Graeme Sutherland

Chief Automation Officer at

[cat @_catwu]

Excited to share our most powerful new Claude Code feature: dynamic workflows!

Mention "workflow" in a prompt and Claude will dynamically create an orchestration plan that it strictly follows, allowing you to confidently trust that every stage happens in the right order even across 100s of agents.

[Andrej Karpathy @karpathy]

LLM Knowledge Bases

Something I'm finding very useful recently: using LLMs to build personal knowledge bases for various topics of research interest. In this way, a large fraction of my recent token throughput is going less into manipulating code, and more into manipulating knowledge (stored as markdown and images). The latest LLMs are quite good at it. So:

Data ingest:
I index source documents (articles, papers, repos, datasets, images, etc.) into a raw/ directory, then I use an LLM to incrementally "compile" a wiki, which is just a collection of .md files in a directory structure. The wiki includes summaries of all the data in raw/, backlinks, and then it categorizes data into concepts, writes articles for them, and links them all. To convert web articles into .md files I like to use the Obsidian Web Clipper extension, and then I also use a hotkey to download all the related images to local so that my LLM can easily reference them.

IDE:
I use Obsidian as the IDE "frontend" where I can view the raw data, the the compiled wiki, and the derived visualizations. Important to note that the LLM writes and maintains all of the data of the wiki, I rarely touch it directly. I've played with a few Obsidian plugins to render and view data in other ways (e.g. Marp for slides).

Q&A:
Where things get interesting is that once your wiki is big enough (e.g. mine on some recent research is ~100 articles and ~400K words), you can ask your LLM agent all kinds of complex questions against the wiki, and it will go off, research the answers, etc. I thought I had to reach for fancy RAG, but the LLM has been pretty good about auto-maintaining index files and brief summaries of all the documents and it reads all the important related data fairly easily at this ~small scale.

Output:
Instead of getting answers in text/terminal, I like to have it render markdown files for me, or slide shows (Marp format), or matplotlib images, all of which I then view again in Obsidian. You can imagine many other visual output formats depending on the query. Often, I end up "filing" the outputs back into the wiki to enhance it for further queries. So my own explorations and queries always "add up" in the knowledge base.

Linting:
I've run some LLM "health checks" over the wiki to e.g. find inconsistent data, impute missing data (with web searchers), find interesting connections for new article candidates, etc., to incrementally clean up the wiki and enhance its overall data integrity. The LLMs are quite good at suggesting further questions to ask and look into.

Extra tools:
I find myself developing additional tools to process the data, e.g. I vibe coded a small and naive search engine over the wiki, which I both use directly (in a web ui), but more often I want to hand it off to an LLM via CLI as a tool for larger queries.

Further explorations:
As the repo grows, the natural desire is to also think about synthetic data generation + finetuning to have your LLM "know" the data in its weights instead of just context windows.

TLDR: raw data from a given number of sources is collected, then compiled by an LLM into a .md wiki, then operated on by various CLIs by the LLM to do Q&A and to incrementally enhance the wiki, and all of it viewable in Obsidian. You rarely ever write or edit the wiki manually, it's the domain of the LLM. I think there is room here for an incredible new product instead of a hacky collection of scripts.

[Codie Sanchez @Codie_Sanchez]

The best small businesses of the next decade will run on a barbell:

One side: leverage AI for ruthless speed and scale.

The other side: be aggressively, inconveniently human. Show up in person. Send the note. Remember the name. Do the thing that doesn't scale.

The ones who nail BOTH ends will eat everyone else’s lunch.

[JJ Englert @JJEnglert]

I just finished restructuring all my skills based on @AnthropicAI 's latest recommendations for how to build them.

Here's what's different, why it matters, and how you can do it too.

Our #1 AI newsletter (https://t.co/Aigg5QxrBp) skill used to be one long file. Voice rules, examples, subject line logic — all crammed into one document.

It worked pretty well. But Claude would drift every now and again. It would nail the tone in one section and lose it in another. And every time I wanted to fix one thing, I risked breaking something else.

The problem isn't Claude. It's how you feed it information.

When everything lives in one file, Claude tries to hold it all at once. Rules compete with each other. Examples get buried. The longer the file, the worse the output.

The fix: break one file into a folder of specialized files.

1. SKILL.md is the boss. It doesn't contain any rules itself — it just tells Claude which files to read and when. Like a playbook.

2. instructions/ holds the actual rules. One file for voice. One for subject lines. One for section-specific guidance. They never compete because Claude only loads what it needs for the current step.

3. examples/ is where Claude learns what good and bad look like. Good examples from real shipped work. Bad examples showing 12 common AI writing patterns to avoid. Claude reads these right before writing so the voice is fresh.

4. eval/ is the quality check. After every draft, two things run automatically:

5. A checklist with 9 pass/fail tests

6. An advisory board — 3 AI personas (Exec, Builder, Lurker) review the draft in parallel and give feedback

The workflow it runs:

- Load the rules
- Gather inputs
- Read relevant examples
- Write the draft
- Run the checklist + eval (9 tests)
- Run the advisory board (3 reviewers in parallel)
- Revise based on feedback
- Save and queue for human preview / review

Every step loads only what it needs, when it needs it.

Want to restructure your own skills? Paste this into Claude Code:

--

I want to restructure my Claude Code skill files. Right now my skills are single files that try to do everything. I want to break them into a folder system like this:

SKILL.md — the orchestrator that tells Claude which files to read and when

instructions/ — one file per set of rules (voice, formatting, section guides)

examples/good/ — annotated examples of great output

examples/bad/ — anti-patterns to avoid

eval/checklist.md — pass/fail tests that run after every draft

eval/advisory-board.md — AI reviewer personas that evaluate drafts in parallel

templates/ — output format templates

Phase 1: Read my existing skill files and identify every distinct concern (voice rules, formatting, examples, evaluation criteria, templates). Show me the audit before building anything.

Phase 2: Create the folder structure and move each concern into its own file.

Phase 3: Build SKILL.md as the orchestrator — it should contain no rules, just the step-by-step workflow pointing to the right files.

Phase 4: Build the eval layer with a checklist and 2-3 reviewer personas.

Phase 5: Run the skill on a real task and verify everything works.

Start with Phase 1.

[Sam Parr @thesamparr]

How to live a miserable life in 5 easy steps:

1. Don’t decide. Spend tons of time planning and thinking, very little doing.

2. Don't set goals. And certainly don't track progress (or lack thereof).

3. Switch from thing to thing every quarter or year. Mastery and compounding is for suckers. Starting from scratch will be easier.

4. Don't have a best friend. Tons of acquaintances will suffice! Loving a buddy like family and caring for them is gay!

5. Buy individual stocks. Index funds are dumb. You're smarter than 90% of hedge funds with billions in budget and thousands in headcount. Your picks will help you sleep easy.

[Graeme Sutherland @graemesuth]

@Alyn Eek! Note to self - maintain healthy relationship with Claude….. and my wife.

[Graeme Sutherland @graemesuth]

@godofprompt Market

[Graeme Sutherland @graemesuth]

@gregoryjohn @lovable @boltdotnew @supabase @buildcamp bugs

[Justin Brooke ❤️‍🔥 @IMJustinBrooke]

Whats cooler than building AI agents?

LICENSING AI agents for profit!

Licensing is what made Bill Gates so rich.

Want my AI Agent licensing agreement?

Just like + comment "send it" and I'll DM you the editable Google Doc..

[Graeme Sutherland @graemesuth]

Visiting with @IMJustinBrooke and @chaunna while they offer free mentoring to local small businesses at https://t.co/kS1UlLcxrE in Elizabethton TN

[Graeme Sutherland @graemesuth]

Interested in https://t.co/XSvLvbCciu Automations, AI (obvs 🙄) & Productivity? Over the next 30 days, I'll be writing 30 Atomic Essays.

Follow my Social Blog on @typeshare_co: https://t.co/mDAmHsB63t

[Graeme Sutherland @graemesuth]

Kicking off the new year. Maker mode activated. #ship30

[Graeme Sutherland @graemesuth]

I just hopped aboard Ship 30 for 30 from @dickiebush and @Nicolascole77!

Can't wait to learn the fundamentals of digital writing & start publishing every day alongside hundreds of others in the community.

Check it out!👇🏼🚢

https://t.co/12RvjZXqfU

[Graeme Sutherland @graemesuth]

@c_spags Congrats Chris... from a happy customer. 🙂

[Graeme Sutherland @graemesuth]

@itsedaxe Outreach

[Graeme Sutherland @graemesuth]

Update: She jumped. Twice. ❤️

[Graeme Sutherland @graemesuth]

My kid wants to jump off a big ledge thing in a forest park up North. They won’t let her. Asks daily if she’s tall enough. She’s not. But she is. She totally is... We’ll try again today. What’s your thing? You can’t, won’t, don’t because… But you can. Mine is writing this. 🙄

[Graeme Sutherland @graemesuth]

@IMJustinBrooke @neotyk Hit me up @neotyk 👋

[Justin Brooke ❤️‍🔥 @IMJustinBrooke]

Copywriters don’t want you to know this trick…

You can outsell some of the best copywriters with this trick and be just awful at writing.

Billy Mays, famous TV pitchman used to use it all the time.

To sell a saw, he once cut a car fully in half with that saw.

To sell glue he glued a semi truck to a huge boat by a rope and pulled the boat.

This trick is called “dramatic demonstration.”

And you don’t have to cut a car in half to make it work for you.

Another way of doing it is just loading up your page full of testimonials, reviews, or before & after photos.

Now if you can write copy, great, you’ll be even better with this.

But…

You could literally put a dramatic demonstration of your results on a page with an order button and outsell 80% of copywriters out there today.

Results do all the talking you’ll ever need.

Shhh 🤫

Don’t share this with any copywriters they’ll get all pissy.

They love their words.

Our secret…

Mmmmkay!

✌️❤️

[Graeme Sutherland @graemesuth]

@itsedaxe Automate.