Step 3: Build AI-Powered Workflows¶

Where You Are¶

You've just finished Deconstructing your workflow. You should have:

• Workflow Definition ([name]-definition.md) — every step in your workflow broken down with decision points, data flows, context needs, and failure modes

This file is your input. Build has three parts:

• 3.1: Design — decide how the workflow should be built
• 3.2: Construct — build the components
• 3.3: Run — launch and operate the workflow

3.1: Design¶

Before building anything, decide how the workflow should be built. The Design phase takes your Workflow Definition and produces an AI Building Block Spec — a complete blueprint that tells you exactly what to build.

Design covers:

1. Architecture approach — Choose between no-code (build in a platform UI) and code-first (build with APIs and SDKs). The model recommends based on signals like integration needs, deployment requirements, and technical comfort. This choice shapes which tools and building blocks are available in subsequent steps.

2. Architecture decisions — The model confirms your platform (the one thing not in the Workflow Definition), then extracts tool integrations, trigger/schedule, and browser access flags directly from the definition. It presents a single confirmation block with the platform, integration mapping, and trigger implications — you review and adjust. These gate all subsequent recommendations.

3. Execution pattern — Choose from four patterns based on what your workflow needs:

   Pattern	When to use
   Prompt	Sequential steps, human drives the process and provides all inputs
   Skill-Powered Prompt	Steps with repeatable sub-routines or moderate complexity
   Single Agent	Tool use required, autonomous decisions, multi-step reasoning
   Multi-Agent	Multiple expertise domains, parallel execution, review gates

4. Interaction mode — Interactive (real-time collaboration), Autonomous (unattended execution), or Hybrid (mix of both) — determined by your architecture decisions

5. Autonomy classification — Classify each step (Human → Deterministic → Semi-Autonomous → Autonomous)
6. Building block mapping — Map each step to AI building blocks (Prompt, Context, Skill, Agent, MCP, Project, API, SDK)
7. Skill candidates — Tag steps that should become reusable skills, with generation-ready detail
8. Agent blueprints (when applicable) — Platform-agnostic specification for each agent (name, description, instructions, model, tools, context, goal) — built into working agents by the model in 3.2

Design Your AI Workflow — the full Design guide with execution pattern decision flow and output format

Produces: [name]-building-block-spec.md — your AI Building Block Spec with architecture approach, architecture decisions, execution pattern with interaction mode, step classifications, skill candidates, agent blueprints (when applicable), code-first selections (when applicable), and implementation order.

3.2: Construct¶

The AI Building Block Spec tells you exactly what to build — and the execution pattern determines which steps you follow. The model uses your architecture decisions (platform, integrations) and resolves deferred decisions (specific platform offering, shareability, code comfort) to generate artifacts in the right format for your specific setup. Work through only the steps that apply to your pattern:

The model handles each step based on your execution pattern. See Construct for details on what you'll need to do yourself (gathering context, configuring MCP connections, building agents on your platform).

3.3: Run¶

The final Construct deliverable is a Run Guide ([name]-run-guide.md) — a plain-language walkthrough tailored to your platform, architecture approach, and technical comfort level. It tells you exactly what to do with the artifacts that were built:

1. What was built — Every artifact listed with what it does and where it lives
2. Setup steps — Numbered instructions for getting each artifact into the right place on your platform (menu paths, button names, what you should see when it's working)
3. First run — A guided test with sample input, expected behavior, and common first-run issues
4. What to do next — How to run it again, share with teammates, and when to revisit

The Run Guide is saved to [name]-run-guide.md so you can reference it later or share it with your team.

The first run is a test, not a final product. Expect to run, evaluate, go back to 3.2 Construct to adjust, and run again. This cycle is normal — most workflows need 2-4 iterations before they produce reliably good output.

You're evaluating: Did the output match what you expected? Were any steps skipped? Was the output specific to your business, or generic?

Run guide — how to choose between a normal chat and a project, and how to troubleshoot first-run issues

Test, iterate, repeat¶

Each time you run, you're testing. When something is off, go back to 3.2 Construct, adjust the relevant building block, and run again:

The workflow is ready when you can run it on a new scenario and trust the output without heavy editing. Until then, keep iterating.

Run guide > Iterating — detailed troubleshooting for common first-run issues

Your deliverables across Build:

The Building Block Spec is the design document — it captures what to build and why. The platform artifacts are the implementation — generated by the model using your architecture decisions and current platform knowledge. The Run Guide bridges the gap between "artifacts exist" and "workflow is running."

Plus any context artifacts and tool connections you set up along the way.

Many workflows stay at the prompt-plus-context level permanently — pasted into a chat whenever you need it. That's a feature, not a limitation.

Track and Version Your Work¶

As you build, two background practices keep your work organized and recoverable:

Register building blocks in your AI Registry. Each time you create a skill, prompt, or agent, register it in your AI Registry Notion database — name, type, description, and which workflow it belongs to. If you registered the workflow during Deconstruct, these building blocks link back to it. This gives you a searchable inventory of everything you've built, and makes it easy for your team to discover and reuse building blocks across workflows.

Commit source files to GitHub. The .md files for your skills, agents, and prompts are source code — they should live in version control, not just on your local machine. After creating or updating a building block, commit it to your GitHub repository. This gives you a history of changes, makes it easy to share with collaborators, and protects against losing work.

These aren't separate steps — they're part of the rhythm of building. Each time you finish a building block in 3.2 Construct: test it, register it, commit it.

Quick Reference¶

For deep dives on individual building blocks, see the Agentic Building Blocks reference pages.

Worked Examples¶

These three examples show complete AI-powered workflows at different levels of the autonomy spectrum — from deterministic automation to collaborative workflows to fully autonomous multi-agent pipelines. Each includes working building blocks you can install and study.

All Building Blocks¶

These are the working building blocks included across all three examples. Each one links to its source on GitHub — read the full definition, understand how it works, and adapt it for your own workflows.

How to Use These Examples¶

# Add the Hands-on AI marketplace (one time)
/plugin marketplace add jamesgray-ai/handsonai

# Install the business-first-ai plugin
/plugin install business-first-ai@handsonai

Related¶

• AI Use Cases — browse use cases by type (content creation, research, coding, data analysis, ideation, automation)
• Analyze AI Workflow Opportunities — identify which of your workflows are candidates for AI
• Deconstruct Workflows — break down workflows into structured definitions
• Agents & Skills — browse all available agents, skills, and prompts