Where you are

AI work needs too much checking

AI output is creating rework, or you are unsure how people use AI. Establish what is happening, who is responsible and what better work would look like.

Plausible work that someone else has to repair

You receive a proposal with unsupported claims, an answer that misses the customer’s question, or a summary that nobody has checked against its sources. Producing it took minutes. Establishing whether anyone can rely on it takes much longer.

People sometimes call this “AI slop”: plausible output that adds volume while passing checking and correction to someone else. The useful question is where the work fails and who bears the cost. Blaming the person who used AI will not explain whether the task, information, instructions or checks were adequate.

You may have a different starting point: you do not yet know which tools people use, what information they share or which decisions rely on the output. That uncertainty is enough reason to establish the boundaries of use.

When the output is unreliable

Choose an example of work that matters to a recipient. Define what a correct, useful result would look like, then follow the work through preparation, review, correction and use.

Check the sources, missing information and decisions involved. Compare improving those conditions with changing the AI approach or leaving AI out of the task. Track whether the recipient gets a better result, including the work passed downstream. More output alone does not answer that question.

When use and responsibility are unclear

Start by establishing what is permitted, what information may be shared and who can approve consequential use. Involve the people doing the work and those responsible for information and access controls. Use a non-sensitive example to understand the workflow; do not paste confidential records into a tool to demonstrate the problem.

Name who checks work before it is relied on, who handles failures and who can stop use. Written guidance needs appropriate access restrictions and review practices behind it. Training supports those practices; it does not enforce them or prove safety.

Choose the help the problem needs

  • A practical foundation across roles: AI literacy introduces context, output checking and human judgement.
  • Decisions across the organisation: Leadership alignment helps leaders examine purpose, priorities and responsibility.
  • One work problem to investigate: Discovery establishes available evidence, compares alternatives and defines a next test or stopping decision. You can start without existing metrics.
  • Systems allowed to act: Agentic engineering addresses technical permissions, verification and recovery when implementation needs those controls.

You can enter at the point your situation needs. A workflow you can investigate and improve with your own people may not need an engagement.

Start with what you know

Tell me where AI work is causing difficulty, or what you cannot yet see about its use. You do not need a formal rollout or a set of metrics to begin.

Not this? The other situations · or just tell me what's happening