The Workbench

Training Programme

The Familiar Relay Protocol

A hands-on team context engineering programme — teaching development teams how to build shared context infrastructure for AI, so knowledge compounds across people and sessions instead of resetting to zero. Complements your existing agile process, doesn't replace it.

The Problem

Your team uses AI every day. But every conversation starts from scratch. No one knows what the AI told someone else yesterday. Decisions get lost between sessions. New team members get no benefit from months of accumulated AI interactions.

Your standups, retros, and sprint planning already generate valuable knowledge — but none of it reaches the AI. The individual tools exist — prompt engineering, context files, memory features. But they don't solve the team problem. Everyone's still working alone, duplicating effort, and producing work that conflicts with decisions already made.

Amnesia

Every AI session starts from zero. Context built in one conversation is gone in the next. Multiply that across a team and the waste is enormous.

Duplication

Three engineers ask the AI the same architecture question. Three different answers. No one knows the others asked — or what was decided.

Conflict

AI produces code that contradicts decisions already made. It doesn't know about them. Nobody wrote them down where the AI — or the new hire — could find them.

No Compounding

Six months of AI usage produces no institutional knowledge. No shared memory. No accumulated intelligence. Every day is day one.

The question isn't "how much does AI training cost?"

It's "what's the cost of your team using AI without methodology for another 6 months?"

What We Teach

Context engineering as a team discipline that complements your agile process — not individual prompting tricks. Your team already produces documentation through standups, retros, and decision-making. This programme gives that knowledge a second purpose: the working memory of an AI squad-mate that participates meaningfully in your existing workflow.

For Engineers

  • Build persistent context that survives across sessions and people
  • Capture decisions with reasoning — so the AI understands why, not just what
  • Create handoff-ready work that others can pick up from context alone
  • Progress from basic context to genuine AI autonomy on real tasks

For Product Managers

  • Shape AI context from the product perspective — requirements, priorities, constraints
  • Contribute domain knowledge that makes AI output more relevant
  • Validate and verify AI work against product standards
  • Bridge the gap between what the team builds and what the business needs

Programme Structure

Four weeks. Three phases. Real work, not slides.

1

Foundation

Week 1 · 4 days on-site

Teams build shared context infrastructure from scratch on a hands-on exercise project. Every participant builds something real — interdependent services that require coordination, shared standards, and documented decisions. By the end of the week, someone else on the team can pick up your work and continue from context alone.

2

Sprint

Weeks 2–3 · Your real projects

Teams apply the methodology to their actual codebase, inside their existing sprint cadence. This is where habits form — the training moves from exercise to practice. Context engineering becomes part of your standups, your retros, your pull requests. Twice-weekly office hours provide guidance as the team adapts what they've learned to their own architecture, conventions, and workflows.

3

Refinement

Week 4 · 4 days on-site

Grounded in what actually happened during the sprint — not hypotheticals. Teams learn to maintain, prune, and evolve their context. Introduces the path toward autonomous AI workflows, built on the foundation they've already proven works.

The Progression Model

Most AI training stops at "better prompts." This programme builds toward a fundamentally different relationship between your team and AI.

Level 1

Orientation

"The AI knows this codebase."

Persistent context that gives AI genuine understanding of your project — architecture, conventions, decisions, and standards. Not a blank slate every session.

Level 2

Continuity

"The AI remembers what we decided."

Shared memory across sessions and team members. Decisions compound. Context is maintained, pruned, and evolved — not just accumulated until it's noise.

Level 3

Autonomy

"The AI has opinions, not just outputs."

AI takes on larger tasks with genuine understanding of team standards. Not blind delegation — informed autonomy built on proven context, verified against explicit criteria.

What Teams Walk Away With

Shared context infrastructure

Built by the team, fully understood — not installed from a template. The team owns it because they created it.

Explicit development standards

Processes, conventions, and "what good looks like" — written down where both humans and AI can follow them consistently.

Decision logging as habit

Not a chore. Reasoned decisions captured with why, not just what — creating institutional memory that compounds.

Faster onboarding

New team members — human or AI — orient from shared context without requiring someone to explain everything from scratch.

Confidence in AI output

Explicit standards enable verification against criteria, not blind hope. You trust AI the same way you trust a new hire — through shared, explicit process.

A clear path forward

Not a one-off workshop. A progression model the team can continue to develop, with each level building on the last.

Why This Is Different

Team methodology, not individual skill

Most AI training teaches individuals to prompt better. This teaches teams to work differently — with shared context that compounds across people.

Practice on real work

Week 1 uses a structured exercise. The sprint moves to your actual codebase. Week 2 is grounded in what actually happened. No hypotheticals.

Four weeks, not four hours

Habits don't form in a day. The sprint phase is where the methodology becomes practice — with guidance, not just a certificate.

Above the tooling

Context files are table stakes — everyone has them now. The gap is the methodology: how teams share, maintain, and evolve context together.

Works with your agile process

This isn't a new methodology to adopt — it's a complementary discipline that gives AI the ability to participate in sprints, retros, and standups as a genuine team member. You don't change how you work. You make AI capable of working the way you already do.

Who This Is For

Development teams

Teams of approximately 8 — engineers and product managers working on shared codebases. Not individual contributors; teams that need to coordinate.

Already using AI

Your team already uses Copilot, ChatGPT, Claude, or similar tools. You've seen the value individually — now you need it to work as a team.

Agile teams ready for the next step

You already have sprints, retros, and working agreements. You're past "should we use AI?" and into "how do we make AI part of how we already work?" This programme is the how.

Delivery

Duration 4 weeks (8 days on-site + 2-week sprint)
Team size Up to 8 participants per cohort
Location On-site at your offices
Tracks Engineering + Product (parallel or combined)
Sprint support Twice-weekly office hours during the 2-week sprint
Includes All materials, exercise environment