agentos.guide › blog

How To Learn More About AI Agents: The 30-Minutes-A-Day Loop I'd Run Inside An Agent OS

By Julian Goldie · 13 October 2026 · agentos.guide

How to learn more about AI agents once you know the basics is a routine problem, and the fix is thirty minutes a day inside a simple loop.

I'm Julian Goldie, and I test new agents and models inside my Agent Operating System almost every day.

Here is the path in five lines.

• You follow the release notes of the one agent you use most.
• You watch one build video a day and copy one thing from it.
• You join one community so you can ask when you get stuck.
• You test every new tool with the same fixed prompts.
• You turn each thing you learn into a small build within 48 hours.

The rest of this page shows where to learn, how to set the loop up and how to keep it running.

Where to learn more about AI agents, ranked by format

This is my own list, so my own channels are at the top, and I'm telling you that plainly.

I ranked the formats by how quickly they turn into a build.

Videos come first because seeing a tool run is the fastest way to judge it.

Papers come last because they rarely change what you build this week.

You don't need all nine, and three is enough to start.

The learning loop inside an Agent OS

An Agent OS is a dashboard that runs your agents under one shared memory.

That makes it a very good place to learn, because the system does the collecting and you do the building.

I think of it as a loop with four parts.

If you don't have an Agent OS yet, you can run the same loop with a notes app, a text file of prompts and a to-do list.

The parts matter more than the software.

If you want to build the real thing, my Agent OS build guide does it in five steps with the exact prompts.

Quick start: set the loop up in one sitting

This takes about an hour, and you only do it once.

• First, you write down the one agent you use most, because that is the tool you'll go deep on.
• Second, you open its releases page on GitHub, click Watch, choose Custom and tick Releases.
• Third, you subscribe to one video channel and join one community.
• Fourth, you create one notes folder called "agent learning" in the vault your agent can read.
• Fifth, you save three to five fixed prompts in a file called "test prompts".
• Sixth, you add one column called "learning builds" to your task board.
• Seventh, you put the weekly plan below into your calendar as seven thirty-minute blocks.

After that hour, the loop runs on thirty minutes a day.

The 30-minutes-a-day weekly routine

Each day has one job, and each job feeds one part of the loop.

Only two of the seven days are about taking in new information.

The other five are about building, testing, asking and reviewing, because that is where learning sticks.

I'd keep this routine for four weeks before changing anything.

🔥 Want the exact learning more about AI agents setup I run inside my Agent OS?

Inside the AI Profit Boardroom you get the installable Agent OS, step-by-step video tutorials, four coaching calls a week and 3,400+ members building real automations.

→ Get access here

Let your agents do the reading

This is the part most people miss when they try to learn more about AI agents.

Your agents can do a lot of the collecting for you.

My own example is a radar.

It is an agent on a schedule that wakes at six, reads live posts on X and scores every story.

When I published the guide to it, it had logged 49 mornings and scored 294 trends.

One morning I checked its ranked list against the stories I would have chosen by hand, and my two picks were sitting at number two and number three.

You can build a smaller version in an afternoon.

• You can schedule an agent to open three release pages each morning and summarise what changed.
• You can use the Hermes loop command to re-check one page on a timer, and my Hermes loop guide shows real sessions.
• You can point the Hermes learn command at a PDF or a book so it becomes a skill your agent keeps, which my learn command guide walks through.
• You can save every summary into a notes vault so nothing is learned twice, as shown in my Hermes second brain guide.

The agent brings you the short list, and you still choose what to build.

Official docs and changelogs worth bookmarking

Docs and release notes are written by the people who shipped the change, so they are the most accurate source you have.

They also move fast.

When I checked the Hermes Agent releases page on 13 October 2026, it listed four releases in about three weeks.

Version 0.21.4 came out on 21 September, version 0.21.5 on 24 September, version 0.21.6 on 8 October and version 0.21.7 on 12 October.

The Hermes Learning Path is a good place to go deeper.

Its intermediate tier covers sessions, messaging, tools, skills, memory and cron, and the page estimates two to three hours for it.

That is roughly five Thursday slots.

If you want a guided version, my Hermes crash course covers the same ground with screenshots from the system I run.

GitHub repos to watch

A repo shows you when a project was last updated, which no launch post can fake.

Every repo below had commits in October 2026 when I checked.

You don't need to read the code.

You set Watch to Releases only, and GitHub tells you when a new version ships.

Three or four watched repos is enough for one person.

🔥 Want the exact learning more about AI agents setup I run inside my Agent OS?

Inside the AI Profit Boardroom you get the installable Agent OS, step-by-step video tutorials, four coaching calls a week and 3,400+ members building real automations.

→ Get access here

Newsletters, blogs and X accounts

I kept this list short and neutral.

Each source below had published in the week before 13 October 2026.

One newsletter is enough, so you pick the style you like.

On X, I'd follow official accounts that each organisation links from its own website.

• @NousResearch is the lab behind Hermes Agent.
• @AnthropicAI is Anthropic's account.
• @GoogleDeepMind is Google DeepMind's account.
• @JulianGoldieSEO is mine, where I post launches and screenshots from my tests.

X tells you that something launched, and the docs tell you what it is.

How I test a new tool the day it launches

Running a tool teaches you more than reading about it.

My routine has three steps, and it is published on my GoldieBench about page.

• First, I wire the new model into the Agent OS, so I can send it work from the same board my other agents use.
• Second, I run my fixed prompt set through it with one shot each, no second tries and no hints.
• Third, I score each result from 0 to 10 on camera, based on whether it ran, how close it got to the brief and how good it looks.

Inside an Agent OS the first step is quick, because I keep one profile per model.

A new model gets a new profile, and nothing else in the setup has to change.

Your version uses the "test prompts" file from the quick start.

You run every prompt once, score each result from 0 to 10 and write one line about why.

The prompts must stay the same every time, because that is what makes two tools comparable.

How to avoid hype

I use three checks, and together they take about ten minutes.

The first check is the repo and the README.

A real tool has install steps you can follow and updates you can see.

The second check is the date.

Anthropic's engineering post, Building effective agents, was published on 19 December 2024.

Anthropic has since added a note at the top saying much of the tooling it describes has changed.

The ideas are still worth reading, and the note is a reminder to check the date on everything.

The third check is your own test.

On my leaderboard, Claude Opus 5 averages 8.27 and the newer Claude Opus 5.5 averages 7.57 on the same one-shot prompts.

Opus 5.5 was scored in a separate, stricter run than Opus 5, so I treat that gap as a reason to test, not as proof.

That is my prompt set and not yours, which is exactly why you run your own.

How to turn learning into a build

My rule is that one thing learned becomes one thing built within 48 hours.

I teach the same order every time.

• You start with one agent doing one job.
• You keep one profile per model, so you can swap the brain without breaking the job.
• You use free models first and pay only when a job needs it.
• You connect the jobs into an Agent OS once two or three of them run well.

The "learning builds" column on your board is where this happens.

Monday's new feature becomes Wednesday's card.

Friday's test result becomes a row in your score sheet.

After a month you can open the board and see what your learning produced.

Papers for the curious

You can skip papers and still get very good with agents.

If you enjoy knowing why things work, these three are free on arXiv.

You read the abstract and the introduction first, and you stop if it won't change a build.

Where to go next

If you want the path from the start, read my how to learn AI agents guide.

If you want structured lessons, my best AI agent courses online list and my best AI agent course for 2026 compare the options.

If you want to know who is behind this page, my Julian Goldie page covers my background.

FAQ

How do I learn more about AI agents after the basics?

Follow the release notes of the one agent you use most, watch one build video a day, join one community, test new tools with the same fixed prompts, and turn each thing you learn into a small build within 48 hours.

How does an Agent OS help me keep learning?

It gives the learning a home. A scheduled agent collects updates, a shared notes vault stores what you learn, one profile per model makes testing quick, and a task board turns each lesson into a build.

Is thirty minutes a day enough to stay current with AI agents?

Yes, if each day has one job. A simple week is changelogs on Monday, a video on Tuesday, a build on Wednesday, docs on Thursday, a test on Friday, community on Saturday and a review on Sunday.

How does Julian Goldie test a new AI tool on launch day?

He wires the new model into his Agent OS, runs his fixed prompt set through it with one shot each, and scores each result from 0 to 10 on camera. The routine is described on goldiebench.com/julian.

How do I avoid AI agent hype?

Check that the repo and README exist and are active, check the date on what you are reading, and run the tool on your own fixed prompts before you repeat a claim about it.

Do I need to read research papers?

No. Papers are optional. ReAct, Toolformer and Generative Agents are three well-known ones if you are curious about the ideas behind modern agents.

Final word on how to learn more about AI agents

Set the loop up once, give it thirty minutes a day for a month, and your own board will show you how to learn more about AI agents.

Related guides on agentos.guide

→ How To Build Your Own AI Agent OS

→ Hermes AI Agent: The Full Crash Course

→ Hermes Oracle AI: The Radar

→ Best Skool Courses

📺 Video notes + links to the tools 👉

🎥 Learn how I make these videos 👉

🆓 Get a FREE AI Course + Community + 1,000 AI Agents 👉

← all guides on agentos.guide