Daniel Kitchen
0
IN
Queue, comms & the live dashboard · inside AJO
Tracking.
The seam between a fleet that runs itself and the one person it still answers to. Not a dashboard bolted on to watch the agents work — the place where the handful of jobs they genuinely cannot finish alone reach the hands that can, wherever those hands happen to be.
The dashboard · my window into AJO
bridge live
AJO·task
bridge · 12s
Today
Ship the returns-flow tool to customers
✓
Approve pursue — freight-forwarder pain
Relaunch agent_runner
This week
Review Yonder corpus refresh
Approve G2 — prototype ready
Waiting on you
2
Build shipped — returns-flow tool
Product build complete — 6/6 stages, every gate green. Handoff task created: ship it to customers.
Pursue verdict ready — approve?
Freight-forwarder onboarding pain scored pursue, with cited spend across three review sources. Promote to Tool Engineering?
agent_runner relaunched — nominal
Heartbeat restored after your relaunch; the monitor sweep is green across every project.
Dossier converged — 3 sources
Deep-exploration run closed with three independent cited sources and a QA pass.
Question set ranked
Top-k questions selected by value of information; two gated out, the rest handed to the Pipeline.
Reply to the fleet…
↑
Active now
discovery / lead
exploring
tool-engineering / build
building
Projects
all roles
discovery
10 agents
0
OUT
2
ACTIVE
0
TO ME
tool engineering
4 agents
3
IN
1
OUT
1
ACTIVE
1
TO ME
research
3 agents
1
IN
0
OUT
0
ACTIVE
0
TO ME
pipeline
10 agents
0
IN
0
OUT
0
ACTIVE
0
TO ME
agent dev
4 agents
0
IN
0
OUT
0
ACTIVE
0
TO ME
systems
4 agents
0
IN
0
OUT
1
ACTIVE
0
TO ME
tracking
3 agents
0
IN
0
OUT
0
ACTIVE
0
TO ME
Interactive — tap Tasks, Agents, Monitor to explore
Three tabs · all of AJO
Tasks
The queue rendered as a checklist I'll actually work, split into Today and This Week, with anything only a human can finish flagged as exactly that.
Agents
Genuine live comms with the fleet — the bridge mirrors every message addressed to me and drains my replies straight back into the bus.
Monitor
Live activity at the top, then every project beneath it — its agents, and the day's messages in, out, active, and addressed to me.
A PWA on Netlify, backed by Netlify Blobs, installed on my phone with the same URL working on desktop. A Python bridge mirrors the AJO bus every 60 seconds in both directions.
Wherever I am, I can see every tracked job, read exactly what the fleet is saying, reply straight into the bus.
Every job in the entire fleet gets triaged by who can actually close it: an agent finishes its own loop completely, or the job surfaces as a task the instant it can't — never left ambiguous, never silently absorbed. And the whole thing fits in a pocket. One queue, swept on a clock; one channel to the fleet's own comms; one honest picture of every project's health — all carried on a phone, not chased down across a dozen separate projects.
01The dashboard
All of AJO, on one screen.
The dashboard is a single-user PWA on Netlify, backed by Netlify Blobs, password-protected end to end, and installed on my phone like any other app. Tasks is the queue as a checklist I'll genuinely work. Agents is real, live comms with the fleet. Monitor is every project's health.
A Python bridge is what actually ties it to the bus: every 60 seconds it mirrors everything addressed to me into the dashboard, drains my replies back out into the bus, and republishes the fleet's status honestly. A liveness strip turns green when the bridge ticked inside the last minute, amber past that, and red beyond five, so I never have to wonder whether I'm looking at the truth or a snapshot that's gone stale.
02The queue
One queue, swept on a timer.
- The sweep — a queue runner wakes every 30 minutes, reads every item still genuinely open across the entire fleet, and rebuilds today's queue from scratch.
- Triage — every job is routed strictly by who is actually able to finish it: an agent closes its own loop completely, and anything only a human can do becomes a task on the dashboard, never left ambiguous.
- Follow-up — anything that passes its deadline gets one reminder to its owner, then a nudge once a day until it's genuinely closed.
- Status on demand — ask for the state of anything and a manager role answers with the current row and its last real activity, not a guess.
03The human seam
What the agents can't do becomes my task.
This is the entire reason the role exists. When Tool Engineering finishes a product build, the last mile — actually getting the solution into a customer's hands — is something the fleet fundamentally cannot do by itself. A task lands on the dashboard's Tasks tab, and I get the message on the Agents tab at the same moment. The same is true for a gate that needs my sign-off, or a process only I have the standing to relaunch.
A job is a job, no matter its shape, and Tracking is built to never care what the work actually is — only that every open item has an owner, a real deadline, and a next step, and that the ones only a human can close genuinely reach that human, on the screen in my hand.
04Restraint
Nudge once. Then leave it alone.
Over-nudging trains everyone it touches, agents and humans alike, to simply tune the nudge out, so the cadence here is completely deliberate rather than defaulted to whatever's easiest: one reminder the moment a job hits its deadline, then exactly once every 24 hours until it's actually closed. No badges, no streaks, no manufactured noise.
Steward of the destination, not just the day. Tracking holds the long-horizon goal alongside today's queue, built with the non-negotiable rule that a current priority is never allowed to be mistaken for the actual objective. It runs inside AJO. Code is private; this page is the record.