Opportunities come into my consultancy from a lot of directions — client inquiries, partner introductions, vendor pitches — and for a while my triage process was whatever I felt that day. Something exciting jumped the queue. Something that arrived on a busy day sank. Nothing was written down, so nothing could be checked.
I built a Rovo agent to fix that, and the design that finally worked splits the job in two:
Rovo handles conversation and interpretation. It asks the follow-up questions and makes sense of messy answers.
Jira holds the state. Every fact the agent gathers lands in an explicit field. Every stage the item sits in is visible.
Automation validates and sequences. An item moves forward only when the required fields are complete and the conditions are met. No step gets skipped because the agent felt confident.
A human confirms before anything goes out. No outreach, no commitment, leaves my desk without my say-so.
Before that split, the rules lived in the prompt. The agent skipped required questions, gathered facts out of order, and returned confident answers with no ground truth behind them. After the split, every transition is a rule, every state is a field, and every failure can be traced. That is the whole trick, and it is not specific to my pipeline.
The Rubric Comes First
Here is the part I would put ahead of any tooling: an agent needs a rubric before it needs a prompt.
I score every opportunity against a fixed filter — a set of checks an item has to pass before it earns attention — inspired by Kate Judina's Six-Point Filter Framework. The six checks: frequency, manual effort, error-proneness, cross-system work, business criticality, and measurability. The point of a filter is not the score itself. It is that the same questions get asked every time, the answers sit in fields where anyone can read them, and "interesting" stops being confused with "worth pursuing."
"We are what we repeatedly do. Excellence, then, is not an act, but a habit." — Will Durant, summarizing Aristotle
Without the rubric, the agent just does what I used to do, only quicker. With it, the agent is a junior colleague working from a checklist I can audit.
What I Would Not Trust Yet
Two things, plainly:
An agent grading its own homework. If the score lives only in the agent's reply, it cannot be checked. Scores belong in fields.
"In God we trust; all others must bring data." — W. Edwards Deming
A rubric nobody argued about. A filter you have not debated with yourself is just your biases with numbers attached. The filter earns trust by being argued over, then written down, then enforced the same way every time.
Start Small
If you are staring at your own intake queue — requests, vendor pitches, support triage, all of it — try this:
- Write down the checks an item must pass before it earns attention. Keep it to a handful.
- Put the score in a field, not in a chat reply.
- Keep the human confirmation on the step where a mistake costs something real. Automate the rest.
An Open Question
What is the one check your queue most needs — the question that, if every item had to answer it, would save you the most time? I'd love to hear how you are scoring work today.
#AtlassianChampions
