How to learn building AI agents has a clear verdict from me, which is that you should build seven practice projects on a no-code harness first and move to an agents SDK second.
The seven projects run from a scheduled daily recap up to your own agent OS dashboard, and each one teaches a single concept.
This page is the reference version, so every build gets the same fields, a score and a short verdict.
The scores are my own opinion from building and publishing each project, and they aren't a measured benchmark.
You'll also get the two builder tracks compared side by side and a debugging checklist.
The Verdict In One Table
| Build | Concept taught | Ease | Teaching value | Lasting use | My score |
|---|---|---|---|---|---|
| 1. Scheduled daily recap | The agent loop on a timer | 10 | 7 | 9 | 8.7 |
| 2. Inbox triage agent | Tools and permissions | 7 | 9 | 9 | 8.3 |
| 3. Research and news radar | Live data and structured output | 7 | 8 | 8 | 7.7 |
| 4. Second brain with Obsidian | Memory and context | 8 | 10 | 10 | 9.3 |
| 5. Skill from a book or SOP | Reusable skills | 9 | 8 | 8 | 8.3 |
| 6. Kanban build line | Orchestrating several agents | 5 | 9 | 8 | 7.3 |
| 7. Agent OS dashboard | System design | 4 | 10 | 10 | 8.0 |
Each score is the average of the three columns before it, rounded to one decimal place.
The order of the ladder follows difficulty and not the score, because you can't skip to the best build without the earlier ones.
How I Scored These Builds
I scored every build out of 10 on three things.
Ease is how quickly a beginner can get it working.
Teaching value is how much the concept carries over to other agents.
Lasting use is whether the build keeps earning its place after the lesson is over.
I've built and published all seven, and the receipts sit in my guides on agentos.guide.
What You Need Before You Start
All seven entries below use Hermes Agent, which is a free, open-source agent from Nous Research.
You install it with the official script and connect a model with hermes setup.
curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash
You should also run hermes doctor to check the install and hermes update so the newer commands exist.
A free or local model is enough for the early builds.
Build 1: Scheduled Daily Recap — 8.7/10
Concept: The build teaches the core loop, where a prompt runs, does some work and delivers a result.
Tools: You need Hermes, one model and a delivery target such as Telegram.
Command: You create the job with hermes cron create, a schedule and a prompt.
hermes cron create "0 6 * * *" "Read my task list and send me a short to-do recap for today" --name "Daily 6AM To-Do Recap" --deliver telegram
Definition of done: The recap arrives three mornings in a row with nobody at the keyboard.
What usually breaks: The prompt assumes context the job doesn't have, or the delivery target was never connected.
Verdict: This is the easiest build on the ladder, and it's the right first one because the feedback arrives every day.
🔥 Want my exact How To Learn Building AI Agents setup, step by step? Inside the AI Profit Boardroom you get step-by-step video trainings, my installable Agent OS and four coaching calls a week with 3,400+ members building real automations. → Join the AI Profit Boardroom
Build 2: Inbox Triage Agent — 8.3/10
Concept: The build teaches tools and permissions, which decide what an agent can touch.
Tools: Hermes has a built-in email-inbox-triage skill, and it names himalaya and google-workspace as related mail skills.
Command: You attach the skill to a morning job with the --skill flag on hermes cron create.
Definition of done: The agent's five priority threads match yours for five mornings, and it has sent nothing without you.
What usually breaks: The mailbox isn't connected, the rules for "urgent" are vague, or the agent was allowed to send too early.
Verdict: It scores high on teaching value because it's the first build where a mistake has a real cost.
Build 3: Research And News Radar — 7.7/10
Concept: The build teaches live data, structured output and running an agent from a script.
Tools: You use hermes -z for a one-shot run and hermes send to push the result to a phone.
Command: My own radar pins the model, because only xAI models carry the live X search tool.
hermes -z "Return the six hottest stories in my niche today with a headline, a heat score out of 100, a link and a content angle" --provider xai-oauth --model grok-4
Receipts: My Hermes Oracle AI guide shows 49 mornings logged, 294 trends scored and 46 pages published from those trends.
Definition of done: Six scored signals with working links arrive every morning for a week.
What usually breaks: Live search occasionally returns an empty sweep, and a single retry on empty is the fix I use.
Verdict: It loses a point on ease because it depends on one provider, but it's the build that makes structured output click.
Build 4: Second Brain With Obsidian — 9.3/10
Concept: The build teaches memory, which is how an agent gets your real context.
Tools: You use Obsidian, which is a free download, and the built-in obsidian skill in Hermes.
Method: You make a vault, give Hermes the full folder path, ask for the folders and write one short "about" page.
Receipts: My vault held 614 notes when I measured it for my Hermes second brain guide, and the demo answer listed the 11 notes it used.
Definition of done: A fresh chat answers a question about your business correctly and names the notes it read.
What usually breaks: I skipped the folder path while filming, and Hermes built the folders in its own workspace instead.
Verdict: This is my top-scoring build, because every other agent gets better the moment it exists.
🔥 Want my vault structure, automations and board configs? Inside the AI Profit Boardroom, I share them as step-by-step trainings, alongside my installable Agent OS. You also get four coaching calls a week and 3,400+ members running the same builds. → Get access here
Build 5: Skill From A Book Or SOP — 8.3/10
Concept: The build teaches skills, which are knowledge and procedures an agent loads on demand.
Tools: You use the /learn command inside a Hermes chat.
/learn ~/Documents/your-book-or-sop.pdf
Method: Hermes reads the file once and writes an index, a reference file per chapter, a glossary and a cheat sheet.
Receipts: I ran it on The Art of War in my /learn guide, and the engine's own testing reports 24 to 51 times fewer tokens than pasting a book into the chat.
Definition of done: A new session answers three questions you know the answers to, with no upload.
What usually breaks: Book mode only ships in newer builds, so an old install has the command but not the full layout.
Verdict: It's nearly as easy as the recap, and it's the build I'd give to a team with documented processes.
Build 6: Kanban Build Line — 7.3/10
Concept: The build teaches orchestration, where tickets are handed to separate agents.
Tools: You use hermes kanban, and each worker is a Hermes profile.
hermes kanban boards create my-builds
hermes kanban create "Build a landing page for my newsletter" --assignee builder
Receipts: In my Hermes Command Desk guide three tickets reached done without me, on free models, for $0.00.
Definition of done: Three tickets are done, and you have opened and checked each result yourself.
What usually breaks: My first three dispatches died instantly, and my kanban build line guide documents the three config traps behind that.
Verdict: It has the lowest score because setup is fiddly, but the lesson is the one that separates one agent from a team.
🚀 Ready to put How To Learn Building AI Agents to work in your business? Inside the AI Profit Boardroom you get step-by-step video trainings, my installable Agent OS and four coaching calls a week with 3,400+ members building real automations. → Join the AI Profit Boardroom
Build 7: Agent OS Dashboard — 8.0/10
Concept: The build teaches system design, which is how memory, agents and tasks share one home.
Tools: You use Claude Code as the builder, Obsidian for memory, Hermes for long jobs and Tailscale for phone access.
Method: You describe the dashboard in plain English, review what Claude built and refine it.
Receipts: My free Agent OS guide lays out five steps with the exact prompts, and my own dashboard holds Claude Code, Hermes and several other agents as pages in one app.
Definition of done: One page shows your agents, your memory and your board, and it opens on your phone.
What usually breaks: People install every agent at once, skip the vault or automate a task they've never done by hand.
Verdict: It's the hardest build and the most valuable one, so it belongs at the top of the ladder.
The Two Builder Tracks Compared
| Field | No-code harness route | Code route with an agents SDK |
|---|---|---|
| What you use | A finished agent such as Hermes Agent | A framework inside your own program |
| What you write | Commands, prompts and skills | Python or TypeScript |
| Time to a first working agent | One evening | Longer, because you assemble the loop yourself |
| Control | You work within the harness | You control every step |
| Best for | Operators, owners, agencies and creators | Developers shipping agents in a product |
| My verdict | Start here | Move here second |
A harness gives you the loop, the tools, the memory and the scheduler on day one.
An SDK gives you the parts and leaves the assembly to you.
These are the code-route frameworks I confirmed have live documentation today.
| Framework | Who makes it | Where to read about it |
|---|---|---|
| OpenAI Agents SDK | OpenAI | The OpenAI Agents SDK docs |
| Claude Agent SDK | Anthropic | The Agent SDK overview |
| LangGraph | LangChain | The LangGraph overview |
| Agent Development Kit | The ADK docs | |
| smolagents | Hugging Face | The smolagents docs |
| CrewAI | CrewAI | The CrewAI docs |
I haven't scored those frameworks, because I teach the harness route and I won't rate tools I haven't published builds with.
My honest advice is that the concept matters more than the framework.
A scheduled loop, a tool with permissions, a memory, a skill and a ticket all exist in every one of them under different names.
How To Debug An AI Agent: The Checklist
| Symptom | What to run | What it tells you |
|---|---|---|
| Nothing works at all | hermes doctor | Whether the configuration and dependencies are healthy |
| A scheduled job never arrived | hermes cron status and hermes cron runs | Whether the scheduler is running and what each attempt did |
| A job failed with an error | hermes logs --level ERROR --since 1h | The actual error from the last hour |
| A kanban ticket is stuck | hermes kanban show and hermes kanban diagnostics | The ticket's events and any active problems on the board |
| The bill looks too high | hermes insights | Token use, cost and tool patterns |
| An agent is doing something you don't like | hermes pause | It stops cron and kanban dispatch until hermes resume |
After the checklist, you change one thing and run the job by hand.
Most failures I've hit were a vague prompt, a missing connection or the wrong model for the tool.
Common Mistakes When You Learn Building AI Agents
- Starting with the kanban build line is a mistake, because orchestration only makes sense once a single agent works.
- Studying frameworks for weeks is a mistake, because one small agent that runs teaches more than ten tutorials.
- Skipping memory is a mistake, because an agent without context gives generic answers.
- Giving write access too early is a mistake, because a drafting agent can't embarrass you.
- Paying for the biggest model on day one is a mistake, because the early builds run fine on free ones.
💡 Want a second pair of eyes on your first builds? The AI Profit Boardroom has courses on Hermes Agent and coaching calls where you can share your screen and get unstuck. → Join the AI Profit Boardroom
A Weekly Plan That Follows The Scores
In week one, build the daily recap, because it's the easiest.
In week two, build the inbox triage agent on drafts only.
In week three, build the radar.
In week four, build the second brain, which is the highest-scoring project here.
In week five, turn one SOP into a skill.
In week six, ship three tickets on a kanban board.
In week seven, start the dashboard.
Related Reading
- The best AI agent course in 2026 is my ranked list of courses to pair with these builds.
- How to learn AI agents is the broader path for a complete beginner.
- How to learn creating AI agents is the sibling guide in this series.
- The best Hermes Agent use cases, ranked and scored gives you more projects after the ladder.
Also On Our Network
- 🌐 Every command for the seven builds, step by step
- 🌐 What each agent build takes off a business owner's desk
- 🌐 The seven builds as rooms in one Agent OS
- 🌐 Model picks for each build, with real bench scores
FAQ: How To Learn Building AI Agents
Is it better to learn building AI agents with no-code tools or with code?
I'd start with a no-code harness such as Hermes Agent, because the loop, the tools and the memory already exist.
I'd move to an agents SDK when you want agents inside your own product.
Which practice project teaches the most about AI agents?
In my scoring, the second brain with Obsidian teaches the most.
Memory and context affect every other agent you build.
Which agents SDKs exist for the code route?
The frameworks with live documentation today include the OpenAI Agents SDK, the Claude Agent SDK, LangGraph, Google's Agent Development Kit, smolagents and CrewAI.
I link each one's own docs in the table above.
Are the scores on this page a benchmark?
No, the scores aren't a benchmark.
They're my own first-person scores from projects I've built and published.
What is the most common mistake when learning to build AI agents?
The most common mistake is starting with the hardest project before a single scheduled agent works.
The second is studying frameworks for weeks without shipping one small agent.
📺 Video notes + links to the tools 👉
🎥 Learn how I make these videos 👉
🆓 Get a FREE AI Course + Community + 1,000 AI Agents 👉
About Julian
I'm Julian Goldie, an SEO entrepreneur, author and founder of the AI Profit Boardroom, which has 3,400+ members.
I help business owners scale with AI agents, automation and SEO.
- I founded Goldie Agency, a 7-figure SEO and link-building agency based in Bangkok.
- I've grown a YouTube channel to 400,000+ subscribers.
- I wrote "Link Building Mastery", which is available on Amazon.
- I run GoldieBench and agentos.guide, where I publish my AI model tests and free Agent OS guides.
→ Get my best AI training inside the AI Profit Boardroom
Start on the harness, climb one build a week, and you'll have my full verdict on how to learn building AI agents from your own experience.
