At a glance
WHAT IT’S REALLY ABOUT
DoorDash rolls out Claude Code to boost productivity across company
- DoorDash rolled out Claude tools broadly to “raise the floor” of AI fluency beyond chat-style usage, resulting in notable increases in engineering throughput and experimentation across functions.
- Fang describes a personal return to shipping production code by relying on Claude Code to write nearly all code, enabled by an inflection in model capability that made end-to-end workflows “just work.”
- As code generation accelerates, bottlenecks shift to CI/CD, review cycles, and security, prompting DoorDash to invest in AI code review agents, security scanning, and rethinking merge and approval processes.
- DoorDash built internal infrastructure (“Flux”) to run security-approved Claude sessions in the cloud and power agents (including code review) using Agent SDK and Claude models.
- The organization is learning that small, empowered teams plus standardized, shareable “skills” and written artifacts can compress timelines 3–5×, but cross-functional reviews and alignment become the new constraint for user-facing work.
IDEAS WORTH REMEMBERING
5 ideasTreat AI adoption as “raising the floor,” not just enabling power users.
DoorDash saw many employees still using AI like a chat; connecting tools to work systems (email, calendar, Slack) and rolling out Cowork broadly helped baseline productivity and AI fluency across roles.
Model upgrades can turn failed workflows into working ones—retry aggressively.
Fang’s first attempt to ship production code with an agent failed due to local environment setup; months later with newer models it worked, highlighting the need to revisit previously discarded approaches.
As coding speed increases, processes—not coding—become the main bottleneck.
Higher code volume stresses merge queues, CI/CD, and reviews, so DoorDash is investing in AI code review agents and reimagining pipelines to keep delivery speed aligned with generation speed.
Security must be embedded early, with fast procurement and clear guardrails.
DoorDash partnered closely with IT/security to enable rapid access to evolving AI tools while keeping security “non-negotiable,” avoiding slowdowns that would kill experimentation.
Create “agent-friendly” codebases with explicit architectural principles in-repo.
A successful team wrote dozens of architectural rules into markdown inside the repo so agents could consistently follow intended patterns, reducing drift and review churn.
WORDS WORTH SAVING
5 quotesIt wasn't until I was using Claude Code that I actually like was shipping production code in our code base again. So I like had a major comeback.
— Andy Fang
I set a goal for myself to like not write code manually. So, um, I try to have the agent, Claude Code, basically write everything.
— Andy Fang
There's really been an inflection point with the latest models of being able to just figure things out.
— Andy Fang
A challenge I've posed to the entire team is, like, try to get projects done in three to five X less time using AI to do so.
— Andy Fang
Security is, like, non-not negotiable.
— Boris Cherny
High quality AI-generated summary created from speaker-labeled transcript.
