AI + Duo Strategy
Overview
As we work toward greater Duo adoption and dogfooding GitLab in an AI-native world, I want to collaboratively develop an AI strategy for our organization. Below is a draft framework—I'm sharing this to spark discussion, not to dictate direction. I'd love your input, pushback, and ideas.
Draft Principles
-
Duo-First Development: Duo-First is a product and development principle. Before starting a task, Duo is mandated for:
- Breaking down issues into sub-tasks
- Generating Unit Tests before work starts
- Generating MR descriptions after work completes
- Reviewing code submitted
- And more Duo-First Strategy Recommendation
- Getting started: We pick 1-2 clear problems, decide which is most urgent, and make that mandatory first. We can delegate repeatable tasks until we have a broader AI ecosystem.
- Discussion: What tasks should be prioritized? Where would Duo add the most value for your workflow today? These answers are already evolving as part of the AI Transformation initiative that @dsatcher leads.
- First Duo, Second Claude: First Duo, then Claude if Duo doesn't work, is used at least 1x weekly on a previously manual task. Every team member needs to select 1 task and document it. This is to contribute towards a reduction in administrative toil and build muscles for AI-assisted work.
- Customer Zero Mindset: To be Customer Zero, we must be capable of selling the value of Duo to customers. When given a use-case for GitLab, every team member will be capable of suggesting how to use Duo to accomplish it.
- Agent-Paired Workstreams: On large projects with workstreams (Project Studio with 5 workstreams, Panel-based architecture with 5 workstreams, Harness Compete strategy with 3 workstreams), workstream DRI’s will be 1 agent alongside the 1 human.
- Monthly AI Hackathon for EMs: 1 day every month, we will run a Hackathon to jump start adoption from the EM level, getting them familiar with what’s possible within the context of our own company for team member encouragement. Then we’ll have show & tell. This will allow EMs to “be in the details” with their teams throughout every week.
- AI-First Resourcing: All headcount requests and software spend will be analyzed against trying something different with the use of AI or agentic AI.
How to Contribute
Please share:
- Comments on what resonates, what concerns you, or what's missing
- Suggestions for how to measure success
- Ideas for pilot programs or quick wins
Edited by Lee Tickett