A Ferrari on Gravel

In a lot of development environments, applying AI tools is like driving a Ferrari on gravel. You can still do it, but you'll probably spin out when you try to go as fast as you can. Here's the foundation you need in place to get the most out of the tooling:

• CI/CD set up for multiple safe deploys per day
• Feature flags + (ideally) trunk based development
• Testing as a first class citizen - but balanced with how long they take to run
• Effective + fast suite covering the most critical flows + smoke tests
• More localized conditional suites where possible to get the right coverage on what's being changed
• Full regression suites to run less frequently
• Linters / static analysis
• Quick workflows to go from git pull to working dev environment
• locally, in a git worktree, or in a cloud environment
• Solid observability tooling for problems in production
• Mature, field tested technology choices that ideally can run locally on a fast computer
• (helpful but not required): typed, safe by default languages for compile time feedback

Don't forget:

• Some people with deep knowledge of how software works
• An understanding of an effective software development lifecycle + the whole value stream map and how to measure its effectiveness

The secret? it's the same foundation for effective software development with humans. The other secret? You can use AI tools to help you establish these foundations in situations where it might have seemed untenable before.

Get the right mix of humans and ai agents, and you'll be able to go real fast. Making sure you're going fast in the right direction? That's another set of problems for another post 😄

Join the conversation on LinkedIn