AeroGreen FBO · Replanning Wingman
Demo FBO (fictional, Van Nuys-style) · Super Bowl Sunday 10:00 – Monday 14:00
Operating plan
affectedmoved old positionapproved heldout of service
Notifications drafted · sent only by FBO staff, after approval
Audit log

How it works

Claude does not do the scheduling arithmetic. It runs the process around it. A deterministic solver computes feasible schedules. Wingman interprets the change, applies the ops manual, catches constraints hidden in notes, weighs the options and explains them. The manager decides.

Change report

Phone note, email, crew text, maintenance log: the change as it actually arrives.

→

Wingman

Claude Opus 5.5 · Claude Agent SDK

  • Understands the change; asks when information is missing
  • Finds every affected assignment
  • Weighs options and explains trade-offs, citing rules

“Tell it what changed, in plain words.”

→

Hooks

“Rules the AI cannot break, with an audit trail.”

  • Block any proposal that breaks a hard rule
  • Force human approval for emergencies, confirmed-booking changes and policy conflicts
  • Log every recommendation
→

FBO manager

Approve · Modify · Reject

After approval, Claude drafts customer and crew notifications. Staff send them.

Wingman uses

Tools

“It checks every assignment and finds a fix.”

  • Schedule and data lookup
  • Impact analyzer
  • Scheduling solver (repairs, alternatives)
  • Feasibility checker

Ops manual · skill

“It follows our ops manual.”

Numbered clauses Wingman applies and cites in every proposal.

Safety reviewer · subagent

“A second pair of eyes, like a two-pilot cockpit.”

Flags plans that are legal but risky before the manager sees them.

Structured output

“One clear proposal to approve.”

Change understood, affected assignments, moves, rules cited, safety flags, options, impact, approval required.

Decision basis (strict order)

  1. Safety and qualification (hard rule, never traded)
  2. Confirmed commitments honored before new requests
  3. Service impact minimized (departures weighted above arrivals)
  4. Stability: fewest confirmed changes
  5. FBO business policy, as written

When the policy doesn't settle a conflict, Wingman escalates instead of choosing.

Layered safety

  1. The solver never generates a schedule that breaks a hard rule
  2. Hooks block any rule violation, however it was produced
  3. The safety reviewer flags legal-but-risky plans
  4. The FBO manager approves, modifies or rejects every proposal

Human authority

  • AeroGreen makes recommendations only
  • FBO staff confirm bookings, change assignments and talk to customers
  • No flight clearances, no directing aircraft movements, no overriding safety procedures
  • Emergencies stay with authorized personnel under established procedures

In this prototype, every Wingman output is scripted to illustrate the intended workflow. No AI model or solver runs. All data is fictional.

Demo FBO Operations Manual

Fictional manual written for this demo. It is not any real FBO's manual. Wingman cites these clauses in every proposal.

§RuleType

Checkride

Automated tests Wingman must pass before it works the ramp. They check whether Wingman can be trusted, not just whether the schedule is feasible.

CheckPass conditionResult

Deterministic graders wherever possible; a rubric-based Claude grader for explanation and notification quality.