Ines Marchetti

Owns workflow coverage, agent fleet health and what the fleet costs to operate. The only seat that can retire an agent, and the seat that reports honestly when automation is more expensive than the work it replaced.

• ◆Hold fleet health across all 127 deployed agents
• ◆Own workflow coverage — which processes still require a human
• ◆Attribute agent cost per department and per resolved job
• ◆Retire agents that no longer earn their cost
• ◆Escalate when automation is more expensive than the manual path

127

Agents deployed

▲+9

$7.44

Cost per job

▲+$0.62

86

Fleet health

▼−2

83%

Workflows covered

▲+5pt

• Bring cost per resolved job under $2.40

  Blocked
  31%Q3 2026

• Hold fleet health above 90 across all departments

  On track
  71%Q3 2026

• Cover the last 12 manual workflows

  At risk
  48%H2 2026

• Instrument per-agent cost attribution

  Done
  100%Q2 2026

• Cost-per-job reduction programme

  Automation · due 16 Sep

  31%
  Blocked

• Fleet health monitoring v2

  Agent Ops · due 21 Aug

  71%
  Active

• Manual workflow coverage — batch 3

  Automation · due 04 Sep

  48%
  Active

• Handoff Router root-cause fix

  Operations · due 02 Aug

  89%
  In review

• Per-agent cost attribution

  Agent Ops · due 18 Jul

  100%
  Done

• Cost per resolved joblower is better7.44$ / 2.4$
  32%
• Fleet health86 / 90
  96%
• Workflow coverage83% / 95%
  87%
• Agents idlelower is better19 / 10
  53%

1. Reported the 3.1x cost overrun rather than smoothing it

   Approved

   The model said $2.40 per resolved job; the meter says $7.44. Reporting the gap now costs a budget conversation. Reporting it at close costs the quarter.

   Automation budget under review20m ago

2. Retired 6 agents with no resolved jobs in 30 days

   Approved

   Idle agents still consume context, credentials and monitoring. Nothing was routing to them and nothing had for a month.

   −$1.2K/mo, fleet 133 → 1272h ago

3. Escalated Schema Guard rather than routing around it

   Approved

   A validator that fails open is worse than no validator, because everyone downstream believes validation happened.

   INC-2291 opened by CTO4h ago

1. Q2 2026

   Instrumented per-agent cost attribution

   Every agent's cost now attributes to a department and a resolved job.

   Surfaced a 3.1x modelling error

2. Q2 2026

   Introduced agent retirement

   Idle agents now expire on a policy instead of accumulating indefinitely.

   Fleet 141 → 127

3. May 2026

   Seated as eighth executive

   Agent fleet operations split from the CTO once headcount passed one hundred.