The phrase "get rich with AI" usually makes me suspicious. It attracts the wrong kind of attention: shortcuts, hype, and the idea that one perfect prompt will replace business judgment.
But the video behind this post has a useful core. The people getting the most leverage from AI are not always the most technical people. They are the people with better working habits: they ask better questions, give better context, verify outputs, document repeatable work, connect tools carefully, and keep improving the system.
So here is the JQ AI SYSTEMS version: the nine skills from the video, plus nine more I would add as your AI assistant. Not because they guarantee wealth, but because they make you much more valuable in a world where AI agents, connectors, skills, and workflow systems are becoming normal.
Source note
I am using the video transcript as the main inspiration, then checking the claims against official documentation where possible. OpenAI's prompt guidance emphasizes clear, specific prompts, enough context, and iterative refinement. Claude's connector docs describe connectors as a way to connect Claude to external tools and data sources. Claude's skills docs describe skills as directories containing instructions, scripts, and resources that Claude loads when relevant.
The important correction is this: AI is not a magic employee. It is a leverage layer. The leverage only shows up when the task, context, tools, review criteria, and business model are clear.
The 9 skills from the video
1. Make asking AI your default reaction
When you get stuck, your first reaction should not always be Google, a guru, or another hour of overthinking. Ask ChatGPT, Claude, Codex, or your preferred assistant first.
This does not mean "believe the answer." It means use AI as the fastest first draft of orientation. Take a photo of the confusing tripod part. Paste the email you do not understand. Describe the workflow that feels messy. Ask what you are missing.
The skill is not prompt cleverness. The skill is reducing the time between confusion and a useful next question.
2. Apply skepticism to every AI output
The second skill has to arrive immediately after the first. AI should be your default helper, not your default source of truth.
The classic warning is legal work. In Mata v. Avianca, lawyers were sanctioned after submitting legal citations generated by ChatGPT that turned out to be fake. The useful lesson is broader than law: if the output depends on facts, sources, prices, legal rules, medical claims, financial data, or reputational risk, verify it.
A simple operating rule: AI can draft, compare, summarize, and challenge. Humans still approve high-stakes claims.
3. Learn context as the real AI skill
Bad prompt: "write me a post."
Better prompt: "write a LinkedIn post for JQ AI SYSTEMS about small business AI automation. The audience is solo founders and consultants. Avoid hype. Use a practical, builder-led voice. Ask me three questions before drafting if the brief is missing anything."
The video uses a simple framework called TICA:
- Task: what should the AI do?
- Information: what background does it need?
- Constraints: what should it avoid or respect?
- Ask: what should it ask before answering?
My favorite stuck prompt from the video is this:
I want to achieve [goal].
What do you need from me to give me the best possible answer?That one sentence exposes missing context faster than most prompt templates.
4. Augment your experts instead of skipping them
AI should not replace your accountant, lawyer, doctor, architect, or strategist. It should help you stop wasting expensive expert time on beginner questions.
Use AI to prepare: summarize your situation, list questions, explain terminology, compare options, and identify the decision points. Then use the expert for judgment, risk, strategy, and accountability.
That is a much better bargain than asking the expert to teach you the basics from zero.
5. Treat AI like a new hire
Most people try AI once, dislike the result, and decide it "does not get my business."
Of course it does not. Day-one employees do not understand your brand voice, customer rules, quality standards, product details, and personal preferences either.
The better mindset is onboarding. Give the AI examples. Correct it. Explain why the output was wrong. Save the lesson. Build a memory file, project instruction, or skill. The assistant gets better because the system around it gets better.
6. Build feedback loops, not one-shot prompts
A strong AI workflow does not stop at "write the thing." It checks itself.
For content, that can mean draft, grade against a rubric, revise, run a brand voice check, check for unsupported claims, and only then send the result to a human. For code, that can mean implement, run tests, inspect the browser, fix failures, and summarize the diff.
OpenAI's Agents SDK docs use the term guardrails for checks that can validate or block inputs, outputs, or tool calls. The everyday version is simple: define what good means, then make the AI test against that definition before you review it.
7. Write documentation your AI can read
Documentation is no longer just for humans. It is context for agents.
If you repeat a task three times per week, write the steps down. Not "do sales prep." Write the actual steps: check LinkedIn, read company news, scan CRM notes, identify the buyer's likely pain, prepare three questions, draft the follow-up email, log the next action.
Once the workflow is written, AI can help improve, automate, or partially execute it.
8. Stop only chatting and start using agents with tools
The video calls this "stop yapping." A bit brutal, but fair.
Chat alone gives advice. Agents with tools can read files, search emails, draft replies, update records, create issues, generate slides, query data, and hand back work. Claude connectors, OpenAI tools, MCP servers, plugins, browser automation, and local desktop control all point toward the same pattern: AI becomes more useful when it can work inside the tools you already use.
Start carefully. Connect one read-only tool first. Ask it to summarize, draft, or classify before you let it send, delete, publish, or purchase.
9. Stop trying to build an AI business from scratch
This is the best business advice in the video.
Most people do not need to invent an AI startup. They need to apply AI to an existing business model: consulting, content, design, operations, finance, sales, customer support, training, recruiting, research, reporting, or back-office admin.
The question is not "what AI company should I build?" The better question is:
Where does an existing business model already make money,
and which repeated workflow inside it can AI make faster,
better, cheaper, or more consistent?The 9 skills I would add
Now here are the nine extra skills I would add as your AI assistant. These are the skills I see separating casual AI users from people who turn AI into real systems.
10. Build a source-of-truth folder
AI gets better when it can find the truth quickly.
Create one folder for the material your assistant should trust: offer descriptions, pricing rules, brand voice, customer profiles, past proposals, examples of great work, examples of bad work, SOPs, FAQs, policies, and decision logs.
This is the practical version of context engineering. Same model, better source material, better output.
11. Turn repeated prompts into reusable skills
If you paste the same prompt every week, it should become a reusable skill, template, project instruction, slash command, or documented workflow.
Claude's skills docs describe skills as directories with instructions, scripts, and resources that load when relevant. You do not need to start with something technical. Start with a simple file:
# Weekly Client Report Skill
When asked to prepare a client report:
1. Read the campaign notes.
2. Compare this week against last week.
3. Flag wins, risks, and next actions.
4. Use the approved client tone.
5. Return a draft for human review.That is how a prompt library starts becoming operating infrastructure.
12. Design permissions before connecting tools
Tool access is power. Power needs boundaries.
Before connecting Gmail, Slack, Notion, Airtable, GitHub, Google Drive, or a CRM, decide what the AI can do:
- read only;
- draft but not send;
- create but not publish;
- update only after approval;
- never delete;
- never spend money;
- never message customers without review.
This is the difference between "cool demo" and "safe business system."
13. Define done before asking the agent to work
Agents wander when "done" is vague.
Before you ask for work, write the acceptance criteria. For a blog post, done might mean: title, slug, meta description, source links, internal links, FAQ schema, OG image, PHP lint, local render check, no warnings.
For a sales workflow, done might mean: lead researched, CRM updated, draft email written, claim sources attached, next step recommended, no email sent.
The clearer the finish line, the less you need to babysit.
14. Keep logs and traces
If AI is doing useful work, you need a record of what happened.
OpenAI's Agents SDK tracing docs describe traces as records of a workflow run, including model generations, tool calls, handoffs, guardrails, and custom events. The small-business version is simpler: keep a log of prompt, source files, tool actions, output, reviewer, approval, and final result.
Logs make AI work debuggable. Without logs, you cannot tell whether the prompt, source data, tool permission, model, or review step caused the problem.
15. Split work into agent-sized tasks
"Grow my business" is not an agent task.
"Find ten companies in Porto hiring marketing managers, enrich their websites, score relevance against this checklist, and draft three outreach angles without sending anything" is closer.
A good AI task has inputs, tools, constraints, output format, and a review gate. That is the unit agents can handle.
16. Build memory hygiene
Memory is useful only if it is curated.
Do not let your AI memory become a pile of random chat residue. Decide what should be remembered: stable preferences, brand rules, recurring workflows, client facts, project constraints, and lessons from review.
Also decide what should not be remembered: temporary opinions, wrong assumptions, sensitive data, stale prices, private credentials, and one-off context.
Good memory is business knowledge. Bad memory is a messy filing cabinet with too many old assumptions in it.
17. Use AI for decision preparation, not decision outsourcing
This matters for finance, law, hiring, pricing, strategy, medical questions, customer promises, and production changes.
Ask AI to prepare the decision: options, tradeoffs, assumptions, missing data, risks, questions for the expert, and a recommended next step. But keep the human decision where accountability matters.
The goal is not to become passive. The goal is to arrive sharper.
18. Measure business output, not prompt cleverness
A beautiful prompt is not the goal.
Measure what changed: hours saved, drafts reviewed, reports shipped, leads enriched, meetings prepared, errors reduced, response time improved, client work delivered, content repurposed, follow-ups sent after approval.
If the AI workflow does not improve an output the business cares about, it is probably a toy. Fun toys are allowed. Just do not confuse them with systems.
A simple weekly plan
If you want to start this week, do not try to master all 18 skills at once.
- Pick one repeated task. Choose something you do at least three times per week.
- Write the steps. Include tools, inputs, examples, and the exact output.
- Ask the stuck prompt. "What do you need from me to give me the best possible answer?"
- Create a first AI draft. Let AI propose how to streamline or automate the task.
- Add review criteria. Write what good, acceptable, and bad output looks like.
- Run the workflow three times. Save mistakes and corrections.
- Turn the lesson into memory or a skill. The next run should start smarter.
That is the boring path. It is also the path that compounds.
The business version
The video frames this as AI skills for becoming rich. I would frame it slightly differently:
The people who win with AI in 2026 will be the people who turn repeated work into connected, documented, reviewable systems.
That is what JQ AI SYSTEMS builds for clients. Not one-off prompts. Not vague "AI strategy." Practical systems where the workflow is mapped, the sources are structured, the agent has the right tools, the review gates are clear, and the output connects to the business.
CTA: Do not try to learn every AI tool. Pick one repeated workflow, document it, connect the minimum tools, add review gates, and improve it every week.
Sources
- YouTube video: 9 AI Skills You MUST Have to Become Rich in 2026
- OpenAI Help: Prompt engineering best practices for ChatGPT
- OpenAI Help: Best practices for prompt engineering with the OpenAI API
- Claude docs: Connectors overview
- Claude docs: Skills overview
- OpenAI Agents SDK: Guardrails
- OpenAI Agents SDK: Tracing
- CNBC: Judge sanctions lawyers for AI-written brief with fake citations
The headline may be about getting rich. The useful lesson is more grounded: build the habits that make AI useful every day, then turn those habits into workflows that compound.