Make AI-assisted architecture and engineering work repeatable and reviewable.
Architecture Workbench is a toolkit of reusable AI skills and structured workflows for architects, technical leads and engineering teams. It preserves decision rationale, agreed scope and check results so another person or AI agent can continue the work.
When technical work stays in chat, its context gets lost
When technical work lives in AI chats, the reasons for a decision, agreed scope and results of checks can get lost between sessions. Teams then have to reconstruct the context before they can continue.
Reuse the workflow. Apply a structured workflow to the next task instead of rebuilding instructions in each new conversation.
Keep decisions explainable. Retain the reasons, trade-offs and evidence so another person can review or revisit the choice.
Hand over without the chat history. Keep the agreed scope and useful state with the project so another person or agent can continue.
Keep people in control. Separate AI recommendations from human decisions, and implementation from review and merge approval.
02 / What is inside
Choose the skill for the outcome you need
Architecture Workbench is an extensible collection of separately available AI skills. Technical Decision helps you prepare and record a technical decision; AI-Assisted Change guides a repository change. Each works independently. You can use an approved specification from Technical Decision in your existing delivery process or pass it to AI-Assisted Change.
Architecture Workbench
Technical Decision
From a clear problem and scope, through comparing options, to a technical decision you can explain.
Choose this if
You are deciding what or how to build and need to clarify the problem, options and the decision rationale.
What it helps you do
A technical choice your team can explain, review and revisit without reconstructing the context from a conversation. When implementation is the next step, an approved specification gives the delivery team a clear starting point.
Useful for
People making or preparing technical decisions: technical leads, architects, product managers and consultants.
Includes
Installable AI skill ✓ Instructions ✓ Workflow ✓ Four playbooks ✓ Templates & prompts ✓ Current release ✓ Updates based on availability ✓ Support
View details
When this helps
You have a technical question to resolve, but the problem and scope are not yet clear. Information is scattered and several options are possible. Before comparing solutions, you need to establish what you are solving, why and where the boundaries are.
Start with a question, a note or ticket, competing options, or feedback that reopens a decision.
Outputs, as needed
A problem brief and option assessment where the choice needs more preparation.
A decision record with rationale, alternatives and consequences; it can serve as an ADR when the architectural context is included.
An approved implementation specification with scope and acceptance criteria when delivery is the intended next step.
Four steps to move forward
1
Define the problem and scope
Clarify what you are solving and why, the outcome you need, and what is in and out of scope.
2
Compare the options
Compare credible options based on the agreed problem, scope and constraints. Weigh their benefits and trade-offs.
3
Make the choice
Record your choice, reasons and consequences—or a decision to defer.
4
Prepare the handoff
If implementation follows, agree on what to build and how to check it.
Guidance for your own AI agent
Use the installable AI skill or start with the supplied manual prompt. The package includes reusable guidance and templates; no hosted account is needed.
aw-technical-decision
You stay responsible
AI helps clarify the problem, compare options and draft the record for the chosen solution. You approve the important assumptions and make the choices, including rejecting or deferring a proposal. This package prepares decisions and specifications; it does not implement code.
You still need to check the evidence and use relevant expertise. The package does not guarantee a better decision.
Polar handles the secure checkout.You’ll leave this site for payment. Polar processes the purchase and then returns you to the confirmation and access details.Purchase information.
Architecture Workbench
AI-Assisted Change
Change code with AI. Keep the scope and checks clear.
Choose this if
The change is already defined and needs implementation or review in the repository.
What it helps you do
The change, its agreed scope, current state and check results stay together. Another person or agent can continue the work, and a person has the evidence needed to decide whether to merge.
Useful for
Developers, technical leads and reviewers working with AI coding agents.
Includes
Installable AI skill ✓ Instructions ✓ Workflow ✓ Four playbooks ✓ Templates & prompts ✓ Current release ✓ Updates based on availability ✓ Support
View details
When this helps
A change can drift from the original request. When another person, session or reviewer takes over, the agreed scope and reasons can get lost in the chat history.
Start with a change request, an approved specification, an existing branch to review, or feedback on work in progress.
Outputs, as needed
An implemented change and observed check results when implementation is in scope.
Agreed scope and acceptance criteria, according to a change specification or step-by-step plan.
Review findings, unresolved risks and evidence for a human merge decision; a continuation record when work crosses sessions.
Four steps to move forward
1
Agree on the change
Check the codebase, scope and what would count as done.
2
Plan small steps
Arrange the work into steps that can each be checked.
3
Build and check
Implement approved steps and run their checks.
4
Review the evidence
Assess a fixed version of the change so a person can decide whether to merge.
Guidance for your own AI agent
Use the installable AI skill or the supplied manual prompt with your coding agent. The package includes guidance for scoping, planning, implementation and review, including templates for the outputs.
aw-ai-assisted-change
You stay responsible
You approve the scope and plan, assess the findings and decide whether to merge. Only the current capability’s permissions apply. Review assesses the change; repairs return to implementation. The skill never merges or deploys your changes, grants access or enforces permissions.
Check the agent’s work against your project’s evidence and tests. The package does not guarantee absolute correctness.
Polar handles the secure checkout.You’ll leave this site for payment. Polar processes the purchase and then returns you to the confirmation and access details.Purchase information.
03 / How to use it
Start with the work you already have
Describe a question, change or existing repository branch and provide the relevant project context. The skill selects the steps still needed, reuses approved work and keeps useful decisions, specifications and verification evidence with your project. You approve the scope and important choices; AI output alone is not approval.
What you need to use it
Bring a capable AI agent that can read the packaged Markdown files and your project context. Code changes also need repository access and project checks. Each versioned package includes a skill entry and a manual prompt path. Installation and behaviour depend on the AI environment; compatibility is not verified across every tool or model.
Architecture Workbench provides guidance and reusable workflows, not a hosted app or an enforcement system. Your team retains responsibility for expertise, review and permissions.
Next step
Tell me which decision or repository change you are working on and which AI environment you use.