AI can make bad work harder to see
AI can remove visible friction while making unnecessary or poorly designed work less likely to be questioned.
Insights
One Idea
A short observation about AI, work and organisations.
Before asking AI for answers, ask it for questions
Generative AI can expose gaps in organisational understanding by asking what must be investigated before improvements are recommended.
AI is not observing your organisation
AI analyses representations of organisational work, so evidence, interpretation, hypotheses and recommendations must remain distinct.
Start with one piece of work
A practical way to investigate one recurring piece of work before deciding whether AI or another intervention is appropriate.
Capability does not confer authority
AI capability must be separated from the authority an organisation deliberately delegates, including clear escalation boundaries.
A human in the loop is not enough
Meaningful human oversight depends on retaining independent judgement, not merely placing an approval checkpoint in a workflow.
Data access is not organisational understanding
Access to organisational systems cannot capture all the history, context and situated judgement needed for consequential work.
Who is actually managing the work your AI does?
Delegating work to AI still requires domain, managerial and technical supervision focused on the organisational outcome.
Automation should not be the first question
Understanding real work—including exceptions, waiting and judgement—must precede decisions about what to automate.
Faster does not necessarily mean better
Automation should be assessed by its effect on organisational outcomes, not only by time saved or tasks completed.
Make judgement visible before you automate it
Making judgement visible reveals where AI can support decisions, where authority can be delegated and where humans must remain accountable.
A better sequence for automation
A disciplined automation sequence starts with observing work and defining outcomes before choosing technology or measuring system-wide effects.
Did the process work, or did the people make it work?
Successful outcomes can conceal the checking, correcting and judgement people use to compensate for weaknesses in formal processes.
What would stop working if people stopped quietly fixing things?
Invisible employee compensation can make weak processes appear healthy and must be understood before AI or automation removes it.
The five-minute check that costs 1,400 hours
Small repeated checks can consume substantial capacity, but their accumulated cost should prompt investigation rather than automatic removal.
Don’t automate the workaround
Workarounds are evidence of underlying weaknesses or legitimate needs that must be understood before their compensating work is automated.
Come back with three processes that give you grief
Prioritising valuable organisational problems keeps the intervention space wider than a portfolio framed around AI use cases.
AI should be a conclusion, not a premise
Starting with the work allows AI to emerge as an appropriate conclusion of diagnosis rather than a predetermined solution.
When the evidence runs out, don’t ask AI to guess
When organisational evidence is incomplete, AI should help design the next investigation instead of supplying a plausible guess.
Tool literacy is not improvement capability
Using AI tools effectively is different from knowing what organisational work is worth improving and whether change made it better.
Reveal how your work
really works.
A first conversation is free, and usually enough to tell you whether there's a leverage point worth pulling.