Skip to main content

Developing with Agents

Overview​

We treat AI tools as development partners, not just code generators. This handbook provides practical guidance for effective AI-assisted development.

Living Document

AI workflows evolve rapidly. These patterns reflect our current learning. Experiment, adapt, and share what works for your team.

The Problem​

Without structured approaches, teams experience:

  • Inconsistent results from the same AI tool
  • AI-generated code that doesn't follow team patterns
  • Knowledge gaps where effective workflows stay with individuals
  • Wasted time in endless back-and-forth with AI tools

Our Approach​

This guide helps teams discover effective AI collaboration through:

  • Proven workflows you can adapt to your style
  • Safe ways to experiment and learn what works
  • Structure for sharing successful patterns across the team
  • Gradual approach to building AI collaboration skills

What Good Looks Like​

Teams developing agentic workflows often find:

  • More efficient development with the right AI-human balance
  • AI-generated code that aligns with team standards
  • Growing confidence in AI collaboration over time
  • Better focus on strategy vs repetitive implementation
Goal

Each engineer finds their optimal AI collaboration style while contributing to team learning.

Workflow Framework​

Start here, then adapt: Spec → Plan → Build → Test → Review

What we've learned works:

  • Clear specs before coding (AI needs context)
  • Small, testable chunks (easier to validate)
  • Frequent feedback loops (catch issues early)

Adapt this based on your experience:

  • Skip steps that don't add value for your use case
  • Add steps where you find gaps
  • Change the order if it fits your workflow better
  • Share what works with the team so we all improve

Key Principles​

  1. Plan before coding - AI produces better output with clear context
  2. Work in small pieces - Incremental changes with clear success criteria
  3. Manage context actively - Clear AI context between phases, use documentation as single source of truth
  4. Safety first - Configure permissions, use feature flags, commit frequently
  5. Learn and improve - Track what works, share patterns with the team

Our Tool​

We use Claude Code as a team. To request access, contact your manager or check internal docs for setup instructions.

Getting Started​

Start with our quick start guide for a structured 5-day introduction.

Guides​

"Agentic workflows should become as reliable as using Git or CI/CD. Teams should trust that AI outputs are predictable, high-quality, and aligned with our principles."