Developer Workshop · Session 2

Making It Real

Running the full AI workflow — together, today

02

Why We're Here

The Goals

Structured context and collaboration — shared artifacts, consistent standards, AI gets the right information every time
Feedback loops — AI that validates its own work, not just generates output
Team-based AI workflow — one consistent process every developer runs, every task
Agent-ready codebase — PRDs, rules, and tests an AI agent can reason over independently

02

Recap from Session 1

12 Principles. One Workflow.

01Start with clear intent and context

02Brain-dump requirements

03Establish rules and guardrails

04Create a phased implementation plan

05Iterate with precision

06Test early and often

07Communicate clearly with AI agent

08Refine each layer systematically

09Maintain continuous documentation

10Deploy frequently

11Reflect, learn, adjust

12Up your own game

Now encoded as a /ai-workflow skill in Claude Code

03

Your Daily Driver

The /ai-workflow Skill

1

Intent & Context

What is this task? What does done look like?

2

Requirements Brain Dump

Must-haves, non-goals, edge cases, unknowns

3

Create the PRD

Foundation document — AI reads this first

4

Phased Plan

2–4 phases, each independently testable

5

Guardrails Check

PRD confirmed, conventions understood, tests ready

6

TDD Cycle

Write test → RED → implement → GREEN → refactor

7

Update the PRD

Code refs, decisions, phase marked ✅ COMPLETED

8

Continue or Close

Next phase → step 6, or final PRD review → ship

Type /ai-workflow in Claude Code to start any task

04

Mindset Shift

Go Slow to Go Faster

Invest upfront in clarity — ambiguity is the enemy
Structure before generating — artifacts first
PRD first, code second — every AI agent reads it before writing a line
Testability from the start — not bolted on later

Slower at the start → dramatically faster in every sprint after

05

Continuous Documentation

The Living PRD

You define at task start

Intent & goal
Business requirements
Phases — all ⏳ PLANNED
Success criteria

AI populates as it works

Code references (filepath:line)
Implementation decisions
Tasks marked ✅ COMPLETED
Troubleshooting & gotchas

Shared memory — not a spec doc. A new session reads the PRD, not the chat history.

06

The AI Workspace

A Dedicated Repository

A separate repo — not buried inside a product codebase
Every PRD lives here — created, maintained, and updated by the team
AI agents read from it at the start of every session
Versioned in git — decisions are never lost

docs/ ← all technical PRDs, one folder per feature

docs/ai-grading/PRD_AI_GRADING.md

docs/user-auth/PRD_USER_AUTH.md

templates/ ← PRD template the team uses

skills/ ← Cursor rules & Claude skills

agent_docs/ ← conventions, context, guardrails

06

Self-Correcting Systems

Feedback Loops

Single point of failure

One-shot prompt
No validation
Mistakes compound

Self-correcting loop

Tests
Linters
Validation
Iteration

The PRD is the contract — AI agents check their work against it at every step.

09

What Happens Now

Next Steps

Starting now

Create the AI Workspace repo on GitHub
Run /ai-workflow on the next new task
PRD-first for every new capability from here

Existing code (gradual)

Discover — map existing capabilities, identify boundaries
Modernize — document decisions and patterns already in place
Baseline — create technical PRDs for each existing feature area

New work: full workflow from day one. Existing code: backfill structured context, gradually.

07

Starting Today

The Commitment

Create the PRD before writing any code

docs/[feature-name]/PRD_[FEATURE_NAME].md — every task, no exceptions

Run /ai-workflow to start every task

Intent → brain dump → PRD → plan → guardrails → TDD → update

Update the PRD when the task is done

Code refs, decisions, phase marked ✅ — leave it current for the next session

One consistent workflow. One shared memory. One team.