules broadly hold: be persuasive, be clear, give a person reasons to choose you. Near the bottom your brand is talking to the agent itself, and persuasion stops mattering. What matters is whether the machine can understand who you are, verify what you claim, and trust you enough to act. There’s no salesperson in that room to smooth over a gap, no charm to fall back on, only the evidence you’ve left for the machine to read.

That’s the shift worth preparing for. As the boundary deepens, brand work moves from convincing people to making the truth about you legible and corroborated for the machine that now acts in their place. It’s the same discipline aimed at a harder, more literal reader.

Argue with me

The rungs are mine, and I’ve drawn them cleanly on purpose so there’s something to disagree with. Tell me where your buyers actually sit today, and where you think they’ll sit in two years. Read this alongside the agentic spectrum: the spectrum tells you how many of your buyers are machines, the boundary tells you how far each one is trusted to go. Put the two together and you can size, quite precisely, how much of your future depends on a machine understanding you correctly.

Similar Posts