Uploaded July 2026 | Updated September 2026, 3 days ago
The last 12 months of AI adoption were about token maximization: stuffing everything into massive context windows because you finally could. Remy Guercio from Tailscale says that era is ending, and ROI maximization is next.
In this conversation, Remy breaks down what he's seeing across companies drowning in AI bills: why consolidating onto a single model provider is the wrong instinct, how open-source and Chinese models are reshaping the cost-per-task curve, and why he thinks about frontier model efficiency the way engineers think about the Concorde, burning more fuel to go twice as fast.
Topics covered:
• Why "token maximization" defined the last year of AI spend
• The wrong way companies are responding to rising AI bills
• Open-weight and Chinese models vs. frontier model efficiency
• The Concorde analogy for test-time compute
• How Tailscale's own engineering team uses AI internally
• What ROI maximization actually looks like in practice
To learn more about Tailscale, visit tailscale.com/. To learn more about WorkOS, visit https://workos.com.
The last 12 months of AI adoption were about token maximization: stuffing everything into massive context windows because you finally could. Remy Guercio from Tailscale says that era is ending, and ROI maximization is next.
In this conversation, Remy breaks down what he's seeing across companies drowning in AI bills: why consolidating onto a single model provider is the wrong instinct, how open-source and Chinese models are reshaping the cost-per-task curve, and why he thinks about frontier model efficiency the way engineers think about the Concorde, burning more fuel to go twice as fast.
Topics covered:
• Why "token maximization" defined the last year of AI spend
• The wrong way companies are responding to rising AI bills
• Open-weight and Chinese models vs. frontier model efficiency
• The Concorde analogy for test-time compute
• How Tailscale's own engineering team uses AI internally
• What ROI maximization actually looks like in practice
To learn more about Tailscale, visit tailscale.com/. To learn more about WorkOS, visit https://workos.com.
