Aviana Le · Creator + Founder · Building in publicInstagram @avianaleMogul Mastery waitlist

Mogul Maxxing  /  Blog  /  AI Second Brain

02 · AI Second Brain

What an AI second brain should do for a founder, and what it shouldn't

Most founders set up an AI assistant and then ask it whatever comes to mind. A second brain works better with a job description. Here are the seven jobs worth giving it, the ones worth keeping for yourself, and what the research says about where AI helps and where it quietly gets things wrong.

By Aviana LeNovember 202622 min read
Woman founder at a cream studio desk in morning light writing in a chocolate leather notebook beside an open laptop and a ceramic mug
The AI holds the context. The notebook holds the decisions. Keeping those two jobs separate is most of the design.

An AI second brain for founders should do the remembering, sorting, retrieving and first drafting that eat your work window, so your limited attention goes to judgment: choosing priorities, talking to customers and making the calls only you can make. In practice that means seven jobs: capturing loose thoughts, briefing you on context before you start, keeping your open loops, logging decisions and the reasons behind them, finding things with sources, drafting, and helping with a weekly review. It shouldn't set your priorities, be the only record of anything important, or be trusted on facts it can't show you a source for.

The short answer

Give your AI second brain a written job description. Its jobs: capture, brief, track open loops, log decisions, retrieve with sources, draft and review. Your jobs: decide what matters, verify anything factual, keep the master record in a place you control, and do the thinking you want to get better at. Research on AI at work suggests large gains on tasks inside what the models do well and real accuracy losses on tasks outside it, so the division of labor matters more than the tool.

Last month I wrote about how to use AI as a second brain without letting it do your thinking. That piece was the setup: what to put in, how to structure it and the six steps to get started. This one asks a narrower question that came up as soon as the setup existed: once the second brain is there, what exactly is it for? Without an answer, an AI assistant tends to become a very capable place to procrastinate.

I'm working through this in my own build. I have limited work windows, a first offer I'm validating in public, and more context than fits in my head. I'm not going to claim this job description has been proven in my business, because it hasn't been yet. What follows is the research behind it, the jobs I'm testing, and the reasoning for where I've drawn the lines.

Why your AI second brain needs a job description

The phrase "second brain" was popularized by Tiago Forte's 2022 book Building a Second Brain, which described a personal system of notes for capturing and reusing what you learn. The idea is much older. In 1945, Vannevar Bush described the "memex" in The Atlantic: a desk-sized device in which a person would store their books, records and communications so they could be consulted "with exceeding speed and flexibility", as "an enlarged intimate supplement to his memory".

What's new is that the supplement can now talk back. A note-taking app stores what you give it. A language model can summarize it, connect it, draft from it and answer questions about it. That's powerful, and it's also why the role gets blurry. An assistant that can do almost anything tends to be used for whatever feels most pressing in the moment, which, as the article on deciding what to work on first covered, is often the urgent thing rather than the important one.

A job description fixes that in the same way a founder operating system fixes recurring decisions: you decide once what the assistant is responsible for, and stop renegotiating it every time you open the chat window. It also makes it obvious when the assistant is doing a job it shouldn't have.

What the research says about handing thinking to tools

Offloading is normal, and it changes what you remember

Psychologists Evan Risko and Sam Gilbert reviewed the research on cognitive offloading in a 2016 paper in Trends in Cognitive Sciences. They defined it as using physical action to alter the information-processing requirements of a task, such as writing a reminder, tilting your head to read rotated text or saving a file instead of memorizing it. People offload constantly, and it often improves performance. Their review also described a trade-off: when people expect information to stay available externally, they tend to remember less of it themselves, and decisions about when to offload are driven partly by how confident people feel in their own memory, which isn't always accurate.

For a founder, that's the honest starting point. Offloading context to an AI second brain is a reasonable thing to do. It also means that whatever you offload, you'll increasingly depend on the system to get back. That's fine for meeting notes. It's worth thinking twice about for the reasoning behind your strategy.

The "Google effect" is weaker than its reputation

You'll often see a 2011 Science paper by Betsy Sparrow, Jenny Liu and Daniel Wegner cited as proof that the internet is ruining memory. It reported that people were less likely to recall information they expected to be able to look up later. The finding became known as the "Google effect" or "digital amnesia".

The evidence is shakier than the headlines suggest. A 2018 project led by Colin Camerer, published in Nature Human Behaviour, tried to replicate 21 social science experiments published in Nature and Science between 2010 and 2015, and the Sparrow study was among those that did not replicate. Later attempts have produced mixed results. A fair reading is that offloading probably does shift what we bother to remember, as Risko and Gilbert describe, but the dramatic version of the Google effect isn't settled science. I'd treat warnings about AI "rotting your brain" with the same caution.

AI helps a lot inside its frontier and hurts outside it

The most useful study for drawing the line is a 2023 field experiment by Fabrizio Dell'Acqua and colleagues at Harvard Business School, run with Boston Consulting Group. In it, 758 consultants worked on realistic tasks with or without access to GPT-4. On tasks within what the authors called the AI's "jagged technological frontier", consultants using AI completed 12.2% more tasks, finished 25.1% faster and produced results rated more than 40% higher in quality. On a task deliberately chosen to fall outside that frontier, consultants using AI were 19 percentage points less likely to reach the correct answer than those working without it.

The "jagged" part is the point. The frontier isn't a neat line between easy and hard tasks; some tasks that seem simple fall outside it, and some that seem hard fall inside. The study also observed two working styles among successful users, which the authors called "Centaurs", who split tasks between themselves and the AI, and "Cyborgs", who integrated the AI throughout their work. Both styles kept the human actively involved.

These were consultants on set tasks, not founders running their own businesses, so the exact percentages won't transfer. The direction does: an AI second brain should be given the jobs it's demonstrably good at, and kept away from the ones where a confident wrong answer is costly.

Gains tend to be biggest where experience is thinnest

Erik Brynjolfsson, Danielle Li and Lindsey Raymond studied 5,179 customer support agents at a software company that rolled out an AI assistant. Published in the Quarterly Journal of Economics in 2025, their study found that access to the tool increased the number of issues resolved per hour by 14% on average, with gains of 34% for novice and lower-skilled workers and minimal impact on the most experienced. The assistant seemed to spread the know-how of the best workers to newer ones.

There's a founder lesson here too. Your AI second brain will probably help most in the parts of the business where you're newest: a first sales page, a first pricing conversation, a first round of customer interviews. In the areas where you're already the expert, it's more likely to be a fast typist than a better thinker.

Confidence in AI can crowd out critical thinking

In a study presented at the CHI 2025 conference, Hao-Ping (Hank) Lee and colleagues at Microsoft Research and Carnegie Mellon University surveyed 319 knowledge workers who shared 936 real examples of using generative AI at work. Higher confidence in the AI was associated with less critical thinking, while higher confidence in their own abilities was associated with more. Workers also described shifting their effort from producing work to verifying and integrating it.

This was a survey of self-reported perceptions, not a measurement of thinking skills over time, so it shows an association rather than proof that AI makes anyone worse at thinking. But it supports a design choice: the more you trust the assistant on a given job, the more deliberately you should check its work on that job.

Even specialized tools still make things up

Finally, accuracy. Researchers at Stanford and Yale, led by Varun Magesh, tested AI legal research tools from two major legal publishers that were marketed as reducing or avoiding fabricated answers. In their study, published in the Journal of Empirical Legal Studies in 2025, the tools still produced incorrect or misgrounded answers, often called hallucinations, between 17% and 33% of the time on their benchmark queries. Those were purpose-built tools working over curated databases.

A general-purpose assistant answering questions about your business is in a harder position, because it only knows what you've given it plus what it learned in training. That's why "retrieve with sources" is on the job list below and "tell me the facts" isn't.

An AI second brain earns its place by remembering for you, not by deciding for you.

The seven jobs an AI second brain should do for a founder

These are the jobs I'm testing, chosen because they sit inside what current AI tools tend to do well, they eat a lot of founder time, and a mistake in any of them is usually easy to catch.

JobWhat it looks likeWhy it's a good fit
1. CaptureYou dump voice notes, half-ideas and meeting scraps; it tidies them into short, labeled entriesTurning messy input into clean text is squarely inside the frontier
2. BriefBefore a work window, it summarizes where a project stands and what was decided last timeSaves the context-rebuilding minutes that eat a short window
3. Keep open loopsIt maintains one list of unfinished items, each with a next step or a parked dateWriting plans for open loops can reduce their mental pull
4. Log decisionsIt records each meaningful decision, the options considered and the reason, in your wordsStops you from re-deciding the same thing; makes later review possible
5. Retrieve with sourcesIt finds what you said or saved, and points to the note, file or link it came fromRetrieval is useful; retrieval you can check is safe
6. DraftFirst versions of emails, outlines, FAQs and summaries that you then editDrafting is where AI showed some of its biggest speed and quality gains
7. ReviewOnce a week, it compares your plan, your decisions and what actually happenedSurfaces patterns you'd miss, while you decide what they mean

1. Capture: be the inbox, not the archive

The first job is the least glamorous and probably the most valuable. Founders generate a steady stream of half-formed thoughts: an objection a customer raised, an idea for a post, a thing to check with a supplier. Most of them get lost or sit in five different apps. Let the AI second brain be the single inbox. Paste or dictate whatever comes up, and ask it to turn each item into a short entry with a label, such as offer, content, admin or customer. The point isn't to organize perfectly. It's to make sure nothing important lives only in your head.

2. Brief: rebuild the context so you don't have to

The decision fatigue article described the familiar pattern of sitting down with sixty minutes and spending the first twenty working out where you left off. A briefing is the fix. At the start of a window, ask for a five-line summary of the project you're about to work on: the goal, the last decision, the next step and anything unresolved. Because it's summarizing your own notes, you can check it against them in a minute.

3. Keep open loops: one list, each with a home

Research by E. J. Masicampo and Roy Baumeister, which I discussed in the decide-first article, found that making a specific plan for an unfinished goal reduced its intrusive effects in lab studies, even though the goal wasn't done. An AI second brain is a good keeper of that list. Each open loop gets one line and either a next step or a date when you'll look at it again. You still choose the dates; it just makes sure none of them disappear.

4. Log decisions: the job most founders skip

This is the job I think matters most, and the one most AI setups leave out. When you make a meaningful decision, such as which offer to test first, what to stop doing or which channel to focus on, write two or three sentences on what you chose, what else you considered and why. Ask the assistant to file it with the date. Later, when the same question comes back, and it usually does, you can ask "what did I decide about this and why?" instead of reopening the debate.

Write the reasoning in your own words. The value of a decision log is that it captures your judgment at the time. If the assistant writes the reasons, you end up with a plausible-sounding record of a decision you may not have actually reasoned through.

5. Retrieve with sources: show me where it came from

Ask for answers that point back to your own material: "Based on my notes, what did the last three customer conversations say about price? Quote the relevant lines." Many current tools can search files or notes you've connected and cite them. Make citations a standing rule. If the assistant can't show where something came from, treat it as a guess, given the hallucination rates even specialized tools show.

6. Draft: first versions, never final ones

Drafting is where AI showed some of its clearest gains in the research, and it's a natural job for the second brain because it already holds your context. Outlines, email replies, FAQ answers, meeting summaries and the first pass of a sales page are all good candidates. The rule is that a draft is the start of your work, not the end of it. Anything with your name on it gets your edit, and anything factual gets checked.

7. Review: patterns, not verdicts

Once a week, ask the assistant to compare three things: what you planned, what you decided and what actually happened. It's good at noticing that a task has been postponed four times, or that a decision from three weeks ago hasn't been acted on. What those patterns mean is your call. This fits naturally into the weekly rhythm described in the founder operating system article.

A one-paragraph job description you can paste in

"You are my second brain for [business]. Your jobs: (1) turn anything I capture into short, labeled entries; (2) brief me on a project's goal, last decision, next step and open questions when I ask; (3) keep one list of open loops, each with a next step or a review date; (4) file my decisions with the date, options and my reasons, in my words; (5) answer questions from my notes and always show the source; (6) draft when asked, marked as a draft; (7) help me compare plan, decisions and results weekly. Do not set my priorities, do not state facts you can't source, and tell me when you're unsure."

What an AI second brain shouldn't do

The other half of the job description matters just as much. These are the jobs I'm deliberately keeping away from the assistant, and the reasons.

Not its jobThe riskWhat to do instead
Setting your prioritiesIt optimizes for what you asked, not for what matters most to the businessLet it organize options; you choose the one goal and the first task
Being the only recordTools change, chats get lost, accounts get closedKeep the master notes in files you own and can export
Stating facts without sourcesFluent, confident errors are hard to spotRequire a source, or treat the answer as a lead to verify
Replacing customer conversationsIt can only guess what your buyers thinkTalk to real people; use the AI to summarize what they said
Writing your reasoningYou end up with decisions you didn't actually think throughWrite the "why" yourself, even in one sentence
Holding sensitive data carelesslyClient, financial or personal details may be stored or used under the tool's termsCheck data settings; keep sensitive material out or anonymized
Thinking you want to get better atOffloading a skill you're building can mean you don't build itDraft it yourself first, then ask for critique

The last row is the subtle one. If you're learning to write sales copy, price an offer or structure a pitch, having the AI do it every time may get you a decent result today while skipping the practice that makes you better. One approach I'm testing is "me first, then the machine": I write the rough version, then ask for questions and criticism rather than a rewrite. It's slower. For the skills that compound, I think it's worth it.

Overhead view of seven blank cream index cards laid in a row on a linen tablecloth beside a brown pen, a ceramic cup and the corner of a closed laptop
Seven jobs, written down once. The useful part is deciding what's on the cards, and what isn't.

Mogul Mastery · founding waitlist

An AI second brain, set up for your business

Mogul Mastery is the implementation program I'm building in public for founders with limited work windows: decision rules, operating rhythms and AI used for context and decision support, set up around your own business. It's in development and shaped by the people on the founding waitlist.

Inside or outside the frontier? Founder tasks sorted

The jagged frontier idea is useful precisely because you can't always guess which side a task falls on. Here's how I'm sorting common founder tasks for now. It's a working map, not a finding, and it will shift as tools change.

TaskLikely sideHow to use AI on it
Summarizing your own notesInsideUse freely; spot-check against the source
Drafting emails and outlinesInsideUse freely; always edit before sending
Finding patterns in customer feedbackMostly insideAsk for quotes, not just themes
Market sizing and statisticsOften outsideUse only to find sources you then read
Legal, tax or compliance questionsOutside for decisionsUse to prepare questions for a professional
Choosing what to build or sellNot its callUse to lay out options; decide with real demand
Predicting what customers will payNot its callTest with real offers, as in the validation guide

How to tell if your second brain is doing its job

I'm not going to give you a target number, because I don't have one I'd stand behind. What I'm watching instead are a few simple, observable signals. They're the kind of thing you can notice in a week without a dashboard.

  • Window starts get shorter. When you sit down, you're working within a few minutes instead of reconstructing where you were.
  • Fewer re-decisions. Questions you've already answered get settled by looking up the decision log, not by debating them again.
  • Nothing important lives only in your head. When something nags at you, it's already on the open-loop list with a date.
  • You can check what it tells you. Answers about your business come with a source you can open.
  • You're still the one deciding. If you notice you're accepting its suggestions without reading them closely, that's the signal the Lee survey describes, and a reason to slow down.

If none of these are improving after a few weeks, the problem is usually the job description, not the tool. Most often the assistant has drifted into doing everything a little, instead of a few things reliably.

Privacy and ownership: decide this before you start

A second brain is only useful if you put real context in it, and real context includes things you wouldn't post publicly: client names, revenue, plans, personal circumstances. Before you load it up, read the data settings of whatever tool you use. Find out whether conversations are stored, for how long, whether they can be used to train models and whether you can turn that off. Business and team plans often have different terms from personal ones.

Two practical rules help. First, keep the master record in files you own, such as a folder of plain documents or a notes app that exports cleanly, and let the AI read from it rather than holding the only copy. Second, keep genuinely sensitive material out, or anonymize it. "Client A raised a concern about timing" is usually enough context for the assistant to help. This isn't legal advice, and if you handle regulated data, such as health or financial records, check the rules that apply to you.

Common mistakes when setting up an AI second brain

Treating it as an oracle instead of a memory

Asking the assistant what you should do is tempting, and its answers are often reasonable-sounding. But it doesn't carry the consequences of the choice, and it doesn't know what you haven't told it. Ask it what you've said, decided and saved; decide the rest yourself.

Giving it every job at once

Start with two jobs, such as capture and briefing, and add the others once those feel routine. A second brain that does seven things badly is worse than a notebook.

Skipping the decision log

Without it, the second brain remembers facts but not reasoning, which is exactly the part you'll want back in three months.

Letting chats become the system

A long conversation history isn't an organized record. Ask the assistant to write important outputs into your notes or files, where you can find them without scrolling.

Not checking the confident answers

The answers that sound most certain are the ones most worth verifying. Fluency isn't accuracy.

Outsourcing the skills you're building

If a skill matters to your future as a founder, do the first draft yourself and use the AI as a critic. Save full delegation for the work you never want to get better at.

Woman founder in a charcoal knit sweater reading a notebook by a tall window at dusk under a warm lamp in a minimal cream studio, laptop closed beside her
The weekly review is a human job with a machine assist: the AI lines up the pattern, you decide what it means.

Frequently asked questions

What should an AI second brain do for a founder?

It should handle seven jobs: capturing loose thoughts, briefing you on context before a work window, keeping your open loops, logging decisions and their reasons, retrieving information with sources, drafting first versions and helping with a weekly review. The founder keeps the judgment, including priorities and anything that needs fact-checking.

What is the difference between an AI second brain and a notes app?

A notes app stores what you give it. An AI second brain can also summarize, connect, retrieve and draft from that material. That makes it more useful, but it also means it can produce confident answers that are wrong, so it works best when it reads from notes you own and shows its sources.

Should an AI second brain make decisions for me?

No. It can lay out options, remind you of past decisions and summarize evidence, but choosing priorities and making calls about your business should stay with you. Research on AI at work found that people using AI on tasks outside its strengths were less likely to reach the correct answer than people working without it.

Does using AI make you worse at thinking?

The evidence is not settled. A 2025 Microsoft Research survey of 319 knowledge workers found that higher confidence in AI was associated with less critical thinking, but it measured self-reported perceptions, not skills over time. Popular claims about technology destroying memory, such as the Google effect, have also had replication problems. Keeping the reasoning and verification yourself is a sensible precaution.

What is the jagged technological frontier?

It is a term from a 2023 Harvard Business School study with Boston Consulting Group. It describes the uneven boundary between tasks AI does well and tasks it does poorly. In the study, consultants using AI did much better on tasks inside the frontier and worse on a task outside it, and the boundary was not always obvious in advance.

How do I stop my AI second brain from making things up?

You cannot fully stop it, but you can reduce the risk. Ask it to answer from your own notes and quote or link the source, treat any unsourced fact as a lead to verify, and keep it away from decisions where a wrong answer is costly. Even specialized AI legal research tools produced incorrect answers 17 to 33 percent of the time in one 2025 study.

What is a decision log and why should my AI keep one?

A decision log is a dated record of meaningful decisions, the options you considered and your reasons. Keeping it in your AI second brain lets you look up what you decided and why when the same question returns, instead of debating it again. Write the reasons in your own words so the log reflects your actual judgment.

Is it safe to put business information into an AI second brain?

It depends on the tool and its settings. Check whether conversations are stored, for how long and whether they can be used for training, and turn that off if you can. Keep your master records in files you own, and leave out or anonymize sensitive client, financial or personal details. If you handle regulated data, check the rules that apply to you.

Where should a founder start with an AI second brain?

Start with two jobs, capture and briefing, for a couple of weeks. Once those are routine, add the open-loop list and the decision log, then drafting and the weekly review. Write a short job description for the assistant so its role does not drift into doing everything a little.

Where this goes next

This article goes deeper inside the AI second brain theme that started with how to use AI as a second brain, and it builds on the decision rules in how to decide what to work on first. Next in the series: a 60-minute work window plan that puts these jobs inside a single hour, and then the difference between a tool, a workflow and a business system. If you run more than one project, the multifaceted entrepreneur article covers how to keep their contexts separate; if you're turning an audience into a business, start with from content creator to founder.

To get each new article as it's published, join the Mogul Movement. If you'd like an AI second brain and decision rules set up around your own business, the Mogul Mastery founding waitlist is open. You can browse every article on the Mogul Maxxing blog, or go back to the homepage for the full series.

Sources

This article is general information about research on AI, memory and decision making, not professional, legal, financial or data-protection advice. Study findings describe averages in specific settings and may not apply the same way to every person, tool or business. Check the terms and data settings of any AI tool before you use it with business information.