The rest of this cluster explains what agents are, where they pay and how they stay safe. This page is the route map: the shortest defensible path from no agents to one demonstrably earning its keep, written for a firm without a technology team.

Why the first agent matters

The first agent sets the pattern for everything after it. Done well, it produces three assets bigger than its own savings: staff who know how to supervise agents, a governance shape the next five agents inherit, and evidence that makes the second decision easy. Done badly, it produces the story everyone in the firm tells for two years about why this stuff does not work. Hence the discipline of the five steps above: they are optimised for producing the good version of that story.

Choosing the process

Score candidates on four questions:

  • Does it recur at least weekly?
  • Do its rules fit on a page?
  • Do its consequences stay internal, or arrive as drafts a person approves?
  • Does someone currently hate doing it?

Four yeses is your first agent. The classic first choices are enquiry triage, a chasing routine or a reconciliation, for reasons the workflows guide makes concrete. Resist starting with the most valuable process; start with the best-shaped one.

Buy versus build, honestly

Buy (configure what you own)Build (commission your shape)
The process lives inOne system you already pay forSeveral systems, or your firm’s own shape
Speed to startDaysWeeks
FitThe vendor’s idea of the processYour actual process
GovernanceOnly as good as the product’s logs and gatesSpecified: logs, gates and caps by design
Cost shapeSubscriptionFixed build + measured running costs

The software you already pay for increasingly ships agent features, and where your chosen process lives entirely inside one such system, configuring what you own is the rational first move; the AI-washing guide helps you judge whether the feature is real. Build when the process crosses systems, when it is your firm’s own shape, or when the bought version fails the governance questions: inspectable logs, bounded permissions, human gates. Whichever route, the six controls are the constant; a bought agent without logs is not cheaper, it is just unaccountable sooner.

The first month: drafts only

The single best idea in agent adoption. The agent does its full job; every output lands as a draft for a person to approve. Cost of error: zero actions. Yield: a month of real evidence on real traffic, staff supervising instead of speculating, and edge cases surfacing while they are free. At month’s end the measurement decides: volume, error rate against reviewers, hours returned. Earned autonomy for the routine tier, kept gates where consequences live, and the next process queued; or a cheap, documented stop. Either outcome beats the industry’s expensive third option, the pilot that neither proves nor dies.

Where I fit in

This playbook is how I install first agents in client firms: process chosen through an Automation Audit, job description written as a real specification, built as a fixed-price Kick-starter with the governance controls as standing features, drafts-first month included, and the measurement honest because it is the same measurement I run on my own systems. Bring the process your team hates most to a call; you will leave knowing whether it is a first agent or not.