AI assistants are the strongest cognitive augmentation we have had so far. Not in the abstract sense. In the sense that they close specific gaps, quietly, for specific people.
We are not equal in raw cognitive power. You did not choose your working memory any more than you chose your height. Some people hold a whole lecture in their head. Some can quote the Iliad. Most of us got something narrower, and no amount of character or study moves that ceiling much. Same as with the body: muscle mass, explosive power, metabolism all have limits that training reaches and then stops at.
To be clear, this does not predetermine your life choices, your success or your happiness. It defines which tasks cost you more than they cost the person next to you.
On top of that, there are conditions, inherited or acquired, that affect how clearly you think and how well you focus.
People have always looked for ways to adapt. Cognitive chips, the pill from Limitless. What we actually got is less cinematic and available today.
For me, the useful part is how these tools fit the limits I already work around. Here are mine, and what I do about each.
1. Working memory
Problem: I catch the key parts and lose the detail. For years I solved this by taking notes I never processed, which left me with a pile of context-free fragments instead of a memory.
Solution:
- I keep a daily note in raw form, using a template with sections for call notes, task candidates and open questions.
- A Claude scheduled task runs every evening for note triage. It pulls the note apart and files each item where it belongs: projects in Obsidian, project docs and tasks in Linear, personal tasks on an Obsidian Kanban board.
2. Attention regulation
Problem: on a video call, my attention drops after 10-15 minutes unless I am the engaged speaker. If something is not addressed to me and not written down, it is gone.
Solution:
- I move conversations toward text wherever I can.
- I capture live calls with Fathom or Read AI.
- Recordings and anything with audio go through MacWhisper.
- I extract what I need afterwards, and the result lands in project knowledge or personal notes instead of in my memory.
3. Learning modality
Problem: I do not learn from video or lectures, partly because of the point above. I prefer to read at my own pace.
Solution:
- For specific content, I take the transcript where one exists and generate one where it does not, then read at my own speed and apply it immediately.
- For online courses, I look at the curriculum and use Perplexity to find available text alternatives, such as articles and books.
4. Information load
Problem: with several contexts at once, I get information overload, then mental fog.
Solution: I hand the work to agents.
- I use scheduled tasks in Claude Cowork for anything recurring, invoice submission for example. Claude Code when it touches files, Claude in Chrome when it touches a browser.
- I set detailed and tested guardrails.
- I review outputs and set new tasks according to one rule: one task at a time.
5. Abstract reasoning speed
Problem: blockchain, stock market terms, macroeconomics, higher math. All of it lands slowly for me.
Solution:
- I use an LLM to adapt content for me by simplifying it and explaining it step by step.
- I create Mermaid diagrams and HTML prototypes.
- I create short questionnaires to test my knowledge.
That's me. If none of this sounds familiar, you got a good hand. Worth knowing that about yourself too.
I built the first version of all of this by hand, long before any of it could be automated. It worked, and it cost me a great deal of time, and nobody around me saw the cost because the output looked normal. That is the part worth naming. The workaround was always there. What changed is that it stopped being expensive.
I think this is one of the least discussed effects of the technology, far behind claims of a millionfold increase in efficiency, because “we helped someone understand the concept” does not make a pitch.
If you recognise your own limits, the useful move is small: write down what one of them costs you in a normal week, then find the single tool that removes that specific cost. Not a system. One limit, one tool.