Using AI Effectively
as a Development Team

The Problem

• Everyone is already using AI — mostly as individuals
• Fragmented context, different tools, different approaches
• No shared understanding across the team
• Results are inconsistent — quality varies

What Needs to Change

A Shared AI Workspace

What This Enables

• Better product thinking
• Better system thinking
• Clearer collaboration across roles
• More reliable delivery

The Simple Model

Why This Matters

• Understanding — shared clarity on what we're building
• Architecture — system-level decisions require human judgment
• Alignment — product, dev, and QA must move together
• Maintenance — AI output needs to be owned and evolved

What AI Is Good At

• Generating artifacts at speed
• Expanding and stress-testing scenarios
• Iterating quickly within a defined structure
• Maintaining consistency across outputs

What Humans Must Own

• Intent — what we're trying to accomplish and why
• Context — constraints, history, architecture
• Judgment — tradeoffs, edge cases, risk
• Validation — does the output actually work?

The AI Workspace

• Every developer uses a strong agentic coding tool
• Product, Dev, and QA collaborate in a shared AI workspace
• AI helps create and maintain artifacts for each role
• Templates create consistency — rules guide agent behavior

What Makes a Good Workspace

• Shared across roles — product, dev, and QA all contribute
• Versioned — changes are tracked and reviewable
• AI-readable — structured so AI can reason over it
• Continuously updated — it evolves with the project

Layered Workflow

Artifact Structure

Why Artifacts Matter

• They capture intent — what we decided and why
• They record decisions — so AI doesn't reinvent them
• They set expectations — acceptance criteria AI can reason over

Feedback Loops

Product

• Defines problems clearly — before anything is built
• Sets outcomes — what success actually looks like
• States constraints — budget, timeline, technical limits
• Writes acceptance criteria — what AI and devs work toward

Development

• PRD → Technical plan (TPRD) before touching code
• Tests first — define the contract before implementation
• Small increments — AI works better with bounded scope
• Continuous validation — never trust a big-bang AI output

QA

• Contributes test plans before implementation starts
• Expands scenarios — edge cases AI alone won't catch
• Validates early — not just at the end
• QA moves left — a key shift in the AI-assisted workflow

Conversations, Not Prompts

Go Slow to Go Faster

• Invest upfront in clarity — ambiguity is the enemy
• Structure before generating — artifacts first
• Testability from the start — not bolted on later

Principles I Follow

What These Principles Enable

• Better clarity before implementation begins
• Smaller, phased progress — easier to validate
• More reusable context for AI across the team
• Earlier validation — catch problems before they compound

Healthy AI Usage

Treat Instructions & Artifacts Like Code

• Agent instructions shape outcomes — keep them clear, minimal, intentional
• Version your rules and prompts just like source code
• Review changes to artifacts the way you review PRs
• Maintain them — they decay and need updating as the project evolves

• Think before generating
• Reduce ambiguity — iterate safely
• Maintain alignment across roles

Where Are You Today?

Questions &
Discussion