Leveraging the AI-Augmented CODE Workflow

Last Updated on 2026-04-19 by James Croft

Part 5 in my AI-Augmented Second Brain series

Over the past four posts, we’ve built the machinery: a structured vault, a knowledge graph, and custom AI skills that can operate reliably over that structure. Now let’s pull it all together by walking through my workflow to augment CODE with AI over my Obsidian Second Brain.

This post walks through how AI transforms each stage of Tiago Forte’s CODE workflow: Capture, Organize, Distill, Express. The methodology doesn’t change, but the capacity at each stage is fundamentally different when you have AI that understands your Second Brain, can traverse the knowledge graph, and follows grounded procedures.

Ownership of AI in CODE

What’s interesting across the various stages of CODE where I apply AI is the ownership. At the Capture end, AI does the heavy lifting. As you move toward Express, the balance shifts decisively. Distill and Express are fundamentally human-driven stages where AI acts as a capable research partner, retrieving knowledge and helping you shape your thinking, but the creative direction remains yours. The workflow moves from AI with human oversight to human with AI support.

Stage	AI Usage
Capture	AI-driven – correlates your activity across the knowledge base and external sources to capture what you experienced, from daily notes through monthly reflections
Organize	AI-driven – explores your journals and PARA notes, scans the knowledge base, discovers topics to synthesize deeper into PARA, and weaves every edit into the graph
Distill	Human-driven, AI-assisted – you bring emerging patterns to the AI; together you retrieve related knowledge and shape novel techniques from your experience
Express	Human-driven, AI-supported – you shape techniques and knowledge into consumable assets, with AI retrieving the raw knowledge and reshaping the content into desired outputs

With this in mind, let’s walk through each stage.

Capture – An AI-Driven Approach

Capturing is the easiest stage of CODE, and the one where the most information gets lost. Not because you forget to write things down, but because so much context lives in places you never think to check: your meetings you attended but didn’t take notes for, Teams threads where decisions were made in passing, emails with attachments you opened and read.

The Daily Check-In

For my knowledge base, I built a daily-checkin-generation skill that produces a concise daily status update by pulling from three sources simultaneously:

1. Journal notes – notes that were captured during the day’s meetings, either by myself, or combined with AI.
2. M365 activity with Work IQ – calendar events, Teams chats, emails, and files from the day, retrieved via the Work IQ MCP server.
3. Azure DevOps work items – work items assigned to me, with current status and discussion history, retrieved via the Azure DevOps MCP server.

The skill correlates these sources, deduplicates where both Obsidian and M365 cover the topics, and produces a structured check-in that captures the full picture of your day.

This is the power of external data context retrieval in action via MCP. WorkIQ is an MCP server that bridges M365 activity data into the skill’s workflow. The skill’s instructions tell it exactly what to query: “What meetings, emails, Teams chats, and files did I work on today?” The MCP server handles the API calls to Microsoft Graph; the skill handles the synthesis.

Following the pattern for Weekly and Monthly Synthesis

Following this same context retrieval pattern, you can start to see where you can apply this, alongside the Progressive Summarization technique, to other areas of note capture.

My weekly-journal-generation skill takes this further, readings the notes from the week, querying WorkIQ for activity, and producing a reflective weekly journal categorizing what went well, what didn’t go well, and what my biggest challenges are.

Following on from that, my monthly-journal-generation skill extends even further, capturing the chain to the monthly level. It reads all the weekly notes from the target month and produces a monthly journal that identifies themes, groups related achievements, elevating them to month level accomplishments.

Together, the journal skills form a temporal capture chain where each level synthesizes the one below. AI handles aggregation at every level; your role at the Capture stage is primarily reviewing the output, confirming the AI captured the right context, and filling in the gaps that only you can see. This is the most AI-driven stage of the workflow, and by design: the work being automated (correlating calendars, cross-referencing sources, deduplicating events) is exactly the tedious work that causes most people to skip capture entirely.

Bridging the Capture gaps with MCP

A thread running through all of this is the Model Context Protocol, the mechanism that lets AI reach beyond the vault to access external data sources.

I use MCP servers, loaded into my GitHub Copilot CLI, and reference their tools in my vault skills to follow the same principles as the knowledge base scan: gather context before acting. The purpose is to enrich my data with outside sources, contextually relevant to what I’m doing.

Plus, using MCP is extensible by design. You can build or adopt MCP servers for the tools and services you’re using today, and plug them into GitHub Copilot and reference them in the same skill instructions. The skill’s structure doesn’t change, only the data sources do.

Organize – AI-Driven but Human-Guided

Organizing is where AI shifts from recording what happened to structuring what it means. AI explores your notes and journals, scans the existing knowledge base for gaps and overlaps, discovers topics worth synthesizing into new PARA entries, and weaves every change into the knowledge graph.

This is the knowledge base scan pattern from Part 4 working at a higher level. You can find the vault-knowledge-retrieval skill in the companion starter template on GitHub.

AI still drives at this stage, but you guide the process. The AI does the heavy lifting of discovery and synthesis; you provide the judgment about what matters.

Modifying your Second Brain with Context

When you prompt the vault-note-creation or vault-note-update skills in the companion repository to work on your Second Brain, it doesn’t just stamp out a template and fill it in. It follows a five-phase workflow:

Phase 1 – Knowledge Base Scan. Before working on anything, the skills search all active PARA folders and Archive for existing notes on a topic. Every match is classified: duplicate, archived, related, or no match.

Phase 2 – Archive Reactivation. If an archived note exists on the exact topic, the skills move it back to its active folder, updates its frontmatter, and enriches it with new information.

Phase 3 – Enrichment. The skills query external sources, like WorkIQ, for internal context, and searches the public web for authoritative context. It synthesizes all sources in priority order: user input → internal context (e.g., journal, WorkIQ, ADO) → web research.

Phase 4 – Note Addition/Modification. The skills read the corresponding note template, populates frontmatter, fills sections following the callout prompts, and adds cross-links to related notes discovered during the scan.

Phase 5 – Reporting. The skill reports what was created, what was reactivated, what enrichment sources were used, and what cross-links were added, plus suggestions for additional notes that would strengthen the graph.

If you have an Area or Technique that’s grown to cover multiple distinct topics, the skills can split them out into focused notes, each cross-linked to the others. The graph gets more precise without losing any connections.

Distill – Human-Driven Technique Shaping

This is where the ownership balance tips decisively toward you.

Forte’s Progressive Summarization is about revisiting content and extracting the core insight, layer by layer, until you’ve distilled the essence. In the AI-augmented workflow, Distill is about pattern recognition: you’ve noticed something across your experiences, and you bring it to the AI to shape.

The pattern recognition prompt

It starts with your experience, something that you noticed. A recurring approach in your work, a pattern, a framework forming. You had the see of a technique, and it’s AI’s role to help you cultivate it.

You can bring a half-formed observation: “I keep seeing the same approach work when I’m delivering my customer engagements on AI resiliency, I think there’s a technique here.”

The vault-knowledge-retrieval skill becomes your research partner here. It searches across your Second Brain, reads matching Areas, Resources, and perhaps related Techniques, traces the graph, and pulls back everything in your vault that relates to the emerging pattern.

Now you have raw material that the AI has assembled for you, and you can guide AI through the synthesis process. Explain how these notes connect in your thought process, decide what the technique actually is, framing it.

Shaping the Technique with AI

This is collaborative knowledge work at its best. You articulate the insight; the AI retrieves the evidence. You decide where the boundaries of the technique are; the AI suggests cross-links to adjacent Areas and Techniques.

The vault-note-creation and vault-note-update skills can then take the shaped technique and embed it properly in the vault. But the intellectual contribution, the recognition that these experiences share a pattern worth naming, is fundamentally human.

This is what separates Distill from Organize. Organize is the AI discovering structure in your captured content and filing it into PARA. Distill is you discovering non-obvious patterns with AI in your accumulated experience and shaping them into techniques with AI’s help. The AI can retrieve, correlate, and suggest.

Express – Deciding how you want to share

The Express stage is where your knowledge becomes impact. Perhaps a blog post that shares a technique with your community, presentation slides for a team knowledge share, a career review that tells your professional story. This is the most human-driven stage of the workflow, and for good reason: expression is fundamentally about choosing what you want to say and how to say it.

AI doesn’t need to own this stage. But it can dramatically accelerate it.

From Knowledge to Artifact

There aren’t dedicated Express skills in my system, and that’s by design. Expression is too varied and too personal to reduce to a fixed procedure. Each blog post I write may require a different tone of voice or structure depending on the audience. One presentation may be for an exec-level audience, the other for deeply technical. You make those choices.

With the existing skills, and the overall capabilities in terms of output generation that AI can achieve, you can query your vault for details, pull in examples from across the vault, and orchestrate AI to generate the desired outputs for the desired audiences in a style of your choosing.

But even here, the AI produces a draft. You shape the final outputs, deciding which themes to emphasize, which stories to tell, what impression to leave. The same pattern applies to any expression artifact: the AI assembles and proposes, you refine and publish.

The Self-Reinforcing Loop

Beneath AI-augmented CODE, there’s a deeper dynamic at work:

Structure enables AI. The PARA organization, templates, and frontmatter give every skill reliable patterns to follow. The skill knows where to read, what format to expect, and how to write. This eliminates an entire class of errors.

AI strengthens the graph. Every skill-driven operation adds cross-links, fills metadata, and surfaces connections that we might miss. The graph gets denser and more accurate with every use.

The graph guides the AI. When a skill creates a new Technique and links it to three Areas, those Areas immediately reflect the new connection. When the skill scans the knowledge base before creating a note, the graph tells it what already exists, what’s archived, and what’s missing. The graph is both input to and output of the AI’s work.

This creates a compounding system. The more you use it, the richer the graph becomes, and the better every skill performs. A knowledge retrieval query today discovers notes that were enriched last week. The system rewards sustained use.

What You’ve Built

If you’ve followed this series from Part 1 through here, you now have:

• A structured vault with PARA folders, templates, and frontmatter metadata,
• A knowledge graph built from wikilinks, backlinks, and base queries,
• Custom AI skills grounded in your vault’s actual structure and conventions,
• An AI-augmented CODE workflow with a clear ownership gradient, from AI-driven capture through human-driven expression,
• MCP integrations that bridge the gap between what you documented and what you experienced,
• A self-reinforcing loop where every operation strengthens the system for the next one.

Finally, we’ll cover how to make this system your own: adapting templates for your domain, writing skills for your workflows, connecting your data sources via MCP, and leveraging the companion GitHub template repository.

What this series will build

Over the following articles in this series, I’ll walk you through building this system. Together, we’ll focus on:

• Why Your Second Brain Needs an AI Companion
• Designing a Machine-Readable Knowledge Base with Obsidian
• Building out your Knowledge Graph in Markdown
• Your First AI Skill for Knowledge Management
• Leveraging the AI-Augmented CODE Workflow (you’re here!)
• Making the AI-Augmented Second Brain Your Own

By the end, you’ll have a working Obsidian vault with your own AI skills that actively participate in your knowledge lifecycle.

And I’m sharing a companion base repository, obsidian-ai-second-brain on GitHub, for you to get started with your own Obsidian vault.

Found this article useful?

Thank you for taking the time to read this article. I’m committing to sharing more of my knowledge through my blog and open-source projects! You’ll also catch me in casual conversation on Bluesky and LinkedIn too.

If you enjoy what you see, please consider subscribing to get a notification when new articles go live!

---

This article is part of my AI-Augmented Second Brain series.

Next up: Making the AI-Augmented Second Brain Your Own.

You can find the companion template repository for an Obsidian vault at obsidian-ai-second-brain on GitHub.