Workflow 24: AI Research Agent: Learn Any New Topic Faster

Anytime I want to learn something, I hand it to an agent first.

It goes and researches across Reddit, Twitter, Threads, YouTube, articles, wherever there's public information on the topic. Then it files everything into one notes database, and that becomes the place I go back to.

There are two versions. One runs once, when I'm learning something new. One runs on a schedule, when the thing keeps changing.

What this replaces

Twelve tabs, three YouTube videos you didn't finish, and a Reddit thread you can't find again.

The information for learning something new is rarely hidden. The trouble is that it's scattered, and you end up collecting it in six places and never reading any of them twice.

For anything moving fast, there's a second problem. By the time the articles catch up, the forums have moved on. So if you're following a tool or a field that changes weekly, you need something that keeps checking.

Blueprint

Pick your version.

The one-off. For learning something new. Triggered manually. It finds videos, articles, and the people worth learning from, then files everything in one place.

The recurring. For staying current. Runs weekly, monthly, whatever fits. It updates its own knowledge, writes a report, and sends it to you.

Both feed the same notes database.

Trigger: manual for the one-off, scheduled weekly or monthly for the recurring. Inputs: a topic, what you want to know and why, and public sources. Output: sources filed into one notes database, plus a report for the recurring version.

Organize

What you need first:

• A notes database. One. This is the piece the whole thing depends on
• A clear statement of what you want to know and why
• A place for reports to land: your inbox, Notion, a doc
• Some sense of who's worth listening to on this topic

One warning about learning from a specific person's approach. If you want to learn someone's method properly, keep their material in its own section rather than mixed with everything else. Blending four people's approaches together gives you something muddled that belongs to nobody.

Systematize

The one-off version:

You are my research agent. When I tell you I want to learn something,
go and find everything worth knowing about it.

For each topic I give you, find:
- YouTube videos worth watching, and who made them
- People genuinely worth learning from on this topic, and why
- Articles and written material
- What people are actually saying in public discussions: Reddit, Twitter,
  Threads, forums

Tell me for each source: what it covers, who made it, and whether it's
current. If something is out of date for a fast-moving topic, say so.

Then file everything into [NOTES DATABASE], organized by topic, with a
short summary of each source so I know what's in it before I open it.

If I want to learn one specific person's approach, keep their material
separate from the general research so their method stays intact.

The recurring version:

You are my research agent for [TOPIC]. Run every [WEEK / MONTH].

Each run:
1. Find what's changed since your last run. Check public discussions
   first, since they move before the articles do.
2. Compare it to what's already in [NOTES DATABASE]. Only report what's
   actually new.
3. Write a short report: what changed, what it means for me, and what's
   worth my attention.
4. Update the notes database with the new information.
5. Send the report to [EMAIL / NOTION PAGE / DOC].

If nothing meaningful changed, say that. A short honest report beats a
padded one.

How to set it up

1. Build the notes database first.
2. Start with the one-off version on something you're currently trying to learn.
3. Check what it files. Fix what's useless, and tell it why.
4. Once the output is good, decide whether the topic is worth a schedule.
5. Point the reports at whichever inbox or page you actually open.

Do I need to pay for all of this?

No. NotebookLM does a lot of this for free. You set up one hub, drop your sources in, and ask it questions. It'll also turn what you're studying into a podcast version or a video version, which is useful if you take things in better by listening.

Worth trying before you spend credits on the same job.

Assign

The agent searches, summarizes, files, and reports.

I decide what's worth learning, and whether a source is any good.

That second one matters more than it sounds. A research agent will happily hand you fifteen sources with no view on which are worth your time. It can tell you a video exists and what it covers. It can't tell you whether the person in it knows what they're talking about.

So the judgment stays with me, and the agent's job is getting me to the point where I can make it quickly.

Rule of thumb from this build: research that isn't filed in one place is the same as no research.

There's a small trap in the scheduled version worth naming. An agent that reports every week will find something to report every week, whether or not anything happened. Hence the instruction to say plainly when nothing changed. Otherwise you end up reading manufactured updates and trusting the reports less each time.

Integrate

The research agent feeds the same place everything else does:

• Everything lands in the notes database, so it's searchable later
• My brain dump sorter reads that same database, which means research and ideas sit next to each other
• Reports go to an inbox or a page I already check, not a new place to remember
• Anything I learn this way usually turns into content, since documenting what I'm learning is most of what I publish

The database is the point. An agent that researches brilliantly into a folder you never open has done nothing.

Next: Albus, the productivity wizard, and how he decides what actually gets done.