AI Second-Brain Projects
Give every recurring project its own persistent AI context.
- Difficulty
- Easy
- Time to result
- ~days to results
- Steps
- 6
- Confidence
- 97%
AI Second-Brain Projects organize recurring work into persistent, project-specific context spaces. Each space contains durable instructions, files, brand guidance, SOPs, prior preferences, and platform memory relevant to one workstream. New conversations opened inside that project inherit the context, allowing the user to move directly into analysis or creation rather than repeatedly explaining the background. Corrections made over time can also become part of the retained knowledge, so the assistant gradually reflects the user's standards. The mechanism turns an LLM from a sequence of disconnected chats into a continuing strategic partner with bounded organizational memory. Keeping one context space per project is important: separation reduces instruction collisions and makes the model's assumptions easier to understand, audit, and update.
Origin
Kevin Hudson described his Claude projects as an AI second brain on Marketing Against The Grain.
Core principles
- 01Persistent context eliminates repeated explanations.
- 02Each distinct project deserves a focused context space.
- 03Stable knowledge belongs in instructions and reference files.
- 04Feedback accumulated over time should improve future conversations.
- 05AI becomes more strategic when it remembers the project's goals and standards.
How to run it
- 1
Define the Workstream
Choose a recurring project with stable goals and a recognizable body of knowledge. State what belongs inside the project and what does not.
Pro tip Use project boundaries that match how you already organize your work.
Watch out A project that covers everything will accumulate conflicting instructions.
- 2
Create the Context Space
Open a dedicated project or equivalent persistent workspace in the chosen AI tool. Give it a clear name tied to its purpose.
Pro tip Use the same naming convention as your folders or operating systems.
Watch out Do not rely on a title alone to explain the project's purpose.
- 3
Install Durable Instructions
Describe the project, desired outputs, audience, voice, working preferences, and recurring rules. Keep instructions stable and testable.
Pro tip Include definitions for specialized terminology.
Watch out Stale instructions can silently degrade future outputs.
- 4
Attach Source Knowledge
Upload brand guidelines, SOPs, examples, and other authoritative files the model should consult. Remove superseded versions when they are no longer valid.
Pro tip Prefer concise canonical documents over many overlapping files.
Watch out Conflicting source files leave the model to guess which one governs.
- 5
Work and Correct in Context
Run related conversations inside the project and state clearly when an output violates your preferences. Allow useful corrections to inform subsequent work.
Pro tip Explain both what was wrong and what the preferred replacement looks like.
Watch out Moving related work into disposable chats forfeits the persistent-context advantage.
- 6
Maintain the Second Brain
Periodically inspect instructions, memory, and reference files for duplication or obsolete guidance. Update the context when the project changes direction.
Pro tip Schedule a review after major brand, strategy, or process changes.
Watch out Persistent context preserves mistakes as effectively as it preserves good guidance.
In the wild
A creator builds one AI project containing channel positioning, audience information, voice guidance, scripting rules, successful scripts, and production SOPs. Each new video conversation begins with this context already available.
→ Scripts become more consistent while repetitive setup prompting is reduced.
An agency creates a separate AI project for each client and uploads the client's brand guide, approved campaign examples, product facts, and review checklist. Team members use the relevant project for all drafting and analysis.
→ The model is less likely to mix client voices or require brand context to be pasted repeatedly.
Common mistakes
Mixing Unrelated Work
Combining several projects in one context allows irrelevant instructions and files to influence the wrong task.
Uploading Without Curating
A large collection of duplicated or contradictory documents is less useful than a small authoritative knowledge set.
Never Updating Context
Old strategy, voice, or process instructions continue shaping answers after the real project has changed.
Is it for you?
Best for
It is best for recurring projects with stable goals, reference material, terminology, or quality standards.
Not ideal for
It is not ideal for isolated questions that do not benefit from retained context or uploaded knowledge.
From the transcript
“I have a cloud project for every singular kind of project that I'm trying to do and I think of it as my AI second…”
“And then it has that baked into every conversation you start.”
“whereas like level two you kind of talk to it in a chat and then you move on to another chat and you just kind…”
From the episode
The AI Skill Ladder (Beginner → Workflow Builder)