How do I learn AI without coding and get from chat user to operator?

You do not need to code to use AI seriously, and the gap between a chat user and an operator is a set of habits, not a programming language. The path runs through six things: how a model actually works, briefing it the way you would brief a new hire, checking what it tells you, picking the right tool for each job, connecting it to your own data, and building one small automation that runs without you. I built AI Academy around that sequence after ten years in growth roles, and this page is the map of the course with the reasoning behind each step.

Updated November 2026: ChatGPT's pricing page now lists "Projects, tasks, and custom GPTs" on Plus, and the Agent Skills site moved its list of supporting tools to a client showcase page; plan prices, module counts and the EBU figures were re-checked and stand.

Do I need to code to use AI seriously?

No. Every module in the beginner volume of AI Academy can be completed with a browser, a chat assistant and a spreadsheet. The first time you see code is in the advanced volume, and even there a coding agent writes it while you read, question and approve it.

What changes between a chat user and an operator is the unit of work. A chat user asks a question and reads an answer. An operator gives a tool a job, a source of truth, a way to check its own output and a place to put the result. None of those four things requires code. They require knowing what the tool can and cannot see.

I came to this from the operator side. I ran growth for a two-sided sports travel marketplace, was the first marketing hire at Fangage through its acquisition by Triller, and now run a consultancy where AI does the sourcing, research and drafting and I do the judging and sending. I did not learn Python to get there. I learned how to brief, how to check, and how to connect tools through settings screens.

The course has 92 modules in five volumes. The table shows who each is for and how long it takes at a pace of one module a day, which is a realistic pace for someone with a job.

Volume	Level	Modules	Who it is for	Time at one module a day
AI Foundations	Beginner	29	Anyone who has used ChatGPT or Claude and wants to stop guessing	About six working weeks
AI Builder	Advanced	21	Marketers, founders and consultants ready to build their first automation, skill and agent loop	About four working weeks
AI Delivery Skills	Expert	7	People who ship AI work to clients or internal stakeholders and need it to hold up	About a week and a half
AI Strategy and Implementation	Expert	11	Operators deciding what to automate, in what order, and how to measure it	About two working weeks
AI-Native Operator's Field Manual	Expert	24	People running AI daily across a business and wanting the full set of working methods	About five working weeks

The three expert volumes are not sequential. Delivery Skills is for people who present AI work to others, Strategy and Implementation for people deciding budgets, and the Field Manual is the one I update when my own stack changes.

What should I learn first?

Learn the 20% of how a model works that explains 80% of its failures, then learn to brief it, then learn to check it. In that order, because briefing without understanding produces confident nonsense, and checking without a method produces anxiety.

The 20% is small. A language model predicts the next piece of text from everything you gave it plus what it absorbed in training. It has no memory of your business unless you put it in the conversation or connect it. It has no idea whether what it wrote is true. It does not look anything up unless a tool is attached and switched on. That is the whole mental model, and How AI works spends one module making it stick with examples you can reproduce.

Once that lands, prompting stops being a trick and becomes briefing. I teach it as briefing a literal new hire who is very fast, very well read and has never seen your company. You would not say "write me a LinkedIn post" to that person. You would say who it is for, what you want them to do after reading it, what tone the company uses, two examples that worked and one that did not, and what to leave out. The prompting module turns that into a brief template you will reuse for everything.

Fact-checking is the third skill and the one most people skip. The European Broadcasting Union and the BBC had journalists from 22 public service media organisations evaluate more than 3,000 news answers from ChatGPT, Copilot, Gemini and Perplexity. The EBU study found that "45% of all AI answers had at least one significant issue," and 31% had sourcing problems. Those are the best-known tools answering questions with public answers, and your internal numbers will not do better.

The fact-check module gives you a three-step routine: ask for the source, open the source, compare the claim to the source. In my own work the routine runs before anything leaves my hands. Every outreach message and every client summary goes through a check for invented numbers and machine-written phrasing. The habit costs minutes. The five-step version with timings, and what each step catches, is in the five-minute hallucination check.

After those three, the beginner volume widens out: the tool map, research with AI, coding agents for non-coders, MCP, no-code automation and the AI operating system, which is where you decide what your own daily setup looks like.

Which AI tool should I use for which job?

Pick one general assistant and stay with it for a month, then add tools by job rather than by hype. ChatGPT and Claude cover the same ground for writing, analysis and research, and the differences that matter to a non-technical operator are in what they connect to and how they handle long, structured work.

I use Claude for most of my own work because of projects, skills and MCP connectors. I keep ChatGPT for image generation and because clients use it and I need to know what they see. Gemini comes in when the work lives in Google Workspace. The wrong choice is three subscriptions and no habit. The job-by-job ChatGPT and Claude comparison goes through the two products row by row.

Features change monthly, so I link the vendor pages and you should read them on the day you decide.

Job	What to reach for	Why	Vendor page
Daily writing, summarising, thinking out loud	ChatGPT or Claude, free tier to start	Both free plans cover text chat, file upload and memory in some form	ChatGPT pricing, Claude pricing
Long documents and repeat work with the same context	Claude Projects, or ChatGPT Projects on Plus	Projects hold files and instructions so you stop re-pasting. Claude's free plan lists "Up to 5" projects; ChatGPT lists "Projects, tasks, and custom GPTs" from Plus	Claude pricing, ChatGPT pricing
Deep research with sources	ChatGPT deep research, Claude Research, Gemini Deep Research	ChatGPT shows "Limited deep research" on Free and "Expanded" on Plus; Claude lists Research from Pro; Gemini lists Deep Research on its free plan	ChatGPT pricing, Claude pricing, Gemini subscriptions
Work inside Gmail, Docs and Sheets	Gemini on Google AI Pro	The paid tiers add "Gemini in Gmail, Docs, Vids, and more"	Gemini subscriptions
Connecting AI to your own apps	Claude connectors (free plan lists "Connect your apps and tools"), ChatGPT apps on Plus	This is where MCP lives, see the next section	Claude pricing, ChatGPT pricing
Building a small tool or script without writing it yourself	Claude Code (from Pro) or ChatGPT Codex (limited on Free, expanded on Plus)	A coding agent writes and runs code while you direct it in plain language	Claude pricing, ChatGPT pricing
Images and short video	ChatGPT for images, Gemini for video	ChatGPT lists image creation from the free tier up; Gemini's paid tiers include video generation	ChatGPT pricing, Gemini subscriptions
Repeatable multi-step workflows that run on a schedule	n8n, Make or Zapier, with an AI step inside	A chat assistant is the wrong shape for "every Monday at 8", an automation tool is the right one	Covered in the automation section

Two things from the table that trip people up. "Memory" means different things on each platform and none of them means the tool knows your business. And the free tiers are enough for the whole beginner volume of the course, so pay for a plan when a specific job in the table needs it, not before.

The tool map module goes through this table with screenshots and carries a dated review note so you know when it was last checked against the live product.

What do MCP, agents, RAG and skills mean in plain words?

These four words are the vocabulary that separates someone who uses AI from someone who runs it, and all four describe plumbing you can set up through a settings screen. MCP is a plug standard, an agent is a worker with a goal, RAG is giving the model a filing cabinet, and a skill is a written procedure the model follows.

Term	Plain definition	What it looks like in practice	Module
MCP (Model Context Protocol)	An open standard for connecting an AI app to outside systems, the way USB-C connects devices. The official docs describe it as "an open-source standard for connecting AI applications to external systems"	You add a connector in Claude or ChatGPT and the assistant can now read your Notion, your calendar or your ad account and act on them	MCP for beginners, MCP as a consumer, MCP in the Field Manual
Agent	A model given a goal and a set of tools, which chooses its own steps, checks results and keeps going until the goal is met or it gets stuck	"Find the 20 companies in this list that hired a head of marketing in the last 90 days and draft a note for each" with the agent running searches, reading pages and writing	First agent loop
RAG (retrieval-augmented generation)	Giving the model a searchable store of your own documents so it answers from them instead of from memory	Upload your pricing document and sales playbook once, and every answer about pricing cites the right page	RAG with one document
Skill	A folder with a SKILL.md file holding instructions, and optionally scripts and templates, that the agent loads when a matching task shows up. The open standard calls it "a lightweight, open format for extending AI agent capabilities"	My LinkedIn connection-note skill holds the voice rules, the banned phrases and the character limit, so every draft starts from the same brief	Your first skill, Skills in the Field Manual
Evals	A fixed set of test inputs with known good answers, run against a prompt, skill or agent every time you change it	Twenty real briefs with the output you would accept, re-run after every prompt edit, scored by hand	Basic evals
API	The door a software product opens for other software. Automation tools and coding agents use it so you do not have to	An n8n node that reads rows from Google Sheets is calling the Sheets API behind the scenes	API first

The Model Context Protocol documentation is readable by a non-technical person and worth twenty minutes. The Agent Skills site shows the folder layout and keeps a client showcase of the agent products that read the format, so a skill you write once travels. The MCP guide in plain words walks through the same idea with the setup you can do from a settings screen.

My own stack is the worked example in the MCP modules. Google Ads and Meta run through MCP servers so I can ask for a performance pull in plain language and get a table back. HeyReach handles LinkedIn sequences, Apollo enriches contacts, Notion is the pipeline CRM, and Claude skills draft the messages. AI sources, researches and drafts, and I judge and send every message myself. If 45% of public answers carry a significant issue, the step where a human reads the output is the step you cannot automate away.

How do I build a first automation without code?

Build one that moves information from a place it arrives to a place it is useful, with a single AI step in the middle, and run it on real data for a week before trusting it. The shape I recommend is trigger, AI step, destination, and nothing else until that works.

A concrete first build: a form submission lands, an AI step classifies it as lead, support or spam and writes a two-line summary, and the result is posted to a Slack channel or appended to a sheet. That is three nodes in n8n, Make or Zapier. It has no branches and nothing that can silently delete data. The beginner no-code automation module builds this exact flow step by step. The advanced n8n first module rebuilds it in n8n with error handling and a log. The guide on your first automation in n8n or Zapier gives the clicks, the brief for the AI step and the monthly cost.

Five rules I apply to every automation I build.

1. Draw it on paper first. Trigger, steps, destination. If the drawing has more than five boxes, cut it in half and build the first half.
2. Give the AI step a brief, not a prompt. The same template as the prompting module: who the output is for, what good looks like, two examples, what to leave out.
3. Write the output somewhere a human reads it before it writes anywhere a customer sees it. Slack channel or sheet first, email or CRM later.
4. Log every run. Input, output, timestamp. When something goes wrong you will want to know whether it was the trigger, the model or the destination.
5. Run it in parallel with the manual process for a week. Compare. Only then switch the manual process off.

The place people get stuck is the brief inside the AI step. A classification step with "decide if this is a lead" fails on edge cases a human would catch. One with "a lead is a message from a company email asking about pricing, timelines or a call; support is an existing customer with a problem; spam is anything selling us something" works. Writing that sentence is a writing skill.

Once the three-node flow runs, the next step is the API first module, where a coding agent calls an API on your behalf and you learn to read the response. After that, RAG with one document lets the AI step answer from your own files, and the first agent loop replaces the fixed steps with a goal.

How do I tell if I am getting better?

You are getting better when you can predict what the tool will get wrong before you read the output, and when things you used to do by hand run without you. Both are checkable, so the 30-day plan below sets a checkpoint each week that you can pass or fail.

Week	What you do	Modules	Checkpoint
1	One module a day from Foundations: how AI works, prompting, fact-checking, the tool map, research	How AI works, Prompting, Fact-check, Tool map, Research	You have a saved brief template you used on three real tasks, and you caught one invented fact or source in an AI answer and can say how
2	Set up one project with your real documents, connect one MCP connector, and use a coding agent once for a small task	MCP, Coding agents, AI operating system	The assistant answers a question about your business from your files, with a citation to the right page, and you did not paste anything
3	Build the three-node automation: trigger, AI step, destination. Run it alongside the manual process	No-code automation, n8n first	The automation ran on at least 20 real inputs, you compared every output to what you would have done, and you can name the cases it got wrong
4	Write your first skill from the brief you have been reusing, build a 10-item eval set, and re-run it after one edit	Your first skill, Basic evals	The skill produces output you would send without edits on at least half of the eval items, and you can show the score before and after your edit

The week 3 checkpoint is the one that matters. Twenty real inputs is small enough to review by hand in an evening and large enough to surface the cases the brief did not cover. In my own builds, the first twenty runs are where the gaps in the brief show up, and fixing them is the lesson.

If you fail a checkpoint, repeat the week. The expert volumes assume the four checkpoints are behind you.

Beyond 30 days, the measure is hours. Keep a list of tasks you did by hand in month one and in month three count how many run without you. The AI operating system module gives you the list template, and the GEO module in the Field Manual is where the same methods get pointed at being found in AI search, which is the next question most marketers ask once the basics run.

How much should an AI course cost and what does free get you?

An AI course should cost less than one month of the tool subscriptions it teaches you to use well, and it should let you finish real modules before you pay anything. AI Academy's paid plans are EUR 9.99 a month, EUR 69.99 a year or EUR 129.99 once, and the first three modules at each level are free with no card.

Free YouTube and vendor documentation teach features, and they are good at it. What they do not give you is a sequence, a project at the end of each module, and someone who goes back and re-checks the lesson when the vendor changes the product. Every lesson in AI Academy carries a dated review note saying when it was last checked against the live tool. That maintenance is most of what the fee pays for.

Plan	Price	What you get	Who it suits
Free	EUR 0, no card	The first three modules of each level: beginner, advanced and expert	Anyone deciding whether the course fits before paying
Monthly	EUR 9.99 a month	All 92 modules across the five volumes, dated review notes, cancel any time	Someone planning to work through Foundations and Builder in two or three months
Yearly	EUR 69.99 a year	Everything in monthly, for about 58% of the price of twelve months	Someone who wants the expert volumes and expects to come back as the tools change
Lifetime	EUR 129.99 once	Everything, including future modules and reviews, no renewal	Operators who will use this as a reference for years

The yearly saving is arithmetic: twelve months at EUR 9.99 is EUR 119.88, and the yearly plan is EUR 69.99.

Compare that to the tools. Claude Pro is USD 20 billed monthly and Google AI Pro is USD 19.99 a month on their pricing pages linked above; ChatGPT's page does not print amounts, so check it on the day. If a course does not pay for itself in the first month by stopping you from paying for a tier you did not need, or by getting one automation running, the course is wrong or you are not doing the exercises.

What free gets you is enough to judge. Three beginner modules will tell you whether the explanation style works for you, three advanced modules whether you are ready to build. If you finish the free beginner modules and the brief template is already in use on real work, buy. If it is not, do not.

I run AI Academy through RLGrowth OÜ, the same company that runs my consultancy, and the course is the written version of what I do for clients. The first three modules are at learn.growthmarketing.solutions/academy.

This guide is also available in German: KI ohne Programmieren lernen.