← All Conversations
July 15, 2026

Oh No! I wasted months learning AI wrong

AI tools Claude ChatGPT Gemini productivity

Summary

We need to produce a summary 250-500 words, covering core insight, arguments, tools/techniques mentioned, actionable takeaways. No preamble, no labels. Just the summary text. We must ensure length between 250 and 500 words. Let's aim ~300-350 words. We need to capture main points: wasted months learning AI wrong, focusing on picking one model, using paid tier, matching to work, vibes, changing default model, stopping prompting, feeding context, OC principle, ways to feed context (framework name, examples, tool connections), using Projects/Gems, markdown files, note about AI IPO rush (maybe mention but not core). Provide actionable takeaways. Let's craft ~340 words. Check word count. We'll write then count. Draft: "When I began studying AI I spent nearly eight months collecting videos, prompts, and tool trials without gaining real skill, because most content is either short‑lived or purely theoretical. The breakthrough came when I asked what 20 % of AI knowledge will still be relevant a decade from now and built a three‑level roadmap around that answer. Level 1 – Pick one model and go deep. Trying to master every chatbot spreads effort thin; today’s leading models (ChatGPT, Claude, Gemini) share nearly identical core features, so learning one transfers to the others. Choose based on three principles: prioritize a paid tier (the gap between free and paid is massive), match the model to your work (ChatGPT for research, Claude for writing/design/coding, Gemini for multimodal or Google‑Workspace workflows), and consider personal vibe—enjoyment drives usage. Crucially, change the default model to the most powerful version you can access; the default is the cheapest, weakest option and limits the AI’s reasoning ability. Level 2 – Stop prompting, start feeding context. Memorizing prompt formulas yields diminishing returns; the right context outperforms any perfect prompt. Adopt the OC principle: state the desired outcome and supply the material the AI needs to infer the rest. Feed context by naming a known framework instead of describing it, providing real examples of good output, and connecting the AI to your live data sources (email, Drive, Slack, Notion) so it pulls what it needs directly. Save this setup in a Project (Claude/ChatGPT) or Gem (Gemini); a Project stores instructions, reference files, and examples, making every new conversation automatically aware of your workflow. Use markdown (.md) files for knowledge bases because they are easier for the AI to parse and cheaper to process than PDFs, and you can ask the AI to convert existing PDFs to markdown. Level 3 – Apply the system to recurring tasks, letting the AI handle routine work while you focus on higher‑level decisions. By concentrating on one model, leveraging paid features, and supplying rich, reusable context, you build durable AI skill that won’t evaporate with the next tool hype." Now count words. Let's count manually roughly. I'll copy and count. "When(1) I2 began3 studying4 AI5 I6 spent7 nearly8 nine? actually "nearly" is 8? Let's just use a tool mental? Might be easier to approximate. Need 250-500. Likely around 300-340. Let's quickly count using rough method: I'll count words per sentence. Sentence1: "When I began studying AI I spent nearly eight months collecting videos, prompts, and tool trials without gaining real skill, because most content is either short‑lived or purely theoretical." Count: When(1) I2 began3 studying4 AI5 I6 spent7 nearly8 nine? Actually "nearly" is 8, "eight"9, "months"10, "collecting"11, "videos,"12, "prompts,"13, "and"14, "tool"15, "trials"16, "without"17, "gaining"18, "real"19, "skill,"20, "because"21, "most"22, "content"23, "is"24, "either"25, "short‑lived"26, "or"27, "purely"28, "theoretical."29 Sentence2: "The breakthrough came when I asked what 20 % of AI knowledge will still be relevant a decade from now and built a three‑level roadmap around that answer." Count: The1 breakthrough2 came3 when4 I5 asked6 what7 20 %8 of9 AI10 knowledge11 will12 still13 be14 relevant15 a16 decade17 from18 now19 and20 built21 a22 three‑level23 roadmap24 around25 that26 answer27. Sentence3: "Level 1 – Pick one model and go deep." Count: Level1 1 –2 Pick3 one4 model5 and6 go7 deep8. Sentence4: "Trying to master every chatbot spreads effort thin; today’s leading models (ChatGPT, Claude, Gemini) share nearly identical core features, so learning one transfers to the others." Count: Trying1 to2 master3 every4 chatbot5 spreads6 effort7 thin;8 today’s9 leading10 models11 (ChatGPT,12 Claude,13 Gemini)14 share15 nearly16 identical17 core18 features,19 so20 learning21 one22 transfers23 to24 the25 others26. Sentence5: "Choose based on three principles: prioritize a paid tier (the gap between free and paid is massive), match the model to your work (ChatGPT for research, Claude for writing/design/coding, Gemini for multimodal or Google‑Workspace workflows), and consider personal vibe—enjoyment drives usage." Count: Choose1 based2 on3 three4 principles:5 prioritize6 a7 paid8 tier9 (the10 gap11 between12 free13 and14 paid15 is16 massive),17 match18 the19 model20 to21 your22 work23 (ChatGPT24 for25 research,26 Claude27 for28 writing/design/coding,29 Gemini30 for31 multimodal3

Hey

I have a real confession.

When I started learning AI, I wasted close to 8 months doing it completely wrong.

I bookmarked 50 YouTube videos. I saved threads about "secret prompts." I downloaded prompt PDFs I never opened. I jumped between ChatGPT, Claude, Gemini, and every shiny new tool that launched that week.

I felt busy. I felt informed.

I was neither.

Because here is what nobody told me: most AI content out there is either outdated within months or pure theory you will never actually use. I was collecting information, not building skill.

The turning point came when I stopped asking "what should I learn next" and started asking "what 20% of this will still matter a decade from now."

That question changed everything. And today, I am giving you the answer.

If I had to learn AI from scratch in 2026, this is the exact roadmap I would follow. Three levels. Each one builds on the last. No fluff, no theory, nothing that expires next month.

#### **Level 1: Pick One Model and Go Deep**

This is where I made my first big mistake. I tried to learn every AI tool at once.

Do not do that.

Here is why one is enough. The top models used to be far apart in capability. Today they are clustered so close together that the difference for the average user is negligible. And because every AI company copies every other AI company, they all share the same core features. Projects, memory, file uploads, connectors. Learn them once, and the skill carries straight over to the rest.

So which one do you pick? Realistically, you have three serious choices: ChatGPT, Claude, or Gemini.

Use these three principles to decide:

Principle 1: Prioritize paid tiers. If you are on free ChatGPT but your job gives you paid Gemini, go deep on Gemini. The gap between free and paid is night and day. Do not learn AI on the weakest version of it.

Principle 2: Match the AI to your work. ChatGPT is the most mature, has the most tutorials, and is excellent at web search and research. Claude is the strongest at writing, design, and coding, and coding matters even if you are not technical because data analysis and diagrams all run on code under the hood. Gemini is the pick if you work across text, images, audio, and video, or if you live inside Google Workspace.

Principle 3: Vibes. I know it sounds unserious. It is not. Each AI has its own personality, and the more you enjoy using yours, the more you will use it, and the better you will get. Switching later is easy anyway. All three now have memory import features.

One more thing, and this single tip is worth the whole email.

Change your default model. Every AI company defaults you to their weakest model because it is the cheapest for them to run. For any real work, manually select the most powerful model you have access to. The difference is not small. The powerful models break down your request, map out steps, and catch nuances you never thought to mention.

I used the default model for months without knowing this. Do not repeat my mistake.

#### **Level 2: Stop Prompting. Start Feeding Context.**

Notice I have not mentioned prompting frameworks yet. That is on purpose.

For my first few months, I memorized prompt formulas like exam answers. Role, task, format, tone. I treated prompting like a magic spell.

Then I realized something that made all those saved prompt PDFs useless.

The right context beats the perfect prompt. Every single time.

Here is a simple example. Say you need to find a restaurant for your boss. You could spend 10 minutes writing a detailed prompt describing everything you think your boss likes. Or you could paste a list of restaurants your boss has loved in the past and let the AI figure out the pattern.

The second approach wins every time. The list is context. Context carries information you would never think to write out.

So forget the 50 frameworks. There is only one worth remembering: **OC. Outcome plus Context.**

Tell the AI what you want. Give it the material to work from. It infers the rest, often better than you would have specified it yourself.

Three ways to feed it the right context:

1. Name a framework instead of explaining it. Saying "rewrite this using the pyramid principle" carries more context than a full paragraph describing what you want. Do not know the right framework? Ask the AI for options first, then pick one.

2. Show real examples of what good looks like. Instead of describing the format and tone for a weekly update, paste the last 2 or 3 updates that got approved, add your raw notes, and say "write this week's update in the same format." Examples contain everything you forget to say.

3. Connect your tools. Your best context already lives in your email, your Drive, your Slack, your Notion. Connect them, and the AI pulls what it needs directly. No more downloading and re-uploading files like it is 2023.

And then save all of it.

This is what **Projects** are for (Claude and ChatGPT call them Projects, Gemini calls them Gems). A Project is a permanent home for recurring work. Your instructions, your reference files, your examples, all saved once. Every new conversation inside it already knows your setup. You stop repeating yourself forever.

Quick pro tip: use markdown (.md) files in your Project knowledge instead of PDFs whenever possible. They are easier for the AI to read and cheaper to process. You can even ask the AI to convert your PDFs into markdown.

### The AI IPO Rush Is Coming

OpenAI and Anthropic could bring a new wave of AI attention to the public markets. But investors don’t have to wait for the IPOs.

MarketBeat’s [7 AI Stocks to Buy Now](http://marketbeat.com/newsletter/pdfoffer.aspx?offer=7aistocks&RegistrationCode=Beehiiv-BUBDNZO6MS-7aistocks&_bhiiv=opp_1e834854-2e18-4b4f-9f8e-a35cf57dcc25_7d4f8797&bhcl_id=30d6c58f-221d-4509-bf0b-34e541a60d70_f672f2b6-1e60-4422-8448-ee13d3aa1099_22c35bc7-5e17-4448-baa5-49c4bf23f93e) report reveals 7 publicly traded companies positioned to benefit from the next phase of AI investment.

[Send My Free Report](http://marketbeat.com/newsletter/pdfoffer.aspx?offer=7aistocks&RegistrationCode=Beehiiv-BUBDNZO6MS-7aistocks&_bhiiv=opp_1e834854-2e18-4b4f-9f8e-a35cf57dcc25_7d4f8797&bhcl_id=30d6c58f-221d-4509-bf0b-34e541a60d70_f672f2b6-1e60-4422-8448-ee13d3aa1099_22c35bc7-5e17-4448-baa5-49c4bf23f93e)

#### **Level 3: Connect Everything Into an AI System**

This is the level most people have not reached yet. Honestly, I only got here recently myself.

Here is the problem with Projects: each one is a silo. My workout Project cannot see what is inside my health checkup Project, even though it obviously should.

An AI system fixes that. It does two things Projects cannot:

First, it connects the dots across everything. When I migrated my separate health, supplements, and workout Projects into one system, the AI cross-referenced my latest checkup with my training plan, flagged that I had zero cardio days despite borderline high cholesterol, and recommended adding cardio to my rest days. No single Project could have caught that.

Second, it learns from your feedback and compounds. My favorite move is called "reconcile." The AI drafts something, I edit it heavily, then I say "reconcile my final version with your initial draft." It dissects every change I made and saves the rules for next time. The more feedback I give, the fewer instructions I need. The system literally gets smarter the longer I use it.

Your three options for building one, from easiest to most powerful:

Gemini Spark is the most beginner friendly. It is already connected to Gmail, Calendar, and Drive, so setup is minimal. The tradeoff is less control.

Claude Cowork is built specifically for non-technical people. More control than Spark, with a small amount of setup. This is the sweet spot for most readers of this newsletter.

Claude Code and OpenAI Codex are Cowork on steroids. Fully customizable and extremely powerful, but you need to be somewhat comfortable around code.

A funny tell for which one fits you: look at the model selector. Codex gives you a dozen intimidating options. Cowork gives you a few. Spark gives you none. Pick the one that matches your comfort level and grow from there.

### 100+ Claude Code hacks to ship code 10X faster

Top engineers at Anthropic say AI now writes 100% of their code.

Are you using AI to write yours?

These [100+ Claude Code hacks](https://magic.beehiiv.com/v1/5f7ce6e3-9a71-416b-99a7-606c5f7e2447?email=bob@reisner.info&redirect_to=https%3A%2F%2Fcodenewsletter.ai%2Fforms%2F14166360-de71-46c4-8722-878d417fab5c&utm_source=beehiiv&utm_campaign=BUBDNZO6MS&redirect_delay=3&_bhiiv=opp_0dc31371-131f-446b-9dba-52fc10e762c0_94e90c2e&bhcl_id=fb20a525-6361-460c-ba5e-b83d6fdfeef3_f672f2b6-1e60-4422-8448-ee13d3aa1099_22c35bc7-5e17-4448-baa5-49c4bf23f93e) show you exactly how. Sign up for The Code and get:

  • 100+ Claude Code hacks — free
  • The Code newsletter — learn the latest AI tools and skills to code faster in 5 mins a day

    • [Claim your free playbook](https://magic.beehiiv.com/v1/5f7ce6e3-9a71-416b-99a7-606c5f7e2447?email=bob@reisner.info&redirect_to=https%3A%2F%2Fcodenewsletter.ai%2Fforms%2F14166360-de71-46c4-8722-878d417fab5c&utm_source=beehiiv&utm_campaign=BUBDNZO6MS&redirect_delay=3&_bhiiv=opp_0dc31371-131f-446b-9dba-52fc10e762c0_94e90c2e&bhcl_id=fb20a525-6361-460c-ba5e-b83d6fdfeef3_f672f2b6-1e60-4422-8448-ee13d3aa1099_22c35bc7-5e17-4448-baa5-49c4bf23f93e)

    #### **The Recap and My final thoughts…**

    Level 1: Pick one of the big three. Use the paid tier if you can. Always select the most powerful model. Go deep, the skills transfer.

    Level 2: Stop memorizing prompts. Outcome plus context wins. Save your recurring context in Projects so you never repeat yourself.

    Level 3: Connect your Projects into one AI system that spots patterns across your life and compounds every time you give it feedback.

    Most people are not at level 3 yet. Honestly, most people are barely at level 1. And that is fine. There is no rush.

    But here is the thing I wish I understood 8 months earlier: the gap between using AI and using AI well is invisible. You cannot see it on anyone's screen. You only see it in their results.

    Now you know exactly where you stand and what comes next.

    Stay curious, talk to you tomorrow.