Stop driving your Ferrari in the parking lot

My feed is painting an uncomfortable picture.

UC Berkeley research on 306 production AI agents:
Everyone’s building agents in “safe mode” to drive efficiency.

Calvin Cheng data on valuation of 5,700+ companies, 2022-2025:
To create value you need to balance investments in efficiency and growth. 

Berkeley researchers reveal:
73% of firms deploy AI agents for "productivity gains"
68% execute ≤10 steps before human intervention
52% keep agents internal-only
70% use off-the-shelf models
Everyone's optimizing for efficiency. Not growth.

Calvin shows what the Market actually rewards:
Efficiency-only companies: 23% valuation compression in 9 months
Efficiency + growth companies: 27% valuation expansion
The gap: 50 percentage points in market value

Translation: You're building Ferrari-grade AI to drive more efficiently in parking lots.

Berkeley documents the "safe" playbook:
• Bounded autonomy (agents take ≤10 steps)
• Internal users only
• Human verification on everything
• Optimize for deployment

Calvin's research shows what wins:
• Deploy 60-80% of savings into growth
• Accept higher risk for competitive advantage
• Build for market capture, not just cost reduction
• Companies doing this achieve 25% returns—more than 2X their peers

Companies that figured this out in 2023-2025 built 18-24 month capability leads.

If everyone's still playing it safe, now's the time to create real advantage.
The window's open. But closing fast.

Are you using AI to reduce costs? Or capture markets?

The data's clear. 
Safe deployment may get through risk committees. 
But strategic deployment gets valuation premiums.

—
Source:
Measuring Agents in Production, Pan et al, Dec 2 2025
Connecting the Dots, Calvin Cheng, Dec 9 2025