Wooldridge Connected AI Design Workflow

// When should you use the Connected AI Design Workflow?

Use this skill whenever you are starting or managing a design project (UX, UI, or product) that involves client meetings, design documentation, and iterative execution. Especially relevant when you feel behind on AI tooling or overwhelmed by too many disconnected tools.

// What do you need before starting this workflow?

• Project typerequired
  What kind of design project is this? (e.g. mobile app UX, dashboard UI, brand design, design system)
• Meeting or brief contentrequired
  Any notes, transcripts, or summaries from kickoff calls, client meetings, user research sessions, or design critiques
• Existing documentation
  Any design specs, PRDs, user interview notes, or briefs already written
• Deliverable targetrequired
  What needs to ship? (e.g. first-pass UI, prototype, design spec, component)
• Team context
  Are you working solo or with collaborators who need access to documents?

// What principles make a connected AI design workflow work?

// How do you apply the Connected AI Design Workflow step by step?

1. 1

   Capture the meeting passively — do not interrupt it

   Run an ambient meeting transcription tool (e.g. Granola) in the background during any project-relevant meeting: kickoff, client call, user research session, design critique. Do not join a bot to the call. Let it run locally and quietly. After the meeting ends, retrieve the structured summary — key decisions, action items, open questions — not a raw transcript. For in-person interviews where you are too focused to take notes, use a mobile version of the same tool.

2. 2

   Drop the structured meeting output into your Single Source of Truth

   Paste the enhanced meeting notes into a dedicated doc inside your interconnected workspace (e.g. Notion). This is not just archiving — this is building the context foundation every subsequent tool will draw from. All project documents (design specs, PRDs, interview notes) must live here too. Add collaborators so the whole team shares context, just as they would in a shared design file.

3. 3

   Enrich and refine documentation using built-in AI assistance

   Use the AI features inside your hub workspace to summarize long documents, answer specific questions about project content, or improve rough writing. The intended pattern is: jot things down quickly, then refine later using AI. Do not use this step to generate designs — use it purely to keep documentation clean, accurate, and usable.

4. 4

   Set up MCP connectors so your AI agent has live access to all tools

   Inside your AI coding/agent tool (e.g. Claude Code), navigate to Customize → Connectors → Browse Connectors. Search for and connect: your meeting tool (e.g. Granola), your document hub (e.g. Notion), your design tool (e.g. Figma), your code repository (e.g. GitHub), and any communication tools relevant to the project. Authenticate each. Once connected, the agent can read and write to all of them simultaneously without you copy-pasting. This step is a one-time setup per workspace.

5. 5

   Use voice dictation to write all prompts and document updates

   For any prompt longer than a sentence — especially when the AI is being stubborn and requires heavy guidance — use a local voice-to-text tool (e.g. Spokenly) instead of typing. Hold the hotkey, speak naturally, release. The text appears in the prompt field. This applies to drafting AI prompts, updating Notion docs, and writing emails. Prioritise tools that use local AI models for speed and accuracy over cloud-dependent dictation.

6. 6

   Prompt the AI agent to execute against your live context

   With MCPs connected and your Single Source of Truth populated, give the AI agent an open-ended but context-grounded prompt: tell it which documents to reference (by tool name) and what to produce. Example pattern: 'Based on my design spec and user interview notes from [hub], build the first version of [deliverable].' Do not paste content manually — let the MCP connectors pull it. The agent can reference multiple connected tools simultaneously in one prompt. Expect a first pass, not a final output.

7. 7

   Iterate using voice dictation for continued guidance

   Review the first pass. Where the output misses the mark, dictate correction prompts using your voice-to-text tool. This is where the speed advantage of speaking compounds — long, nuanced corrections are faster to speak than type. Continue iterating until the deliverable is shippable.

// What does the Connected AI Design Workflow look like in practice?

// What mistakes should you avoid in a connected AI design workflow?

• Collecting 15 tools at once with none of them connecting to each other — this leaves you feeling behind despite having more tools than you need.
• Using a raw transcript as project context. A wall of transcript is not usable. Only structured output — key decisions, action items, open questions — should enter the workflow.
• Skipping the Single Source of Truth setup and keeping documents scattered locally or across multiple platforms. Without a hub, MCPs cannot do their job and the agent works from a blank slate.
• Copy-pasting content into your AI agent instead of using MCP connectors. This is slower, error-prone, and defeats the purpose of the connected workflow.
• Using cloud-dependent or low-accuracy dictation tools that produce text far from what you actually said — this undermines the speed advantage of voice input.
• Treating the AI agent's first pass as a final output. The workflow is designed to produce a starting point, not a finished deliverable — iteration via continued prompting is expected.
• Setting up MCPs once and assuming they are always active — verify connections are live at the start of each major work session.

// What are the key terms in the Connected AI Design Workflow?

Single Source of Truth

The one interconnected document hub (e.g. Notion) where all project-related documents — specs, PRDs, meeting notes, interview summaries — live and can be accessed by collaborators and AI agents alike.

MCPs (Connectors)

Model Context Protocol connections — integrations that allow an AI agent to read from and write to external tools (Figma, Notion, GitHub, Granola, Gmail, etc.) in real time, eliminating manual copy-paste and giving the agent live context across your entire stack. Called 'connectors' inside Claude Code.

Structured Summary

The output of a meeting transcription tool that organises a meeting into key decisions, action items, and open questions — as opposed to a raw, unusable wall of transcript.

Context-First Execution

The principle that AI agent output quality is determined by the quality and completeness of context fed in upfront. Capturing meeting notes and documentation early is not busywork — it is what makes the AI agent's execution useful.

Voice-to-Text Dictation (Prompt Accelerator)

Using a local AI-powered dictation tool to convert spoken words into text for AI prompts, document drafts, and emails — leveraging the fact that speaking is three to four times faster than typing and produces fewer errors.

Connected Workflow

A design workflow architecture where a small number of tools are explicitly linked so that each tool's output becomes the next tool's input — making each tool more powerful by virtue of its connection to the others, rather than operating in isolation.

First Pass

The initial design or document output generated by the AI agent from project context. It is a starting point for iteration, not a finished deliverable.

// FREQUENTLY ASKED QUESTIONS