- Jenny Patchen of Levi Strauss said standards only work if engineers remember them, so her team built them into starter repos and agents.
- The same tooling reaches existing repos: an agent finds gaps against the templates and drafts fixes for an engineer to approve.
Jenny Patchen, a senior director at Levi Strauss & Co., described a way to make good practice automatic. Her team builds standards into AI agents and repo templates, so the secure, accessible route is the default one. She spoke on the GitHub show, in a GitHub Universe 2026 session about the company's multi-agent development process.
What was said
Patchen said Levi Strauss had already written down standards for observability, accessibility, security and more. The weakness, she said, was that these rules work only if each engineer remembers to apply them. In her words: "We wanted to make the right path, the easy path".
Repo templates are the main tool. She described them as starter kits shared across the company, giving every team one common engineering baseline. There are ten, and nine are tied to common programming languages. They set up build and release pipelines, documentation and runbooks from day one. Each also carries files that brief AI agents on the repo.
Every pull request then meets gates in the pipeline. These include dependency reviews, tests, coverage and a New Relic check. Patchen added that the tooling also works on existing repos. An agent compares the repo with the templates, finds gaps and drafts pull requests for an engineer to approve.
Why it matters
Our reading: a standard that lives only in a document is a standard that depends on memory. Patchen named that weakness herself. A default path moves the standard into the work, where a template, an agent or a gate applies it without anyone choosing to.
Patchen gave no figures for what skipping this costs, so any price is our inference. Still, a leader can ask pointed questions of their own organisation.
Which of your security and accessibility standards are enforced by a pipeline gate, and which rely on engineers remembering? Do new repos start from a template that already carries them? Could an agent assess your existing repos against that template and draft the fixes? What would that take?
Also ask who approves what the agents draft. At Levi Strauss, an engineer signs off each fix. A budget signer can ask for those three things: the gates, the gap-finding agent and the approval step.
The other side
Patchen offered no data on what a repo without these defaults costs. She also gave no count of existing repos brought up to standard.
Agents only draft the fixes, and she stressed that the human stays in charge. She warned that agents can run ahead with total confidence and deliver something that misses the intent.
Her line "Don't put the new motor in the old factory" is about process, not old repos. Her team redesigned its development process in a three-day workshop before adding agents. She also said the tooling was the easy part, and that getting people to change how they work was harder.
Written by the WebPulse Newsroom with AI assistance, and checked by our editorial review: every quotation was verified against the recording's transcript. How we use AI.
The conversation this talking point comes from
- GitHub: 17 agents, one SDLC: How Levi Strauss reimagined development on GitHub (2026-11-08)





