Sep 12, 2026
One Chat. Five Different Jobs.
A practical operating model for using AI with clearer roles, repeatable workflows, review standards and human ownership.
Most AI work becomes confusing because one chat is expected to do five different jobs at once. The result is usually a useful conversation that never becomes a repeatable way of working.
The 5JOBS model separates the work. It helps a team decide what AI should think through, what it can do, how the work repeats, how quality is judged, and who owns the result.
The five jobs
1. Chat: think it through
Use chat to explore the problem, test assumptions, compare options and improve the brief. This is where judgement develops before action begins.
2. Agent: take bounded action
Give the agent a clear task, approved tools, useful context and a defined stopping point. Keep human review wherever the decision carries real consequence.
3. Workflow: make it repeatable
Turn the successful interaction into a sequence with inputs, steps, handoffs and expected outputs. A workflow makes good work repeatable beyond one person or one lucky prompt.
4. Scorecard: judge what shipped
Define quality before the output arrives. Use a short scorecard to test accuracy, usefulness, tone, completeness and safety.
5. Owner: put a name on it
Every AI-enabled task still needs a human owner. The owner approves the standard, reviews exceptions and remains accountable for what is released.
How to use 5JOBS on a real task
Choose one recurring task that matters to the business.
Use Chat to clarify the goal, inputs, risks and decision criteria.
Assign only the bounded actions that an Agent can safely complete.
Document the repeatable sequence as a Workflow.
Review every output against a simple Scorecard.
Name the Owner who approves the final result and improves the system.
A practical weekly use case
For a weekly campaign review, the workflow can gather the approved inputs, summarise what changed, flag missing information, compare results against the agreed scorecard, and prepare recommendations. The marketing owner still checks the context and decides what happens next.
The takeaway
Better AI work does not begin with a longer prompt. It begins by giving each part of the work the right job, then keeping a human accountable for the result.
Most AI work becomes confusing because one chat is expected to do five different jobs at once. The result is usually a useful conversation that never becomes a repeatable way of working.
The 5JOBS model separates the work. It helps a team decide what AI should think through, what it can do, how the work repeats, how quality is judged, and who owns the result.
The five jobs
1. Chat: think it through
Use chat to explore the problem, test assumptions, compare options and improve the brief. This is where judgement develops before action begins.
2. Agent: take bounded action
Give the agent a clear task, approved tools, useful context and a defined stopping point. Keep human review wherever the decision carries real consequence.
3. Workflow: make it repeatable
Turn the successful interaction into a sequence with inputs, steps, handoffs and expected outputs. A workflow makes good work repeatable beyond one person or one lucky prompt.
4. Scorecard: judge what shipped
Define quality before the output arrives. Use a short scorecard to test accuracy, usefulness, tone, completeness and safety.
5. Owner: put a name on it
Every AI-enabled task still needs a human owner. The owner approves the standard, reviews exceptions and remains accountable for what is released.
How to use 5JOBS on a real task
Choose one recurring task that matters to the business.
Use Chat to clarify the goal, inputs, risks and decision criteria.
Assign only the bounded actions that an Agent can safely complete.
Document the repeatable sequence as a Workflow.
Review every output against a simple Scorecard.
Name the Owner who approves the final result and improves the system.
A practical weekly use case
For a weekly campaign review, the workflow can gather the approved inputs, summarise what changed, flag missing information, compare results against the agreed scorecard, and prepare recommendations. The marketing owner still checks the context and decides what happens next.
The takeaway
Better AI work does not begin with a longer prompt. It begins by giving each part of the work the right job, then keeping a human accountable for the result.