Uploaded July 2026 | Updated September 2026, 3 days ago
Spending 10x more on tokens doesn't make an engineering team 10x more productive. Jellyfish's data, drawn from hundreds of thousands of developers, shows the top 10% are about twice as productive as the median while burning 10x the tokens to get there.
Nicholas Arcolano, Head of Research at Jellyfish, sat down with WorkOS at the AI Engineer World's Fair 2026 to explain what the token-maxing era actually bought, and where it hit a wall.
We get into:
• Why "tokens are rocket fuel and token maxing is rocket science," and the diminishing returns as spend climbs
• The distinct regimes of token use, from fancy autocomplete to 500-million-tokens-a-week multi-agent workflows
• Why most engineers still spend a fraction of their salary in tokens, and why finance hasn't caught up to treating compute as capacity instead of a dev-tools line item
• Why the real bottleneck has moved off coding onto product, review, and go-to-market: teams can't even absorb 2x, let alone 10x
• What a genuinely rebuilt product development lifecycle takes, and why most orgs only move that fast in crisis mode
• Arcolano's advice for anyone trying to modernize a resistant organization: show, don't tell
A grounded, data-backed look at what AI is really doing to engineering capacity, from someone measuring it across roughly 1,000 companies.
To learn more about Jellyfish, visit jellyfish.co/. To learn more about WorkOS, visit https://workos.com.
Spending 10x more on tokens doesn't make an engineering team 10x more productive. Jellyfish's data, drawn from hundreds of thousands of developers, shows the top 10% are about twice as productive as the median while burning 10x the tokens to get there.
Nicholas Arcolano, Head of Research at Jellyfish, sat down with WorkOS at the AI Engineer World's Fair 2026 to explain what the token-maxing era actually bought, and where it hit a wall.
We get into:
• Why "tokens are rocket fuel and token maxing is rocket science," and the diminishing returns as spend climbs
• The distinct regimes of token use, from fancy autocomplete to 500-million-tokens-a-week multi-agent workflows
• Why most engineers still spend a fraction of their salary in tokens, and why finance hasn't caught up to treating compute as capacity instead of a dev-tools line item
• Why the real bottleneck has moved off coding onto product, review, and go-to-market: teams can't even absorb 2x, let alone 10x
• What a genuinely rebuilt product development lifecycle takes, and why most orgs only move that fast in crisis mode
• Arcolano's advice for anyone trying to modernize a resistant organization: show, don't tell
A grounded, data-backed look at what AI is really doing to engineering capacity, from someone measuring it across roughly 1,000 companies.
To learn more about Jellyfish, visit jellyfish.co/. To learn more about WorkOS, visit https://workos.com.
