Agent Library · By function · Triage · Capacity · Status
The work between the work.
Operations rarely fails at any single task. It fails at the seams — the routing, the chasing, the assembling, the noticing. None of that appears on a job description, and all of it consumes the week.
Everything routes through a person
Inbound arrives in one place and has to be read, classified and forwarded by someone senior enough to judge it.
Overrun is discovered, not predicted
Team load versus committed deliverables is knowable weeks in advance. It gets noticed the week it breaks.
Status lives in five tools
Assembling one honest picture of where things stand means asking five people who are mid-task.
The agents
5 agents. Each one earns its autonomy.
Agent
What it does
Autonomy
Human gate
Universal Inbox Triage
Classifies and routes inbound email and DMs to the right owner, with a draft reply where the reply is routine.
Notify
Client escalations always go to a human
Capacity Mapper
Maps team load against committed deliverables and flags overrun before it becomes a missed date.
Flag only
—
Status Synthesiser
Standup-ready and leadership-ready summaries pulled from your actual systems, not from asking around.
Auto
—
Vendor Coordinator
Freelancer and vendor briefs, chasers and handoffs — the coordination overhead that scales badly with headcount.
Notify
—
Meeting-to-Action
Calls end with owners, deadlines and follow-up drafts, not with "someone should send notes".
Notify
—
Autonomy key — Auto: executes above 95% confidence. Notify: executes at 85–95% with notification and undo. Review: always queued for a human. Flag only: never acts, alerts a person.
What changes
Three shifts, not a feature list.
Routing stops consuming judgement
Senior people stop spending their morning deciding who should read what.
Overrun becomes a forecast
You see the capacity collision three weeks out, while it is still a scheduling decision.
Status becomes what the system shows
Not what was said in standup by the person most optimistic about their own work.
The eval bench
What these agents have to pass before they touch real work.
- Routing accuracy against 200 historic inbound items with known correct owners
- Capacity flags backtested against your last two quarters of actual overruns
- Action extraction: recall measured against human-written notes for the same 30 meetings
Written against your data, not ours. The bench is built during Blueprint and runs on every deploy from then on.
Related
Where this connects.
See it on your Slack.
A Blueprint maps which of these agents pays off first in your exact setup — and which to leave alone.
Apply for a Blueprint