What does a working agentic SDLC "factory" look like inside a enterprise today?
I'd like to hear from people who have put agentic workflows into a real SDLC at scale, beyond a pilot team for small and large orgs.
For my own side projects, I built a local pipeline that takes an issue to a reviewed PR. It works well for one developer. Doing the same for hundreds of engineers, with audit trails, security review, and cost controls, is a much harder problem.
A few things I'm curious about:
What runs your agents across the lifecycle? Claude Code, Codex, Copilot's coding agent, Devin, Factory, or something custom on LangGraph or a similar framework? Who owns it: platform, DevEx, or each product team?
How do you handle identity and permissions for agents that open PRs or touch infrastructure? Do they get their own service identities, or do they run with the developer's credentials?
What checks the output before it merges: evals, required human review, policy-as-code, or something else?
How are you measuring whether it's working? I'm interested in cycle time, cost per change, defect rates, or anything else that held up when finance started asking questions.
What did you try and then abandon?
Thanks in advance. I'll share what we land on.
One thoughtful comment unlocks the ability to create threads.
Vercel's is backed by their eve framework, they actually open source a software factory template: https://github.com/vercel-labs/eve-software-factory-template
I think Ramp is also best-in-class for this sort of thing, here are a few good write ups on their setup: https://x.com/_dylanga/status/2067999346780524753 https://x.com/josiah_arcanist/status/2121321789192081583
Stripe's pieces on how they built Minions are also quite good: https://stripe.dev/blog/minions-stripes-one-shot-end-to-end-coding-agents https://stripe.dev/blog/minions-stripes-one-shot-end-to-end-coding-agents-part-2