Keeping Up With AI Isn’t the Same as Learning It

Introduction

We opened this episode with a question for Charlie Hills: he publishes a new AI system every single week, so is he part of the reason why people feel left behind? His answer was the opposite. He sees himself as a lighthouse, guiding people through how to actually use AI in their day-to-day work.

In this episode, we — Gerjon Kunst and James O'Regan — talk with Charlie about AI FOMO, why you don't need to chase every release, how he runs his content with Claude Code, and what he thinks the internet will look like next year. If you've ever felt that keeping up with AI is a full-time job, this one is for you.

Meet the Guest

Charlie Hills is an AI creator and the founder of MarTech AI, a newsletter with one practical AI marketing system every week, and of Charlie Hills' AI OS, a six-week LinkedIn AI cohort built on the system he created in Claude Code. He started out as a B2B marketing manager and was the self-appointed AI champion at his company, sharing new workflows and tools with his boss every day, until he was told there was no time for it. That frustration, and a long-standing wish to escape the nine-to-five, pushed him to start posting about AI on LinkedIn at the end of 2023.

On Christmas Day 2023 he sat down to work out his content strategy, his content pillars and his profile, and committed to posting every single day in 2024. After nine months of doing that next to his job, he quit and became a full-time creator a little over two years ago. His LinkedIn following grew from roughly 30,000 to 100,000 in four or five months once he could focus on it full time. His headline says it all: he helps you actually use AI.

Setting the Stage

Every week there is a new release from Anthropic or OpenAI, a new GitHub repo, a new agent that everybody seems to be talking about. For many people that creates the feeling that they are already behind. Charlie feels it too, and he talks about AI every day.

This blogpost walks you through what he thinks you can safely ignore, what you really do need to learn, and the simple setup he recommends for getting the most out of AI, starting this weekend.

Episode Highlights

Pick one model and master it

The migration from ChatGPT to Claude and back again has been entertaining, but for the average knowledge worker it isn't necessary. Charlie's advice is to pick one LLM you like, whether that's Claude, Gemini or Copilot, and build the harness around it: the skills, the workflows and the MD files. The cost of switching is simply too large, and the return on obsessing over every minute update is small.

“All you need to do now is pick that one LLM you love, and then really build the harness around that.”Charlie Hills

You're never too late, but you do have to start

Are people genuinely late to AI? Charlie says that in the grand scheme we are still early, but the fundamentals of prompting, systems and connecting your tools are the stage that is passing. The good news is that it takes maybe a hundred hours of doing the thing to catch up. As Gerjon pointed out, the gap only gets bigger for everyone who doesn't start.

Deep Dive: Build a Loop That Critiques Its Own Work

The thing Charlie calls the biggest shift of his year is the loop. He creates the infographics on his LinkedIn feed programmatically with Claude Code. When he started, he went through over a hundred variations before getting a good result, correcting padding and columns one detail at a time, which he jokingly calls vibe designing.

What changed was a system:

• **Examples.** Gather a handful of examples of what good looks like. Your own data is the most valuable thing you have, especially now that AI can generate anything.
• **A rubric.** Ask the AI to analyse those examples and turn them into specific criteria, with different weights for different elements.
• **A skill.** Save the scoring criteria as its own skill, so it is invoked automatically, for example when he asks for an infographic.
• **A loop.** Let the AI generate an output, score itself against the rubric and keep iterating until it passes a threshold. Charlie sets it at ninety-five out of a hundred, and only then does it show him the result.
• **Memory.** Every correction is saved to a memory log, so the exact brand style and past feedback are remembered next time.

The result: nine times out of ten it gets it right, and he uses the Figma connector to keep the output quality high. He uses the same approach for LinkedIn captions with a QA loop, which turns a bland first draft into something he only needs to change by about five percent. As Charlie says, this applies to anything, including emails.

Real-Life Stories & Examples

From marketing manager to full-time creator

Charlie shared how scary the leap was and how he timed it: he had at least one paying client, brand partners coming in and about 30,000 followers. The hardest part was the months of working forty hours a week for his employer and creating content in the mornings and at weekends, on the verge of burnout more than once.

The LinkedIn algorithm has changed

Two years ago, even an average post would reach twenty or thirty thousand impressions. Today he puts much more effort in and never knows how a post will perform. Charlie also points out that AI is a hot topic, which helps; finance would be a lot harder.

Why he doesn't touch every new tool

Charlie hadn't even installed some of the tools that the AI world was hyping at the time. His test is simple: if he hasn't tried it and found a practical use for his marketing audience, he won't share it. He also warns that some personal-agent apps store your data, including personal details and credit cards, on servers you know nothing about. We added our own rule: if we ever test one, it's with a separate virtual card with a set limit.

Letting Claude test the new features

When Anthropic released a new feature on a Thursday, Charlie had Claude test it for him on the Saturday. The first outputs were average, so he gave more input and asked for Apple-style motion graphics instead, which came out amazing. That became his newsletter and a post that performed very well. It's a good example of using Claude in Chrome, or computer use, to test at scale without clicking around every button.

Skip the Obsidian knowledge graph hype

Charlie fell for the second-brain hype himself, wrote a newsletter about it, and a few months later asked what the graph was actually doing for him. His takeaway: it's valuable to have all your MD files and repos connected, so Claude doesn't guess. He created a map.md file that connects all his local files, so the AI knows how things relate. He is also building a second brain for LinkedIn that can be plugged into other people's Claudes.

GitHub for marketing teams

GitHub confused him for a very long time. Today he has a team OS on GitHub that gives everyone who works with him access to how he does things across all his social media. That way, someone can create a graphic or edit a video in his way without being an expert, and the knowledge isn't lost when someone leaves the team.

Key Takeaways

• Most people don't need to chase every release. Pick one LLM and master it.
• Wait a few months before you adopt a flashy new tool or agent, and read what it does with your data.
• You're not too late, but you do need to start. Around a hundred hours of practice goes a long way.
• Voice-to-text and a bit of iteration beat fancy prompt frameworks.
• Collect examples of your best work, build a rubric, turn it into a skill and let the AI critique itself in a loop.
• Connect your files, with something like a map.md, so the AI stops guessing.
• A shared repo, for example on GitHub, keeps team knowledge from getting lost.
• Treat AI content like a sandwich: the idea and the final taste check are yours, and AI does the middle.
• Never hand over your real credit card to a new agent; use a virtual card with a limit.

Closing Thoughts

Charlie left us with a prediction: after the year of the agentic web, he expects the internet to feel very different. With more AI agents and AI-generated content than ever, he thinks real human connection will move to curated communities and in-person events. We thought that was both scary and a good reason to keep doing what we do.

Our favourite takeaway might be his simplest one: make a folder in Claude Code, put in your best work and build your first loop this weekend. A big thanks to Charlie for sharing so openly. What did you think? Are you feeling left behind, or are you building your first loop? Join the conversation with us, and tune in to our next episode.