If nobody can write down in plain steps how a task is done today, buying an agent for it is premature. The written method is the real readiness test, and it surfaces hidden disagreements early.
The last piece in this series sorted a proposed automation into a tool, a skill, or an agent by asking whether it needed to decide what happens next. This piece takes the question a step earlier, back to the moment right before anyone starts evaluating vendors or configuring anything. The honest, contrarian version of that moment is this: if nobody in the organisation can currently write down, in plain steps, how the task gets done today, buying an agent to do it is premature. The problem is not automation. It is documentation, and no agent fixes that for you.
This piece argues that the method has to exist first, on paper, before an agent is worth evaluating, and that skipping straight to the tool is one of the most expensive ways to discover that the method was never actually agreed on.
An agent looks like it solves the documentation problem by absorbing it. Hand the task to something capable enough, the thinking goes, and it will work out a sensible way to do it, sparing the organisation the tedious work of writing the method down first. This is backwards. An agent needs a method to operate against, whether that method is explicit or not, and if the organisation has not written it down, the agent will either invent its own version, silently, or fail in ways nobody can diagnose because nobody agreed what correct looked like in the first place.
The method is not a full standard operating procedure with numbered sub-clauses. It is the honest, plain-language answer to a small set of questions: what are the steps, in order, that a competent person actually follows today; what information does each step need; what does a correct outcome look like; and what makes a result wrong rather than merely different. An organisation that cannot answer these four questions for a task does not yet have a method. It has a habit, held in one or two people's heads, and habits are exactly what breaks first when an agent is asked to reproduce them.
Trying to write the method down, before any agent enters the conversation, does something valuable on its own: it surfaces disagreement that automation would otherwise have hidden. Two staff members who both "do the report" may be doing it in genuinely different ways, each defensible, neither written down, and an agent built against one person's undocumented habit will quietly produce results the other person considers wrong. Writing the method first forces that disagreement into the open while it is still cheap to resolve, rather than after the agent has been built and is producing output that half the team distrusts for reasons nobody can quite articulate.
An agent that automates an undocumented method does not remove the documentation problem. It hides it, temporarily, until the first disagreement about what the agent should have done forces the method into the open anyway, at a worse time and a higher cost.
Before evaluating any agent, write the method it would need to follow, in plain steps, and treat the ability to do that clearly as the actual readiness test, since an organisation that cannot write the method down is not ready to automate it.
No external statistic cited; this article presents an internal contrarian argument rather than third-party evidence.
A tool, a skill and an agent cost very different amounts to build, run and govern. One question, whether the task must decide what happens next, tells you which one to choose.
Custom Agents & Tools · 4 minGuide · 24 September 2026A template for the permitted-use table in an AI policy: one row per role and task, with four columns for role, task, safeguard and data boundary, so staff can find answers in seconds.
AI Use Policy · 3 minGuide · 20 September 2026How to break a task into steps and write down the handover point, usually just before the first judgement call, where AI output ends and human responsibility begins.
Team AI Training · 4 minIf this is the question on your desk, a thirty-minute call tells you whether the service fits, or that you do not need us yet.