Over the past few weeks, I’ve been living inside a strange little personal experiment.
On paper, it looks like I’ve been learning a bunch of unrelated tools: Hermes, Obsidian, Notion, Granola, Cloudflare Workers, AI agents, RAG, MCP, Discord workflows, n8n, meeting transcript systems, local knowledge bases, deployment patterns, internal automation, and probably a dozen other things I’ve already forgotten I touched.
But that’s not really what happened.
What happened is that my capacity expanded.
Not in the hustle-culture sense. I don’t mean I became more disciplined, woke up at 5 a.m., built a color-coded productivity dashboard, and finally became the optimized person I was always supposed to be.
Actually, almost the opposite.
I started designing systems around the fact that I am not naturally organized. I get curious. I chase threads. I lose context. I find something interesting, open twelve tabs, have one good thought, then accidentally bury it under three new problems.
For most of my life, that felt like a weakness.
Lately, I’ve started to see it differently. The problem was never that my brain was messy. The problem was that I didn’t have enough external structure to let the mess compound into anything useful.
AI changed that.
Not because “AI will replace thinking” or any of the other lazy takes. It changed things because, for the first time, I can have a system that follows me down the rabbit holes, remembers what we found, turns scattered thoughts into structure, and hands me back a more coherent version of my own curiosity.
That’s a very different relationship with technology.
The shift from tools to thinking environments
Most people still talk about AI as a tool.
A chatbot. A coding assistant. A way to write emails faster. A thing you ask for answers.
That’s fine. It is those things.
But the more interesting shift, at least for me, is that AI can become a thinking environment.
A place where ideas don’t just appear and disappear. They get caught. Connected. Revisited. Filed. Turned into plans. Turned into workflows. Turned into questions you didn’t know you should be asking.
That distinction matters.
A tool helps you complete a task.
A thinking environment changes the way you approach tasks in the first place.
Over the last few weeks, I’ve been building one of those environments around myself. Part of it is Hermes, which gives me an agent I can talk to from Discord, the terminal, and local files. Part of it is Obsidian, which gives me a local markdown knowledge base. Part of it is Notion, which is probably where more polished team knowledge belongs. Part of it is Granola, meeting transcripts, Claude/Codex-style coding workflows, and a growing pile of experiments.
The exact stack matters less than the pattern.
The pattern is:
- Capture what’s happening.
- Turn raw context into structured knowledge.
- Use that knowledge to make better decisions.
- Build small automations where the pattern repeats.
- Keep everything close enough to daily work that it actually gets used.
That sounds obvious when written out. It is not obvious when you’re in the middle of real work, real meetings, real Slack threads, real deadlines, and real life.
Most knowledge systems fail because they ask the human to become a librarian.
I am not a librarian.
So I’m building a system where my only real job is to dump the raw material. The assistant can do the filing.
Surface-level curiosity vs. working knowledge
One thing I’ve noticed about myself: I’ve spent a lot of time being surface-level curious.
I don’t mean that as an insult. Surface-level curiosity is underrated. It’s how you discover new areas. It’s how you collect the first few dots.
But there’s a point where “I’ve heard of that” starts to feel hollow.
I had heard of RAG. I had heard of agents. I had heard of MCP. I had used Notion, Obsidian, Slack, Zapier, n8n, and a bunch of AI tools. I knew the words. I could follow the conversations.
But following a conversation is not the same as owning the concept.
That’s been one of the biggest changes lately. I’ve been pushing past vocabulary familiarity into operational understanding.
Not just “what is an AI agent?”
More like:
- Where does the agent live?
- What tools can it access?
- What memory does it have?
- What permissions should it not have?
- What does it do when it is wrong?
- How does a human review its output?
- What should be automatic, and what should stay manual?
- What happens when this leaves a demo and enters a real team?
Those questions make the topic real.
The same thing happened with second brains. It’s easy to say “I’m building a second brain.” It’s harder to decide what belongs in Obsidian versus Notion, what should be private scratch space versus canonical team knowledge, how daily notes flow into project notes, how meeting transcripts become tasks, and how to keep the whole thing useful when you’re tired.
That’s the difference between collecting ideas and metabolizing them.
I’m trying to metabolize more.
The workshop and the office
One of the clearest mental models I’ve landed on is this:
Obsidian is the workshop. Notion is the office.
The workshop is where things are messy. It’s where you paste transcripts, half-formed ideas, meeting notes, research dumps, stray links, weird diagrams, and “this might matter later” fragments. You don’t need it to be beautiful. You need it to be safe to make a mess. The office is where reviewed knowledge goes. It’s cleaner. Shared. Permissioned. More useful to a team. People should be able to trust what they find there.
That split has helped me a lot.
I used to think the goal was to find the one perfect knowledge tool. Now I think that’s the wrong question. Different kinds of thinking need different rooms.
You need a place to be messy before you can make something clear.
This applies beyond software. It applies to career growth, writing, strategy, research, and probably any kind of creative work.
If you only have an office, you perform clarity too early. If you only have a workshop, nothing ever becomes usable.
The magic is in the pipeline between the two.
Agents as infrastructure, not gimmicks
Another theme I keep coming back to: agents are more interesting when you stop treating them like magic people and start treating them like infrastructure.
A lot of AI demos feel like theater. “Look, the agent booked a flight.” “Look, the agent browsed a website.” “Look, the agent used six tools in a row.”
Cool. But also: who cares?
The better question is where agents fit into existing systems.
For an agency, that might mean:
- A transcript becomes a task list.
- A project status update gets compared against the actual source of truth.
- A support request gets classified and routed.
- A client meeting produces follow-ups with owners and dates.
- A research thread becomes reusable internal knowledge.
- A messy Slack conversation gets distilled into a decision record.
None of this needs to be flashy. In fact, the less flashy it is, the more likely it is to be valuable.
Good internal AI systems probably won’t feel like sci-fi. They’ll feel like less stuff falling through the cracks.
That’s the part I’m most interested in right now.
Not “How do we replace everyone with agents?”
More like: “Where are the tiny leaks in attention, context, and follow-through that make work harder than it needs to be?”
That is a much better question.
The biggest unlock: context that survives me
My working memory is not a reliable database.
This is unfortunate, but true.
I can have a great conversation, pull out five important insights, feel like I totally understand the next step, and then two days later I’m reconstructing the whole thing from vibes.
The last few weeks have made me much more aggressive about externalizing context.
Meeting transcripts. Daily dumps. Project notes. People notes. Workflow notes. Research summaries. “Why we decided this” notes. Not because I want to become obsessive about documentation, but because I want future me to have a fighting chance.
There’s a specific kind of relief that comes from asking, “Wait, what did we decide about this?” and having the system actually know.
That relief compounds. It reduces anxiety. It reduces rework. It makes you more willing to explore because exploration no longer feels like it disappears into the void.
This is where AI feels less like a productivity hack and more like cognitive scaffolding.
It gives you more surface area to think with.
Going beyond the tutorial layer
A lot of people get stuck at the tutorial layer.
They watch the videos. Save the threads. Bookmark the tools. Maybe try one demo. Then move on.
I get it. I’ve done that for years.
But the real learning starts when you try to make something work in your own messy context.
It’s one thing to understand an “AI meeting notes workflow” in theory. It’s another thing to ask:
- Where are my meetings recorded?
- Can I access the transcripts?
- What format are they in?
- Where should tasks go?
- Who reviews them?
- What should never be automated?
- How do I keep sensitive client context safe?
- What does “done” mean?
That’s where your brain expands. Not from consuming another overview, but from forcing the idea to survive contact with your actual life.
The friction is the curriculum.
If a tool is annoying to set up, that teaches you something. If a workflow feels promising but breaks on permissions, that teaches you something. If an agent produces plausible nonsense, that really teaches you something.
You start to build taste.
And taste might be the most important skill in this whole AI era.
Not prompt hacks. Not tool lists. Taste.
Knowing what is useful. Knowing what is fake. Knowing when automation is helping and when it is just adding another layer of nonsense. Knowing when a human should stay in the loop. Knowing which problems are worth solving.
AI as a curiosity amplifier
One of the best parts of this whole experience is that I feel more curious, not less.
That’s the opposite of the fear I sometimes hear: that AI will make people passive, dependent, lazy.
Maybe it can.
But used well, it does something different. It lowers the cost of following your curiosity.
You can ask the dumb question. Then the follow-up. Then the more precise version. Then you can turn the answer into a note, compare it to another topic, ask how it applies to your work, and build a prototype.
The distance between “I wonder…” and “I made a thing” is shorter than it has ever been.
That changes your relationship with learning. You don’t need to wait until you understand everything before you begin. You can begin, and understanding can catch up through the process.
That has been huge for me.
I’m less afraid of new topics now. Not because they’re easy, but because I have a way to approach them. I can map the territory, identify the core concepts, build small experiments, and gradually turn fog into handles.
That’s the feeling I want more people to experience.
How I think others can expand their mind with this stuff
If I had to turn the last few weeks into advice, it would be something like this. Start with your actual life. Do not start with “I need an AI strategy.” Start with the parts of your day that leak energy.
Where do you lose context? Where do you repeat yourself? Where do decisions disappear? Where do you have the same meeting twice? Where do you know something important happened, but you can’t find it later?
That’s the doorway.
Then build a capture habit that is almost stupidly easy. A daily note. A voice memo. A transcript folder. A Discord message to yourself. Whatever you’ll actually use on a bad day.
Bad-day design is important. Anyone can maintain a system when they’re motivated. The system only matters if it still works when you are tired, scattered, busy, or behind.
Once you have capture, add synthesis.
Don’t just store raw notes forever. Have AI help you turn them into:
- decisions
- open questions
- project updates
- people notes
- task candidates
- reusable explanations
- patterns you keep seeing
Then review before you canonize.
This is the human part. AI can draft structure, but you still need judgment. Especially at work. Especially with client context. Especially when people are involved.
Finally, build automations only after you understand the workflow manually.
This is where I’ve had to slow myself down. It’s tempting to build the agent immediately. But if you don’t understand the shape of the work, automation just makes confusion run faster.
Do the thing by hand a few times. Notice the repeated steps. Then automate the boring middle.
The meta-skill: learning how to learn with machines
The real skill I’m building is not “using Hermes” or “setting up Obsidian” or “understanding Cloudflare’s AI stack.”
Those are useful, but they’re not the deepest thing.
The deeper skill is learning how to learn with machines. How to ask better questions. How to keep context alive. How to turn curiosity into artifacts. How to pressure-test ideas. How to build systems that remember. How to move between exploration and execution without losing the thread.
That feels like the frontier.
Not because AI knows everything. It doesn’t.
Because AI gives you a new kind of leverage over your own attention.
And attention is the whole game.
If you can direct your attention, preserve what it finds, and build on it over time, you become a different kind of person. Not overnight. Not dramatically. But noticeably.
You become harder to overwhelm.
You become faster at entering new domains.
You become better at seeing patterns across tools, teams, workflows, and ideas.
You stop saying “I could never understand that” quite so quickly.
That might be the part I’m most grateful for.
I don’t want a smaller internet
There’s a version of the future where AI makes everything flatter.
More generic posts. More fake expertise. More people asking machines to summarize things they never cared about in the first place. More slop.
I don’t want that.
I want the opposite.
I want AI to help people go deeper. To become more specific. More curious. More capable. More able to build private libraries of thought and public bodies of work. More able to connect the weird dots only they would connect.
That’s what I’ve been trying to do for myself.
I’m not finished. The system is messy. The notes are uneven. Some workflows are half-built. Some ideas will probably turn out to be wrong. Good.
That means it’s alive.
A few weeks ago, a lot of these topics felt adjacent to me. Interesting, but separate. Now they feel connected: agents, knowledge bases, workflows, automation, internal tools, memory, creative output, personal infrastructure.
I can feel a map forming.
And once you have a map, even a rough one, you move differently. You explore with more confidence. You notice more. You ask better questions. You become less dependent on someone else’s finished explanation because you know how to build your own.
That’s the thing I want to keep chasing.
Not just better tools.
A bigger mind.
