FOUNDER PLAYBOOK
How to choose your first AI employee
Choose the first AI employee by the work it can reliably support. Look for a recurring task with clear inputs, a checkable output and a human owner. Avoid starting with the most complex or sensitive process in the business.
Look at a real week
Write down the work that repeatedly interrupts your focus. Include preparation, coordination and correction, not just the final deliverable. A task that takes twenty minutes every day may be a better starting point than a large project that happens once a year.
For each task, record how often it occurs, who does it and where it gets stuck. Be specific. “Marketing” is a department; turning one approved idea into a first draft is a task.
Score readiness before excitement
Use a simple low, medium or high rating for frequency, clarity of inputs, ease of review and consequences of error. These are discussion prompts, not a scientific score or a promise of return.
- Frequency: does this work happen often enough to justify a system?
- Input readiness: can the role obtain reliable information?
- Reviewability: can someone recognise a correct result?
- Ownership: is a person available to maintain the process?
- Impact: does a faster accepted result solve a real bottleneck?
Three possible starting points
A research employee can prepare a brief with links and unresolved questions. A content employee can turn approved material into a draft. A meeting-preparation role can organise updates and open decisions. All three can begin with a person initiating the work and reviewing the result.
These are examples, not universal recommendations. If your records are incomplete or your process changes every time, begin by documenting the work before automating it.
Define the pilot on one page
Write the role name, purpose, input sources, expected output, reviewer and stop condition. Add a small set of test cases and the measures you will compare. Keep the existing process available until you know the new one is dependable.
Include correction time in the comparison. A fast draft that creates a long editing task may not improve the workflow. The goal is the time and cost required to reach an accepted result.
Expand after learning
At the end of the pilot, decide whether to improve, expand or stop. A useful finding may be that the process needs better source information rather than a more advanced model. The next role should solve the next observed bottleneck, not simply complete an imaginary AI org chart.