The Second Brain Is the Future of Work

I’m mostly apathetic toward the race to Artificial General Intelligence. We barely understand human consciousness as it is - how would we even recognize it in a machine?

My ambitions are more modest: I want Artificial Special Intelligence first. Systems capable enough to handle the repetitive and operational parts of my work at the same level of quality, while giving me back the time to sharpen the skills that actually matter.

The defining skill of the AI era will be preserving cognitive sovereignty in a world increasingly built on stochastic systems.

Your brain was always your greatest asset. AI is merely making that obvious.

You were never hired solely for your technical skills. You were hired for the judgment behind applying them - the reasoning, prioritisation, context, and intuition that sit underneath execution.

As more work gets delegated to AI agents, it raises a deeper question

How can I teach the AI what I know?

And the emphasis on “I” matters.

The diversity of human thought is what makes the world interesting. Companies scale through standardization, but they evolve through people who see the world differently. AI can replicate patterns, but innovation often comes from the people who deviate from them.

The real opportunity is not replacing human cognition, but externalizing it - building systems that can capture your context, your heuristics, your taste, and your way of solving problems.

One way to do that is to start building your own bespoke second brain today.

Not just a repository of notes, but an externalized model of how you think, decide, and create. Because the people who thrive in the AI era won’t be the ones who compete against machines - they’ll be the ones who learn how to compound their cognition through them.

Modelling a purely logical brain

Why not a RAG? Because it's an opaque black box. I wouldn't be comfortable if I could explain my thoughts and actions. I wouldn't want it to be any different with my artificial brain.

Ethos and pathos are profoundly difficult problems. Logos is tractable.

Human intelligence relies on all three — emotion, identity, intuition, social reasoning. But in the specific domain of work, logic and structured reasoning already create enormous leverage.

My goal is to build a second brain that can reason logically, compress knowledge into abstractions, learn patterns through usage, and externalize the way I think.

Just cognition with inspectable mechanics.

Why RDF, OWL, and Ontologies Are the Perfect Fit

The elegance of RDF is almost unsettling. It provides the common language to represent a concept

That’s it.

From that tiny grammar, you can model entire systems of thought.

Example:

OWL adds logical structure on top, also expressed in the same language of RDF.

You are no longer storing documents.

You are encoding relationships, invariants, and reasoning itself.

The simplicity is what makes it powerful.

Unlike embeddings or black-box neural systems, every fact is inspectable. Every inference is traceable. Every conclusion has provenance.

A logical brain needs a logical substrate.

RDF feels less like a database and more like a grammar for thought.

What makes it possible today?

The concept behind RDF is quite old. It has been around since the late 90s and was confusingly dubbed as Web3 (not to be confused with the fever dream that was Web3.0) predating even Web2.0. It has been used only in very specific domains (Wikipedia, Medicine, Knowledge Graphs) until now because of the laborious process that involves careful curation of data.

AI flips the script. The reason LLMs are so good as knowledge work is because they are modelled on similar lines, just stored as vectors instead of triples. The laborious process of the past can be delegated to an LLM today with a great degree of accuracy.

Most work is tolerant to some inaccuracy, unless you happen to write code for NASA. LLMs are hard workers that can work for you tirelessly until it meets a certain quality. This is what is possible today with tools like Claude Code and OpenClaw.

Perhaps the authors of RDF were prescient in naming it Web3 - it was an idea ahead of it's time.

There are plenty of open-source frameworks that can ingest and query RDF data for you - like Apache Jena and RDFLib. Plugging in an LLM at the input layer allows you to translate raw text into a strict syntax that encodes semantic meaning. Likewise, an LLM can also translate the output into a more human readable form.

Use LLM architecure for the hard stuff - I/O, visualizations Use classical architecture for the fun stuff - logic, inference

You want your second brain to grow with you. You don't want it to stagnate or become outdated. If you are working from a Claude Code terminal for much of your work, you could configure hooks to ingest and query from your second brain in a background agent.

You could also define skills to ingest data from a variety of sources by specifying your custom ontology - aka. your unique perspective on how to form the connections in your second brain. You could also rely on the freely available repositories of ontologies available online

You don't need a degree in psychology or neuroscience to understand your brain. You are your own walking talking laboratory to pick apart the mechanism of your thoughts

Here are some of my cherry picked findings that also translate well to this system

Lazy Evaluation via Inference

The brain is not a cache.

It computes understanding on demand - synthesising triples on demand.

Humans rarely store complete implementations. We store abstractions capable of reconstructing implementations dynamically.

It compresses thousands of experiences into a handful of reusable abstractions.

A senior engineer does not remember every class they have read in their life. They remember the fundamental concept of OOPs which is enough to understand any new class they encounter.

Give me the axioms, and I can derive the rest.

Pattern Matching

If two systems are structurally similar, understanding transfers instantly.

This is why experienced people learn faster.

They recognize shapes they have already seen before.

Intelligence is often analogy at scale.

This is where a human-assisted system works best for the second brain. LLMs could find patterns on it's own, but it often misses the connections that are obvious to a human.

Hebbian Learning

Neurons that fire together wire together.

Knowledge accessed together strengthens together.

The brain continuously reweights importance based on:

• frequency
• recency
• contextual co-occurrence

Understanding is dynamic, not static.

Learning Through Failure

Mistakes are not noise. They are training data.

The brain retains failed experiments because knowing what not to do is part of intelligence.

Good heuristics are often compressed pain.

Contextual Reinforcement

Concepts strengthen when encountered across multiple domains.

Pythagoras in geometry. Physics. Graphics. Signal processing.

True understanding emerges when abstractions survive context switching.

Provenance and Trust

Humans trust knowledge differently depending on source, confidence, and prior experience.

A logical second brain needs the same:

• where did this come from?
• why is it believed?
• how often has it been validated?
• what contradicts it?

Transparency is mandatory for delegation.

Cognitive Alignment

I do not need AI to be superhuman.

I need it to reason in ways I can predict.

The goal is not intelligence in the abstract.

The goal is alignment with my abstractions, heuristics, and mental models.

Not AI that thinks for me.

AI that thinks like me.

Hello dear reader. If you have managed to read this to the end, I have some news for you. I will be releasing an alpha version of this system soon. Drop me a DM if you're interested in learning more about it.