The AI bill arrived as one number — nowhere to cut. Broken into six terms and stripped of 70K idle tokens, it became six independent levers.
One of the world's largest ride-sharing platforms, running 30,000+ AI coding agent skill executions daily across a codebase spanning hundreds of millions of lines of code and thousands of data tables.
As AI coding adoption surged, total spend grew with it, and aggregate cost metrics made it impossible to identify what to fix. A hidden tax...
Also reported by
Mobility and delivery platform for ride-hailing, food, and freight logistics.
Uber's Code generation is part of this use case:
Related implementations across industries and use cases
Every new support agent meant months of engineering handoffs. Now the people closest to customers build and ship their own.
Each new agent took 8 weeks to ship reliably. A factory framework cut that to days, freeing people for the relationship work AI can't do.
Faults took 5–6 hours to diagnose, and retiring veterans were walking out with decades of floor expertise. GearPal answers in minutes.
Legacy migrations took 12.5 man-months. Engineers now set the architecture and let AI handle bulk coding, freeing them to focus on QA.
Feature requests once sat in backlogs. Now, developers let AI generate working preview branches in minutes for real-time customer review.
Fragmented support caused frustrating handoffs. Live AI translation now lets a small English-speaking team handle 14 languages instantly.
Each new agent took 8 weeks to ship reliably. A factory framework cut that to days, freeing people for the relationship work AI can't do.
Large AI training jobs meant fighting for preemptible slots or leaving campus. Marlowe gave any lab guaranteed multi-node access on demand.
Colleagues hunted through knowledge systems for answers; Global Markets hand-sorted emails before ops could act. Claude now handles both.
The AI bill arrived as one number — nowhere to cut. Broken into six terms and stripped of 70K idle tokens, it became six independent levers.
One of the world's largest ride-sharing platforms, running 30,000+ AI coding agent skill executions daily across a codebase spanning hundreds of millions of lines of code and thousands of data tables.
As AI coding adoption surged, total spend grew with it, and aggregate cost metrics made it impossible to identify what to fix. A hidden tax...
Also reported by
Mobility and delivery platform for ride-hailing, food, and freight logistics.
Uber's Code generation is part of this use case:
Related implementations across industries and use cases
Every new support agent meant months of engineering handoffs. Now the people closest to customers build and ship their own.
Each new agent took 8 weeks to ship reliably. A factory framework cut that to days, freeing people for the relationship work AI can't do.
Faults took 5–6 hours to diagnose, and retiring veterans were walking out with decades of floor expertise. GearPal answers in minutes.
Legacy migrations took 12.5 man-months. Engineers now set the architecture and let AI handle bulk coding, freeing them to focus on QA.
Feature requests once sat in backlogs. Now, developers let AI generate working preview branches in minutes for real-time customer review.
Fragmented support caused frustrating handoffs. Live AI translation now lets a small English-speaking team handle 14 languages instantly.
Each new agent took 8 weeks to ship reliably. A factory framework cut that to days, freeing people for the relationship work AI can't do.
Large AI training jobs meant fighting for preemptible slots or leaving campus. Marlowe gave any lab guaranteed multi-node access on demand.
Colleagues hunted through knowledge systems for answers; Global Markets hand-sorted emails before ops could act. Claude now handles both.