Red Robin Died by Spreadsheet. Agents Die by Vanity Metrics.

Garry's List recently ran Red Robin Died by Spreadsheet — a clean operator parable: manage by vanity metrics and the business dies while the dashboard looks green.

Multi-agent fleets die the same way.

Vanity metrics that look like progress

• Tokens burned
• Lines of code shipped
• "We ran 40 tools this turn"
• Dashboard logins as proof of work

None of those book a job, close a ticket, or ship a safe agent action.

What actually saves money

1. Memory before gen — similarity-search prior answers; REUSE or ADAPT before another full model call.
2. Short-lived tokens — runtime scoped credentials beat long-lived secrets sitting in .env.
3. MCP over dashboards — ask tools you already pay for; stop stitching twelve exports into a spreadsheet.

Honest product note

I ship after-hours / missed-call ops scoring for service businesses (HVAC/plumbing After-Hours Leak Score audits) and local multi-agent control on Mac/mobile. No fake traction claims here — the metric that matters is answer rate and reuse rate, not "AI activity."

If you run agents in production, track outcomes. Spreadsheets lied to restaurants. Token tallies lie to agent fleets.

— Igor Ganapolsky

Source spark: https://garryslist.org/ (builder/civic reporting)