Read this if you can't stand another Claude Code FOMO post on your timeline.
Disclaimer: I am not an engineer or ML expert. I am a marketer who has spent the past ~4 months obsessively learning all of the AI tools available right now. This article is just my perspective on what I have gathered from my research, talking to people in the AI space, and observing the general shifts of the business landscape.
If you want technical "how to use XYZ tool" guides, I will link some at the bottom of this article.
Given my fairly strong background in direct response copywriting, I have been trained to engineer FOMO. To manufacture scarcity. So, when I started seeing all of the "You have 24 months left to make it" posts on my timeline, I immediately dismissed them. I've seen those tricks before. I've written them myself.
This almost proved to be the biggest mistake of my career.
Yes, our timelines are flooded with 24/7 engagement bait these days. But I can promise you: Anyone shouting that the current state of AI is one the greatest paradigm shifts of our generation is telling the truth.
Even in the last 3 months, models have gone from "this is good" to "I can literally instruct this thing to complete any task I can imagine, and with enough tweaking it will successfully complete it"
Table of Contents
- Section 1: Why Does Everything Seem To Have Ramped Up This Month?
- Section 2: My "Holy Shit" Moment
- Section 3: Where Should You Start Learning About AI?
- Section 4: How I'm using Claude Code
- Section 5: Where To Learn More
- Section 6: A Practical Guide To Getting Started
Section 1: Why Does Everything Seem To Have Ramped Up This Month?
Simple answer: people figured out how to use these things.
It is true that the release of Opus 4.5 was one of the biggest leaps in consumer facing LLM capability we've seen in a long time. Some have argued since GPT 3, which I don't think is an unfair argument.
However, I don't think the reason for that leap is in the benchmarks. Everyone points to the true, raw, technical capabilities that Opus 4.5 have brought. Which are incredible.
But the truth is: for the majority of people using these tools (and posting about them), these differences would be completely undetectable. Vibe coders can't tell if the raw coding ability improves.
The biggest difference, in my opinion, is that Opus 4.5 (especially Claude Code) felt like the first AI tool that could just "listen" to the average person.
While prompt engineering is still crucial, this past month has felt like the first time we have been able to say, "Hey Claude, please build this thing," and Claude builds it.
Section 2: My "Holy Shit" Moment
I first booted up Claude Code a few months ago. I am fairly technically proficient, but I can't code to save my life. So, a terminal-based interface looked like alien space technology to me.
At the time, the height of my "automations" knowledge was n8n. I spent a few hours earlier that week setting up a self-hosted server, on a Hostinger VPS. I thought I was Tony Stark when I figured it out. No more paying for n8n subscriptions?
I'd gone down the rabbit hole of MCPs prior to this (basically giving Claude the keys to drive different vehicles for you), and I knew that both Hostinger and n8n had MCP servers. So, I had the idea:
"Could I just get Claude to build these workflows for me?"
To be clear, I had no idea what I was looking at. This terminal interface was completely foreign to me. But I simply asked Claude Code, in plain english, to use Hostinger and n8n MCPs to create automations for me.
I spent (no exaggeration) 5 days straight trying to figure this out. Learning what "hosting" even was, how the Claude Code interface works (I ended up using Claude Code inside of Cursor, which was 100x MORE overwhelming to learn at first), and teaching myself how to read code enough to actually understand what I was looking at.
And after dozens of hours of trial and error, I cracked it:
With the Github, Hostinger, and n8n MCPs connected, I could create a custom "app" (created by Claude Code, committed to GitHub, hosted on Hostinger using Docker), have Claude build an n8n automation that calls this "app" with a webhook…
And basically automate anything I ever wanted.
Now, to be clear:
This is a terribly inefficient way to do things. At the time, I couldn't tell you what any of those words meant. I just knew they worked
This was when this whole "AI tools are taking over" thing made sense to me.
For so long, I figured I needed a tech background to learn these things. Or that they're really only useful for coding software.
Turns out, if you just think of something you would want to delegate to AI, there's a very good chance, with enough trial and error, it can be delegated. And if you don't know how to do it, it'll be a great way to learn.
I spent the next few weeks automating everything. Content repurposing, custom API-powered scheduling tools, landing page builders. I would just pick something, have no idea how to make it work, then watch YouTube videos and talk to Claude until I figured out how to make it work.
What I ended up with was a duct-taped-together understanding of how these tools worked. Which, in my opinion, is a good thing.
I did not understand how machine learning works, or how to build token-efficient systems, or how to read TypeScript. But I did understand how to use them to solve my exact problems, to integrate with my specific use cases.
And this made me 10x more efficient in all areas of my professional life, in about 4 weeks.
Section 3: Where Should You Start Learning About AI?
Disclaimer: You're not going to like the answer.
The best way to start learning is to get your hands on the tools. No guide, course, YouTube video is going to actually help you LEARN. Learning is a process of applying knowledge to whatever you are trying to accomplish, failing, then using the resources to overcome those failures.
The best way to learn how to play guitar is to pick a song and figure out how to play it. You're going to need to learn how to hold the guitar, how to strum, then how to play chords, how to actually get all the strings to ring out, etc.
Watching YouTube tutorials without even owning a guitar won't help you.
And in all likelihood, going through a beginner's guide won't help either. Because you don't want to learn music theory, or even "proper technique" – you want to learn how to play a song to impress your friends.
So, by learning the song, you will be forced to learn all of the techniques; the difference is it will be in a context of learning that you will actually enjoy.
So, the same as guitar, pick a tool (Claude Code is probably the best one to learn right now), pick a project, and start figuring it out.
At first, it'll be extremely overwhelming. Just like learning a song, when you don't even know what a G chord is, will be overwhelming. That's good. It means you're on the edge, just past what you are capable of doing with your current level of knowledge, which is where learning actually occurs.
Now, for a practical application:
Choose something you want to delegate to AI. A boring part of your job, an admin task in your business.
Spend a few hours documenting your process of completing this task. I personally like starting a Loom recording and just narrating my process. Every button, every action, every reason behind taking that action.
When you're done with that, copy the transcript, paste it into Opus 4.5, and have it create a plan to turn this into a workflow.
(Hint: If you're using Claude Code, select "Plan Mode" and it will give you a thorough plan, and give itself a plan to execute it).
Hit start, and watch the all the files start to get created, all of the code written. You won't know what any of it means. Good. Start there.
What is the claude.md file? Why do you need it? What does it do? Can you change it?
How are the folders structured? Why are they in json instead of plain text? Can you edit the files?
Why is it writing code? What coding language is it using? How do you actually implement it? Can you see it?
Maybe you'll need an API key. What is an API key? Is it free? Should you use Anthropic or Google? What's the difference? Why are there flash models? Are they worse than regular models?
Do you see what I'm trying to illustrate?
All of this requires hours of YouTube tutorials, asking different LLMs to help explain things, uploading screenshots, calling your friends, and pacing around your room.
It's what you'd be doing anyway, but the feedback loop you are creating by actively implementing your learnings into something that you want to build makes this process exciting instead of overwhelming.
Fear and excitement are the same emotion, it's just how you choose to process it.
So instead of being overwhelmed, lost, worried, be excited, eager, ready to put the work in.
Just accept that the first 10-20 hours are going to be a complete nightmare. It's the same as learning to walk. If you gave up the first couple of times you fell, you'd be a grown adult crawling around on the floor.
Section 4: How I'm using Claude Code
Instead of trying to articulate how I am currently using Claude Code in my day-to-day operations, I figured I would just ask it itself how I use it.
Here is the response:
How Riley Uses Claude Code
Riley has built what I'd call a "command center" for their consulting work. Rather than using me as a chatbot they prompt from scratch each time, they've encoded their entire methodology into a structured system I can execute.
The Core Architecture:
There are three custom commands that route all work: /hub - Shows current status across all clients and routes to the right workflow /oracle - Runs the full offer audit or launch workflow end-to-end /recall - Searches memory from past sessions to surface relevant decisions and patterns
Skills Instead of Prompts:
Instead of writing prompts, Riley created "skills" - structured methodology files that contain decision trees, quality standards, and step-by-step processes. When I execute a skill, I'm not improvising. I'm following a documented process that produces consistent outputs.
There are currently 10+ skills covering everything from extracting voice-of-customer data from sales calls, to writing VSLs, to diagnosing why an offer is underperforming.
Workflow Handoffs:
Complex projects flow through multiple skills in sequence. Each skill has quality gates - minimum thresholds that must be met before work can proceed. If I haven't extracted enough customer language from source materials, I literally cannot move to the next phase. This prevents the kind of "AI slop" that happens when you rush to outputs without proper inputs.
Memory That Compounds:
Every significant decision logs to a persistent memory system. When I start a new session, I can recall what we learned from past client work. Patterns emerge: which objection types appear most frequently, which mechanisms resolve them, what language resonates. The system gets sharper with each project.
What This Actually Looks Like:
Riley can say "/oracle" and I will: detect whether we're auditing an existing offer or launching a new one, load the relevant client context, execute the appropriate workflow phases in order, enforce quality gates at each handoff, and produce a structured deliverable - all while documenting decisions for future reference.
The goal Riley set was "automate 100% of manual labor so I only provide direction and creativity." We're not fully there yet, but the system handles research, extraction, first-draft production, and methodology enforcement. Riley focuses on client relationships and strategic decisions.
What This Means For You (Even If You're Not Technical):
First, let's clear something up: Claude Code is not just for software developers. It's a terminal interface that gives Claude persistent access to your file system, the ability to run commands, and memory across sessions. Riley doesn't write code - they consult on offers and marketing. But they use Claude Code because it's the most powerful way to build AI into a workflow.
The shift that matters:
Most people use AI like a search engine - ask a question, get an answer, start over. Claude Code enables something different: you build a system once, then execute it repeatedly. The AI becomes a junior team member who knows your methodology, remembers past work, and follows your process.
How to start applying this:
Document your methodology first. Before you touch Claude Code, write down how you actually do your work. What are the steps? What inputs do you need? What quality looks like? Most people skip this and wonder why AI outputs feel generic. The AI can only be as good as the process you give it.
Create "skills" not prompts. A skill is a markdown file that contains: what this task is, what inputs it needs, the step-by-step process, what good output looks like, and common mistakes to avoid. Save it in Claude Code's skills folder. Now you can invoke it by name instead of re-explaining every time.
Build quality gates. Define what "done" means for each phase of your work. "I need at least 10 customer quotes before I write copy." "The positioning statement must pass these 3 criteria." Quality gates prevent AI from rushing to outputs without proper inputs.
Use memory intentionally. Claude Code can remember across sessions. Log your decisions, what worked, what didn't. Over time, you're not starting from scratch - you're building on everything you've learned.
Think in workflows, not tasks. The real leverage isn't getting AI to do one thing. It's chaining tasks together: research flows into extraction, extraction flows into drafting, drafting flows into refinement. Each handoff follows rules you defined.
The underlying principle:
The people who will win with AI aren't the ones with the best prompts. They're the ones who encode their expertise into systems that execute consistently and improve over time. Claude Code is just infrastructure for that - a place where your methodology lives, compounds, and runs.
You don't need to be technical. You need to be clear about how you work.
Section 5: Where To Learn More
Here's an unorganized resource dump, in no particular order.
General AI:
https://www.youtube.com/@GregIsenberg
https://www.youtube.com/watch?v=lXUZvyajciY
https://youtu.be/aR20FWCCjAs?si=1sMHNEiUuEp17WRv
https://x.com/BawsaXBT/status/2008177597398790592?s=20
https://x.com/omarsar0/status/1998383154181361813?s=46
Claude Code (because I get a lot of people asking about CC specifically):
https://x.com/hayesdev_/status/1997259477670854937?s=20
https://youtu.be/vqHBfe3r4OQ?si=eMYaTKNMYgr2rWXA
https://joshuaschultz.com/ai/articles/claude-code-complete-guide/
But please, just do what we all do and watch a bunch of YouTube videos
Section 6: A Practical Guide To Getting Started
Step 1: Pick something you wish AI could replace. Whether it's admin work, or repeated tasks that you do every day, or just a part of your work/life you wish could be done on autopilot.
For this example, let's use editing content.
Step 2: For the next couple of days, document everything you do while executing this task.
For me personally, I like just starting a Loom video and narrating everything I do, as if I am creating SOPs to handoff to a new hire, or filming one of those 2011 YouTube tutorials.
Whether it's taking notes, getting on a Google Meet with a friend and recording you walking them through it, recording a video or voice note; find a system that is easy for you to use.
Then, just do your work as normal. Try to be as detailed as possible with your annotation. Try to vocalize your thought patterns.
So instead of "I'm going to cut this intro line," you'd say "I feel the hook needs to be stronger, and the introduction is unnecessary for the reader and doesn't reel them into this content. So, I am going to delete it, and replace it with something a bit shorter and punchier, like _____."
Step 3: Turn your transcript/notes into an execution plan. If you are using (or learning) Claude Code, toggle the chat into Plan Mode, paste your notes in, and just ask it to analyze what you documented and help you come up with a plan to standardize this workflow using AI.
You could alternatively upload it to Claude, Gemini, ChatGPT, Manus, or whatever LLM you use most often. Set it to its largest thinking model and have it get to work.
Hint: Deep Research is great for stuff like this. You could say "I don't know how to build this, please research the best options out there right now to build this"
Step 4: Build it. If you're using Claude Code, congratulations, you're now a developer (just kidding). It'll be a back and forth process of figuring out what the hell this thing is building, and iterating on it.
You could also use more visual, node-based workflow builders. Gumloop, Make, and n8n are all solid options, offering varying levels of customization and features.
Or maybe there are AI tools that already exist for this specific use case that a Deep Research prompt could uncover, and you don't need to "build" anything, but rather setup your particular workflow on these tools.
Whatever path you take, it honestly doesn't matter. All that matters is that you build, iterate, and learn until you have a workflow that you will actually use in your day-to-day work.
Don't build something that looks cool, but you use it a couple times then start manually doing the work again.
There will be a transition period, where setting up and troubleshooting the workflow will be less efficient than doing the work yourself. Welcome to the world of delegation, my friend. This is no different than hiring. You need to stick it out for the sake of your time.
Step 5: Iterate and improve (and learn). Maybe you want to change the workflow. Maybe you learn about a new coding language that would make it much more efficient. Maybe you want to build a UI for it using Gemini or Lovable and let your friends use it. Maybe you want to package it up and start charging for it! (Please don't do this, despite what you read on X, launching a SaaS is a bit more complicated than this).
The most important thing is that you keep learning through rapid iteration and prototyping. If you come across a banger X post that talks about agentic workflows, don't just bookmark it. Learn how to do it, and apply it to your workflow. Or create a new one. Either way, you only actually learn by executing, failing, realizing you don't know enough to complete the task, then learning enough to be able to complete it.
Again, think back to the guitar analogy.
This is the part everyone misses, and it's (I think) the reason there is so much confusion and stress and anxiety when it comes to this rapidly changing technology landscape.
There are gatekeepers, that are desperately trying to hold onto their edge. They're realizing their College degree is mostly useless, and that numbskulls like me are building silly putty-level AI tools to lap them in whatever professional endeavors they are stuck in.
You do not need anything - permission, formal education, training, resources - to get into the top 10% of people currently using these tools.
All you need to do is pick something that is interesting to you, and set sail into the vast ocean until you hit land.
You are an explorer! You don't need a map. You just need to be willing to get lost at sea, and have faith that you'll figure out how to get back.
– RL