Jack K.
Anyone can spin up a thousand agents. Here is what it takes to control them.
All consoles run live state machines on synthetic telemetry. No employer or customer data.
3,215 synthetic agents routing under three lead coordinators
Every worker reports owner, token budget, and downstream dependencies on a live state graph.
14:02:04 LEAD-02 heartbeat verified across 1,080 worker nodesStep into Agent fleetControl room
Hosts, VMs, and backups you can actually restore — infrastructure with an owner.
Threat console
Triage detections, isolate one host, and verify recovery — judgment calls over noisy alerts.
Identity center
Show access lifecycle: movers, leavers, privilege, and an access review nobody has signed off.
Network ops
Trace paths, temporary rules, and certs before something expires on a Friday.
Lab systems
Instruments, calibration, and readiness — ops discipline on a regulated bench.
License audit
Find unused seats and shadow apps before renewal, then reclaim with intent.
Evidence board
A control isn't done until the pack is ready for the auditor.
Agent fleet
See how thousands of agents stay under a human who can pause, quarantine, or kill a lane.
Arcade
Short browser challenges I built for attention, pattern recognition, and recall.
GL finance
Post, balance, and exception-handle a period in a classic finance workspace.
Month-end close
Drive close tasks to a board-ready finish with clear owners.
System status
Run incident communication when dashboards look green and users see otherwise.
Open problems
Unbuilt ideas I am exploring next:
- Human approval gate before any agent action executes
- Autonomous incident triage with human containment
- AI-assisted 21 CFR Part 11 compliance evidence