Do you need an agent, or just an automation?

The short answer

A managed agent is a system that reads what's happening in your business, chooses what to do next, and drafts the work — inside a line we draw with you before any of it goes live. Anything touching money, a dispute, or a promise to a customer goes to a person. The managed part is us: we build it, run it, watch it, and fix what breaks.

Most work doesn't need one. That isn't modesty. For standardized, repeatable work, plain automation is faster, cheaper, and more reliable — and knowing which is which is a sign a shop has done this before, not a sign it's playing safe. Agents earn their keep where the job changes shape every time. (A study of 50+ agentic builds landed in the same place — Rewired, 2nd ed., McKinsey/Wiley, ch. 5.)

Three rungs, and what each one is allowed to do without asking

We publish the rung for every piece of work we take on. The rung isn't about how clever the software is. It's about permission.

Managed automations

One fixed job, the same way every time. Order sync at 4am, a nightly reconciliation, a batch of production files. Decides nothing. This is the majority of what we build, and it's where most of the hours come back.

Managed agents

Reads context, picks an action, drafts the work. Worth it where the job changes shape every time — a customer email that needs a real answer, an outreach sequence that has to fit the account. It works inside the line we drew with you; money, disputes, and promises to a customer go to a person.

Managed apps

The screens your team actually touches — a production queue, an order tracker, a phone line that answers after hours. Here the person is the one acting. We build the surface and keep it running.

Every rung is managed. That's the constant. What changes is how much the software gets to decide on its own.

The same three rungs, on five questions

Structure adapted from the agent-usage spectrum in Rewired, 2nd ed. (McKinsey/Wiley), ch. 22.

That spectrum has a fourth rung we don't sell: fully autonomous agents, running with nobody in the loop. We don't build those for businesses this size. The oversight is what makes the rest of it safe to leave running — and the point of the line was never fewer people. It's that the people you have stop doing the parts a machine should have been doing.

What this isn't

What an honest portfolio looks like

Wetherby Ridge is a working metal shop selling made-to-order steel across two storefronts. Here's what we actually built, tagged by rung:

• Order capture and reconciliation across both storefronts — automation. Roughly 4,000 orders reconciled into one production tracker, with nobody retyping anything.
• Cut-file generation — automation. Ten to fifteen minutes of hand work per file became hundreds of files in seconds; a person approves each one before steel moves.
• Seven scheduled jobs running overnight — automations. Syncs, status pushes, and reporting, all finished before the shop opens.
• The email reply assistant — agent. It reads the customer's question and drafts the answer in the shop's voice. Anything about money, a refund, or a delivery promise goes to a person.
• Track My Order and the voice line — apps. The customer checks status at any hour. Nobody on the floor stops work to answer the phone.

One agent. Everything else is automations and screens — and that mix is what returned 500+ hours of skilled labor and took the owner out of the middle of his own operation. The ratio is normal. A shop that tells you otherwise is selling you a rung you don't need.