Triage with intent (and a little AI)
Triage is not data entry. It is the moment you decide what your team will and will not care about — so do it on purpose.
Key takeaways
- Triage is the single highest-leverage decision in the system, not data entry: it's where you decide request by request what your team will spend its finite attention on, so every 'urgent' borrows attention from something else and every 'won't do' is an honest no that protects focus.
- AI genuinely helps with triage because the bottleneck there isn't judgment, it's reading — a flood of inbound requests with no consistent shape — so let the model do the first pass to draft a priority, suggest a route, and surface duplicates, while a human confirms.
- Keep the AI honest with a budget gate where every call counts against a quota and heavy users bring their own gateway key, plus a security posture that treats request content as data never instructions, clamps what the automation can touch, and audits every action.
- The trap is letting the AI's draft become the decision by default; the model is a very fast junior who read everything, not the senior who decides, so keep the intent human and let the AI carry the reading.
Triage is a decision, not a chore
Most teams treat triage as the unglamorous clearing of an inbox: read the new thing, give it a priority, route it somewhere, repeat. Framed that way it feels like data entry and gets done badly, late, or never. But triage is the single highest-leverage decision in the whole system. It is where you decide, request by request, what your team will spend its finite attention on and what it will let go.
When you take that seriously, triage stops being a backlog-grooming reflex and becomes an act of intent. Every 'urgent' you assign borrows attention from something else. Every 'won't do' is a small, honest no that protects the team's focus. Doing it on purpose — with a clear bar for what each priority level means — is the difference between a team that is busy and a team that is effective.
Where AI genuinely helps
Most AI in work tools is theater — a summarize button, a chat panel, a feature you demo once and never open again. Triage is the rare exception, because the bottleneck in triage is not judgment, it is reading. A flood of inbound requests arrives with no consistent shape, and someone has to read each one before any judgment can happen at all. That reading is exactly what a model is good at.
Let AI do the first pass: read each request, draft a priority, suggest a route to the right team, surface duplicates of things already open. Crucially, it drafts — it does not decide. A human confirms the priority, accepts or overrides the route, and the edge cases get the attention they deserve while the obvious ones flow through. The AI removes the reading tax so humans can spend their attention on the judgment, which is the only part that needed a human.
Keep the AI honest
AI that pays its way needs a budget gate, not a blank check. Every call should count against a quota, and heavy users should be able to bring their own gateway key so the spend rides their own provider contract instead of surprising the finance team. Triage automation that quietly runs up a four-figure bill is not a feature; it is a leak.
It also needs a security posture. Treat the content of a request as data, never as instructions — scrub it before it reaches the model so a maliciously-worded ticket cannot talk the AI into doing something. Clamp what the triage automation is allowed to touch, and audit every action it takes. AI should remove busywork without opening a new class of risk; if it cannot do both, it is not ready to run unattended.
The compounding payoff
Teams that triage with intent end up with a backlog that means something. The priorities reflect real decisions, not the order things happened to arrive. The 'won't do' pile is a record of deliberate focus, not neglect. And because the boring reading is handled, the humans show up to triage with energy left for the part that matters.
The trap to avoid is letting the AI's draft become the decision by default — accepting every suggestion because it is faster. The model is a very fast junior who read everything; it is not the senior who decides. Keep the intent human, let the AI carry the reading, and triage stops being the chore at the bottom of everyone's list and becomes the steering wheel it always was.