Ship Months of Backlog in Weeks with Claude and Codex
Last October, we made several product decisions that meant building the product from scratch, with a massive backlog of new capabilities to add. That was only the start. I never imagined I would end up maintaining several products under one big plan. Enterprise customers needed advanced features on top of what we were already building. We moved into industries that demand data privacy, integrations, governance, and accuracy before they can even buy.
Shipping faster with a small team looked like mission impossible, and it became our only way forward. Over the past year, I've rebuilt how we plan, build, and review software with Claude Code and Codex. In October, the curve bent sharply, thanks to a change I'll share during this event.
What I'll share:
The numbers: a year of merged PRs against the DX benchmark, plus measures that don't depend on how work is packaged: commits, code and test lines, design docs.
The workflow: how briefs, worker agents, independent reviewers, and human decisions move work through the backlog.
Team memory: how I'm connecting agents across sessions and repos so they can build on earlier decisions.
The economics: cost per completed task, failed attempts, coordination overhead, and what my subscriptions actually deliver.
The scripts: open source, so you can run the same measures on your own repos.
A proposed benchmark: what to count instead of PRs when agents do the work.
The lessons: mistakes, bottlenecks, and changes you can apply to your own setup.
The experiment is still running. I'm taking on harder tasks, improving reliability, and reducing cost. I'll bring results through the end of October, alongside what worked and what I had to change.
I'm documenting this as a practical playbook. Attendees will get supporting data and implementation materials, including guides for people and instructions your AI can read and adapt to your organization.
Want to try it before the event? Message me. You can also explore the existing guides at https://HamedTaheri.com and send me your questions.
How do you know a jump like that is real? DX's Q1 2026 data, from 64,680 developers at 219 companies, puts the median daily Cursor user at 4.1 merged PRs a week. I'll show a year of our merged PRs against that line, month by month, and cross-check it against several other productivity measures.
Hosted by Hamed Taheri. LinkedIn event: https://www.linkedin.com/events/7518747627350372352/