Systems and data
AI workflow designer
Redesigning how a real team gets work done once part of the work can be handed to a model, and being accountable for what happens at the handover.
Written by Nivaan, founder of OffLadder · Last reviewed
Test it this week
Take one repetitive task you personally do each week. Write down its steps, hand two of them to a model, and run it for a week. Note where it saved time and the one place it quietly produced something you would not have sent.
What the work actually contains
- Sitting with a team for a day and writing down every step of one process, including the bits nobody documents
- Deciding which steps a model should draft, which a person must sign off, and where the work stops if something looks wrong
- Building the first version out of ordinary tools, then watching real people use it badly and fixing that
- Writing the short document that says what the system must never do
A realistic day
- Morning with a team, mapping one process step by step, including the workarounds nobody wrote down
- Drafting where a model could produce a first version and where a person must approve it
- Testing a rough version with two or three real users and noting where they stop trusting it
- Writing the short rules: what the system must never send, and who can override it
Tools you would use
- Process maps on a whiteboard or diagram tool
- Spreadsheets
- An AI assistant and a no-code automation tool
- A shared document for rules and decisions
The unglamorous parts
- Most of the job is persuading people who did not ask for the change
- The dull edge cases take longer than the clever automation
- You are accountable when the output is wrong, even though a model wrote it
Why it is appearing now
Most organisations now have access to capable models and no agreed way of putting them into daily work. The bottleneck has moved from the model to the process around it.
What AI changes about it
It creates the job. It also makes the hard part more human: the interesting decisions are about accountability, error handling and who is trusted to override the output, none of which a model settles for you.
What AI does, and what stays human
AI tends to do
- Drafts routine text, summaries and classifications
- Suggests steps to automate
- Speeds up building a first version
Stays with a person
- Deciding who is responsible at each handover
- Noticing where a process really breaks
- Earning the team's trust to use it
The human abilities it leans on
- Noticing where a process actually breaks rather than where the diagram says it does
- Asking the awkward question about who is responsible when the output is wrong
- Explaining a change to people who did not ask for it
What it can grow out of
- Operations, admin and coordination roles
- Customer support, where you already know where the process fails
- Analysts, teachers, project managers — anyone who has redesigned a process to survive real users
What nobody knows yet
Whether this settles into a job title of its own or becomes an expected part of every operations and management role.
Try it now: a 20-minute test
- Pick one weekly task you already do, such as a status update or sorting emails.
- Write every step on paper, including the ones you do without thinking.
- Mark each step: a model could draft it, a person must decide it, or it must never be automated.
Afterwards, notice whether
- You found at least one step you had never noticed you do
- You can say who is responsible if the drafted part is wrong
- You enjoyed the mapping more than you expected, or clearly did not
Sources you can check
- Management analysts (nearest established occupation) · O*NET OnLine, US Department of Labor
Guides worth reading