How to build a Claude Code-powered second brain for agency work

For those who run or work in an company, handle shoppers, or personal something greater than a single workflow, you understand how the primary 45 minutes of the morning vanish into Gmail, Slack, Fireflies, the CRM, and no matter doc was final on display, simply attempting to recollect what mattered the day earlier than. 

It was once I’d be enthusiastic about a pricing determination I owed the company workforce, a roadmap name for my habit-tracking app, three Slack threads I’d half-read on my telephone the evening earlier than, and a gross sales follow-up that wanted to exit right this moment.

These days are over. About six months in the past, I rebuilt how I work utilizing Claude Code because the engine of what I’ll name my second mind. The Monday morning synthesis now takes a couple of minute.

Right here’s what I constructed, why, and the way you are able to do the identical.

Why most second-brain setups break down

The “second mind” concept isn’t new. Tiago Forte’s “Constructing a Second Mind,” the PARA technique, Notion, and Obsidian had been all constructed on the identical intuition: externalize what you’d in any other case attempt to bear in mind.

Capturing info works. Recall largely works. However the bit that provides worth is what occurs after recall: turning what’s saved into one thing you possibly can act on.

Three failure modes present up in nearly each implementation I’ve seen:

• Passive storage. Info goes in. It sits there. The one means out is search plus your personal reminiscence of what you tagged it. Assembly notes are an ideal instance.
• Context-switching tax. Even once you discover the correct observe, you’re nonetheless copy-pasting and re-prompting your strategy to a usable output.
• No motion layer. A second mind that may’t draft, retrieve, or execute duties can find yourself supplying you with extra work than you began with. Over time, it builds up into an extra of notes that simply results in cognitive overload.

Documenting duties isn’t the issue. The issue is that the paperwork and conversations are scattered throughout 100 totally different apps, and nothing reads throughout them.

What’s lacking is the layer on prime: one thing that may pull from all of it and really do the work.

Dig deeper: How to turn Claude Code into your SEO command center

How Claude Code adjustments the equation

Most general-purpose AI assistants are caught behind a chat window. They’ll reply questions, however they’ll’t attain into your file system, bear in mind what you stated final week, or do something exterior their very own UI.

Claude Code (Anthropic’s developer-focused agent) ticks 4 bins that, collectively, change how one can function each day:

• Native file system entry: It reads and writes inside an precise mission folder, accessing your native information like several listing.
• Persistent, structured reminiscence: It remembers throughout classes by Markdown information I curate.
• MCP integrations into the instruments you already use: The Mannequin Context Protocol lets it plug immediately into Gmail, Slack, Google Drive, HubSpot, and Scoro. No knowledge migration. No rebuilding workflows round it.
• An motion layer: It drafts inner docs, prepares evaluation, and handles the repeatable elements of my workflow.

The final one is essentially the most helpful: transferring from one thing that simply shops info to one thing that acts on it. That is the place I can save most of my time (and you may, too).

The 4 layers of an AI second mind

The structure of my second mind is organized into 4 layers, stacked on prime of one another.

1. Reminiscence 

That is contained in a small set of plain Markdown information. One describes me and what I work on. One holds operating info price holding, equivalent to pricing choices, shopper preferences, and retainer buildings. One defines the persona I need it to take when working with me.

These load routinely each session, so I by no means re-explain context.

The attention-grabbing bit is that the reminiscence grows by itself. Each dialog will get dumped right into a each day log. As soon as a day, a small job runs over that log and decides what’s price selling into long-term reminiscence.

Throwaway chat doesn’t make the reduce, and over six months, this has constructed a surprisingly correct mannequin of how I work and what every shopper wants.

2. Search

The long-term reminiscence file stays small on goal. So each each day log will get listed into an area database.

Once I ask one thing particular like, ‘What did we agree with this shopper on inner linking again in February?’ it pulls the precise dialog and context to assessment.

3. Abilities

A ability is a centered functionality I’ve outlined as soon as and may invoke by title: draft a shopper transient, construct a gross sales proposal, reply to an e mail in my voice, summarize a discovery name right into a scope of labor.

Every ability is small, single-purpose, and inherits the reminiscence layer beneath.

For instance, the ‘draft e mail’ ability isn’t generic as a result of it is aware of my tone, the recipient, and what we final mentioned, by coaching on all the information it critiques each day.

I intentionally prevented one large do-everything agent as a result of what I actually wished was a complicated assistant that may work on small, composable expertise that enhance every day to assist assist me whereas I work on the larger choices.

4. A heartbeat

Each hour through the working day, a small course of pulls a snapshot throughout my instruments, equivalent to new emails, calendar shifts, Slack threads I’m in, and HubSpot pipeline motion, and decides whether or not any of it wants me.

If it does, I get a Slack ping with a one-line abstract and, the place applicable, a beginning draft for me to work from.

Dig deeper: How a ‘client brain’ gives AI the context SEO work needs

The place it pays again hours each week

Right here’s the place the time really comes again:

Sooner context-gathering for shopper work

When a shopper emails for a standing replace, the second mind has already pulled collectively each Fireflies transcript, Slack thread, and HubSpot deal observe for that account.

What was once half-hour of looking out earlier than I write again is now 30 seconds of context, so I can spend the time on the reply itself, not the digging.

Sooner knowledge evaluation

Whether or not it’s analytics, rank-tracker knowledge, or Search Console info, the second mind can pull what I want with context for assessment.

Discovery to scope

New retainers used to take days of back-and-forth. Now the second mind reads the invention transcript, the e-mail thread, and historic knowledge on comparable engagements, and it produces a scope I can pressure-test relatively than construct from scratch.

All of it provides as much as working extra effectively and giving shoppers higher service as a result of the necessary stuff isn’t slipping by.

The guardrails that make this work

Establishing a device this highly effective wants guardrails. In any other case, you find yourself with an agent doing stuff you didn’t ask for.

Learn-only by default

Each integration begins read-only. So for Slack, calendar, and Gmail, it might probably see and draft, however nothing goes out with out me.

Write entry will get added one device at a time (and solely as soon as I belief the outputs from that device). An agent with broad write entry is one immediate injection away from doing one thing you’ll remorse, and that danger compounds the extra instruments you wire in.

Reminiscence hygiene issues

The temptation is to dump all the pieces into the device. Don’t.

Lengthy-term reminiscence ought to maintain issues that change how the agent acts: shopper pricing, ongoing choices, and methods of working.

Belief the draft, confirm the motion

Learn each e mail it drafts earlier than it sends. Learn each transient earlier than it lands with a shopper.

The purpose isn’t to take away your self from the work. It’s to provide you a head begin primarily based on what it is aware of, whilst you nonetheless use your personal experience to finesse the ultimate choices.

Dig deeper: How to train Claude to sound like your brand

construct your personal second mind

You don’t want my precise stack. Use the instruments you’re most comfy with, together with this course of.

• Decide the 4 or 5 locations actual choices dwell for you. Electronic mail, calendar, one messaging device, your CRM, and your activity device. That’s sufficient.
• Add a transcript layer. Calls are the place essentially the most context will get misplaced in company work.
• Construct reminiscence first. A ‘that is me’ file plus a each day log that will get distilled. Don’t transfer on till speaking to it seems like speaking to somebody who is aware of your online business.
• Add expertise one by one. Decide essentially the most repetitive factor you do and construct a ability for it.
• Add the heartbeat final. As soon as retrieval and expertise each work, give it a schedule. Begin with notifications solely. Add write entry slowly.

This can be a second mind, however don’t let it substitute your precise one

The purpose of a second mind isn’t to interchange your current one. It’s to be sure to will be as environment friendly as attainable together with your day-to-day work, offering extra worth on your workforce and shoppers.

The instruments to construct this didn’t exist 18 months in the past. They do now, and the time it takes to arrange is paid again inside a fortnight.

Dig deeper: How to build custom SEO reports with Claude Code and Google Search Console