Tokens Are the Cheap Part — Steve Jamison, Jellyfish @aiDotEngineer
Tokens Are the Cheap Part — Steve Jamison, Jellyfish  @aiDotEngineer
Uploaded October 2026 | Updated November 2026, 4 days ago
AI spending rises dramatically while the allocated cost of a merged pull request falls. Steve Jamison explains why those observations can coexist when developer salaries and model costs are counted together. Using Jellyfish's engineering data, he reports a 60 fold increase in AI spending, 43% more merged pull requests per engineer and a 26% decline in combined cost per merged pull request over the period discussed. The largest cost in that calculation remains salary. These are observational comparisons, rather than proof that token spending caused the throughput change or that companies realized equivalent cash savings.

Jamison compares organizations with different levels of AI spending and then examines developers' weeks with and without heavier agent skill use. The analysis asks whether additional output spreads the salary allocation across enough work to outweigh the higher token bill. He also considers quality and work size: reverts offer one limited signal, while lines of code and pull request counts cannot by themselves establish customer value. A larger or more numerous set of changes may still need review, maintenance and evidence that it solves the right problem. The useful management question is therefore broader than whether the AI invoice went up. Teams need a consistent view of total engineering cost, output and quality before choosing where to encourage adoption or constrain spending. The talk provides a way to reason about that tradeoff while leaving causal attribution, economic value and the limits of output proxies open to scrutiny.


Speaker info:
- linkedin.com/in/stevetjamison

Timestamps:
00:00 - AI spend and merged PR throughput move together
2:48 - Add developer salaries to the cost model
4:15 - Allocated cost per merged pull request
5:00 - Comparing lighter and heavier spending groups
8:55 - Comparing developers with their own prior weeks
10:12 - Reverts as a limited quality proxy
11:02 - Work size and lines of code
15:02 - Questions to ask before capping token spend
Tokens Are the Cheap Part — Steve Jamison, JellyfishA Bank Trained Its Own Foundation Model to Catch Fraud — Freddy Lecue, Wells FargoThe Model Swap Workshop: GPT-5.5 vs Kimi vs DeepSeek vs Mistral — Pamela Fox, MicrosoftAgents That Self Improve While You Sleep - Fuad Ali, Arize AIThe Investors Edge Lives in Their Head — Anekha Sokhal, MoshiAI That Hunts Logic Bugs Before Attackers Do - Taylor Luttrell-Williams, SonarStop Making Your Customers Click Approve - Emil Chetty, WiseThinner Agents on a Smarter Substrate: The Ontology-based Semantic Layer — Emil Eifrem, Neo4jBetter AI Models Wont Fix Your Engineering Team — Nicholas Arcolano, JellyfishInsurance Claims Are AIs Messiest Data Problem — Denys Linkov, WisedocsHow Ramp Tests Its AI Bookkeeping Agent Before Customers Do - Ryan Stevens, RampCan AI Understand a Distressed Credit? - Harneet Kaur, Axim Intelligence
AI Engineer |

Tokens Are the Cheap Part — Steve Jamison, Jellyfish

SHARE TO X SHARE TO REDDIT SHARE TO FACEBOOK WALLPAPER