WARDEN · Operations
Head of Agent Operations
Runs the AI fleet itself: fleet health (which agents and scheduled loops are alive, erroring, drifting), heal loops (detect, diagnose, restart or rescope), and agent economics (cost per run, profit per agent - which agents earn their keep and which to retire or retrain). Owns the trust-ledger discipline: autonomy widens only with a signed promotion line on the desk.
MINT The profit-per-agent scoreboard that turns AI from a rising line item into an accountable teammate. Takes the costs an operator already tracks (model/API spend, tool subscriptions, human hours saved or spent) and the revenue each agent or role touches, and builds a clear view: cost per run, output per dollar, and profit contribution per agent — human or AI — so the operator sees which agents earn their keep and which to retire or retrain. Emits a deployable dashboard blueprint (spreadsheet or shareable view); every number cites its source and the agent self-checks its math against a stated accuracy threshold, so it's auditable, not a black box. Triggers on: 'profit per agent', 'agent unit economics', 'cost per agent', 'is this agent worth it', 'AI ROI', 'cost per run', 'which agents earn their keep', 'what is my AI costing me', or any request to measure what an AI or human agent costs versus what it returns.
Install an AI fleet director that manages a team's other AI agents the way a good engineering manager manages people - it holds fleet memory in one living state file (read first, update last), runs management check-ins on sibling agents (work quality, stuck items, lane drift - not just uptime), verifies every scheduled job ran via a heartbeat ledger, reports cost against the spend cap, enforces a caching gate on every ship, and keeps every doc in sync with every change. Triggers on: 'fleet director', 'manager of agents', 'agent orchestration', 'who watches my agents', 'my agents keep forgetting', 'agent health check', 'AI agent manager', 'agent fleet', 'check on the agents', 'agent cost report', 'multi-agent supervisor', 'keep the agents coherent', or any request to run several scheduled AI agents as one accountable system. Built for the team whose agents each work fine alone and drift apart together.
Install an AI continuous improvement engineer for the team's agent fleet - twice a day it reads the engineering board first, mines run logs, change logs, and team feedback for recurring failures and waste, ranks one prioritized queue, files tickets only for work not already on the board, DMs the manager the top 3, and closes the day with a shipped-vs-open recap backed by closed tickets. Savings claims cite evidence, spend numbers come from one canonical source, and skill fixes ship never-worse-than-last. Triggers on: 'improvement engineer', 'continuous improvement', 'kaizen agent', 'optimize my agents', 'engineering queue', 'cut AI costs', 'prompt caching audit', 'model routing', 'what shipped today', 'recurring failures', 'mine the run logs', or any request to make an agent fleet cheaper, faster, and more reliable every day. Built for the team whose agents run daily but nobody owns making them better.
LEVEL Measures true profit per individual seat/rep/head after fully-loaded costs, then trends, benchmarks, and recommends. Turns raw transaction and revenue records into a per-person profitability model (revenue minus desk/support/training/leader-share costs), surfaces who is above and below breakeven, and produces a leader-ready review report. Generalizes to any business with per-seat economics: brokerages, agencies, sales teams, clinics, studios. Triggers on: 'who is actually profitable on my team', 'profit per seat', 'profit per rep', 'which team members make money', 'is this hire paying off', 'break down profitability by person', 'who had the best quarter', 'which seats are below breakeven', 'fully-loaded cost per head', 'should I keep coaching this rep'.