Welcome back to the AI Edge (by Miles Deutscher) weekly newsletter!
This week, we’re breaking down how to build an AI side hustle from scratch.
The strongest place to start is finding an overlap between your existing skills and something you can realistically execute.
In this publication, we’re going to guide you through that entire process.
With GPT-6 Astra, the entire process, from finding potential business ideas and stress-testing them to creating the offer, building the website, mapping the workflow, and prototyping the actual product, has become 10x easier. This new model is seriously good.
In this guide, we’re turning that process into a system anyone can follow.
By the end of this read, you’ll know how to:
→ Find a side hustle that fits your existing skills
→ Choose the right business model
→ Validate demand and product-market-fit BEFORE building
→ Create the offer
→ Get your first real customer and how to scale
After our main guide, stick around for the Midweek Edge for this week’s AI news and Looking Ahead, where we compile our top AI research, trends, and workflows.
Let’s get right into things!
Table of Contents
Step One: Find Your Best Side Hustle
The whole point of the discovery process is to find a side hustle you can actually commit to.
Your skills, available time, budget, interests, and how you like working should shape the business you start.
Someone great with people may be suited to consulting, while another person who prefers working behind the scenes may enjoy building digital products.
AI can help you work this out.
Instead of asking GPT for a random list of businesses, have it interview you first.
Copy this prompt:
Help me identify a business I can realistically start and execute. Act as a rigorous business consultant: practical, commercially minded, and willing to challenge my assumptions.
Your objective is to find the best fit between my capabilities, my circumstances, and a problem customers will pay to solve. Do not assume I have an audience, customers, significant capital, or specialist experience.
Work through these stages interactively.
STAGE 1: DIAGNOSE
Interview me to understand:
- My strongest skills and evidence that I can deliver something valuable.
- My available time, budget, and deadline for earning income.
- The customers or industries I understand and can realistically reach.
- My willingness to sell, create content, learn tools, and deliver client work.
- The work I enjoy, dislike, or cannot do.
- Whether I prioritise near term income or building something that can grow beyond my own time.
Ask no more than three questions per message, then wait for my answers. Ask only about gaps that could change your recommendation. Use concrete choices or examples when a question is hard to answer.
If an answer is vague, follow up. If my goals conflict with my resources, explain the trade off. Do not recommend businesses until you have enough information. Summarise what you learn, separating facts from assumptions.
STAGE 2: RECOMMEND
Propose three specific business opportunities ranked by fit. Do not force variety if closely related opportunities suit me better.
For each, explain:
1. The specific customer and problem.
2. The exact offer and what the customer receives.
3. Why it fits me, citing something from the diagnosis.
4. How I could reach my first ten potential customers.
5. The minimum resources and skills required.
6. What GPT can help execute, which tools that requires, and what I must handle.
7. The biggest untested assumption.
8. The cheapest practical test of willingness to pay.
Distinguish evidence from estimates. Research current market claims when tools are available. Otherwise, mark them as unverified. Do not invent demand, customer results, or income projections.
Recommend one opportunity. Explain the trade off and what new information would change your choice. Get my feedback before proceeding.
STAGE 3: HAND OFF
Once we agree on an opportunity, create a concise business brief I can paste into a separate execution chat.
Include the customer, problem, offer, relevant strengths, constraints, acquisition approach, untested assumptions, first demand test, and a measurable seven day milestone.
Finish with the opening prompt for that execution chat.
Begin with your first questions.Answer the questions to the best of your ability.
The better GPT understands your actual situation, the more useful the recommendations become.
And if you already know roughly what you want to build, you can skip most of this and simply say:
“I want to start a digital product business, but I don’t know what the product should be. Interview me and help me find the strongest opportunity.”
Step Two: Choose the Right Business Model
In the AI space right now, there are primary business models.
1. AI Consulting
This is probably the easiest place to start if you want to get practical experience quickly.
You find a business problem, learn how to solve it with AI, then charge someone to implement the solution.
The exact problem could be anything. Examples below:
→ Leads generation
→ Customers not being followed up
→ Staff spend hours on repetitive admin
→ Content production
You do not need to automate the entire company. Solving one painful problem should come first, and ideally you have experience solving it and can use AI to amplify your offer.
A prompt to get started with consulting based on your existing skills:
I want to start an AI consulting business, but I have no consulting experience, clients, or audience.
Help me work out:
- Who to help and what problem to solve.
- A simple service I can learn to deliver.
- What skills and tools I need.
- How to find my first client and earn a genuine testimonial.
- What to charge and how to deliver the work.
Ask about my skills, budget, and available time first. Then recommend a starting point and give me a seven day plan to test it.
Once we have chosen the service, help me build the demonstration, landing page, offer, and outreach messages step by step.2. Digital Products
Digital products give you much more room to scale because you can sell the same product repeatedly.
This could include:
→ Software
→ AI agents
→ Templates
→ Dashboards
→ Courses or information products
→ Small online tools
For example, you can build custom trading indicators with TradingView and sell them.
The difficult part here is usually distribution, and a great product with no clear audience will still struggle to sell.
A prompt to get started with selling digital products based on your existing skills:
I want to create and sell a digital product, but I have no existing audience or customers.
Help me work out:
- Who to help and what problem they would pay to solve.
- A useful digital product I can realistically create with AI.
- How to test demand before building it.
- What to include, what to charge, and how to find my first buyers.
Ask about my skills, budget, and available time first. Then suggest three specific products, recommend one, and give me a seven day plan to test it.
Once we have chosen the product, help me build it, create the sales page, and prepare it for sale on Whop. Prefer the Whop CLI for setting up the product, pricing, and checkout. Check that it is installed and signed in before using it.3. Physical Products
This is the most ambitious (and difficult) path of the three.
Astra can now help with product concepts, 3D models, component layouts, supplier research, and other parts of the development process.
Most beginners shouldn’t start here; we typically recommend the two models above as starting points.
A simple way to choose the right model:
Business model | Best if you want | Main challenge |
Consulting | The quickest path to real customer experience | You are selling your time and delivery |
Digital product | Something more scalable | Distribution and finding a real niche |
Physical product | To build something tangible | Manufacturing, capital, and operations |
Pick the model that fits you, and walk your AI (GPT-6 Astra) through building with it.
Step Three: Validate Before You Build
The most important step.
An AI model can come up with an excellent-sounding business idea in seconds, but the market still decides whether it’s actually valuable, and thus, worth paying for.
Here’s what you should do:
Start by testing the problem
For a consulting offer, speak to a real business CEO and understand how they currently handle the problem.
For a digital product, build a simple demo or landing page and see whether the target user actually wants it (for free).
For a physical product, test the concept with potential buyers before spending serious money on inventory.
A helpful sequence framework for validating PFM:
Problem → Demo → Feedback → Paid Test → Build
A simple place to start is with a business owned by someone you already know.
Most people know a family member, friend, or someone they can help directly with low stakes.
Offer to implement your solution, grab a testimonial, and iterate on their feedback.
Stress Test Your Idea
Once you have a business idea, send this prompt to your AI before building it:
Act as a skeptical business operator.
I am considering selling:
[describe the offer]
To:
[describe the target customer]
Stress test this idea.
Identify:
1. The exact problem I am assuming exists
2. How customers currently solve it
3. The strongest competitors or substitutes
4. Why someone may refuse to pay
5. What I would need to prove before building
6. The cheapest real world test I can run
7. What evidence would tell me to continue
8. What evidence would tell me to change or abandon the idea
Do not assume the idea is good.You can use this as a brainstorming session to generate ideas, then immediately make it attack those new ideas.
Step Four: Build Your Launch Kit With AI
Once you’ve stress-tested an idea, you can move on to the actual launch.
For most side hustles, the initial launch kit is fairly simple:
→ A clear offer
→ Pricing
→ Simple website or landing page
→ Demo or prototype
→ Outreach copy
→ Prospect tracker/CRM
→ Delivery process
Cust
For Consulting
AI can create the business name, landing page, pricing, outreach copy, CRM, and launch materials.
The first version does not need to be perfect, but your website mainly needs to answer:
Who is this for?
What problem do you solve?
What result are you offering?
What should the customer do next?
For Digital Products
The workflow is:
Problem → Prototype → Product → Checkout → Landing Page
Astra can create the first prototype and then help package it into a product people can actually buy.
Whop is especially useful here because its CLI lets you manage products, pricing, checkout, and other parts of the business from the terminal.
Step Five: Land Your First Customer
Your first goal is proof.
With a consulting business, this means completing one useful AI implementation and documenting the result.
For a digital product, it means getting real users or buyers.
Once you have something that works, you need to capture evidence.
What was the problem?
What did you build?
What changed?
Would the customer recommend it?
That becomes your first case study.
From there, your acquisition channels become easier.
Your content also becomes much stronger once you have actually done something.
Instead of posting generic AI advice, you can document a real build:
“I helped a local business automate its customer follow-ups.”
Or:
“I built my first paid AI tool from scratch.”
Or:
“I’m documenting the journey from $0 to my first $10K in consulting revenue.”
Real work → proof → authority and trust → new buyers
Proof also gives you better content (and distribution).
TL;DR: Your first customers come from outreach (cold calls, emails, DMs, etc.) Once you get them results, document it publicly and build online distribution to start getting warm leads.
Final Tips
Aim for one real customer first. You do not need the perfect logo, business name, company structure, and automation stack before proving that someone wants the offer.
Keep your side hustle inside one AI Project. Store your customer research, offer, pricing, objections, website copy, results, and operating notes in the same workspace so your AI keeps the relevant context (GPT projects).
Do the work before automating everything. Understanding the process manually makes it much easier to know what should eventually become an AI workflow.
Be careful with high-risk products. Trading tools, financial products, physical products, and other regulated categories can bring headaches you may want to avoid (legal, restrictions, etc.).
Treat every attempt as skill building. Even a side hustle that fails can leave you much better at building websites, creating products, talking to customers, using AI, and solving business problems. Those skills can become the foundation for the next offer.
The playbook is:
Find a problem → Validate it → Build the smallest useful offer → Get proof → Repeat what works.
Midweek Edge
Our manually curated list of the most important news updates across AI, robotics & tech.
Meta Launches Muse
Meta just launched Muse, a personal AI agent that can browse, email, shop, book travel, and keep working after you close the app.
It runs on its own cloud computer and asks for approval before sensitive actions.
Available now in the US on iOS, Android, WhatsApp, and muse.ai.
OpenAI Releases ChatGPT Images 2.5
OpenAI just released ChatGPT Images 2.5.
→ Up to 50% faster generation
→ Better image quality and reference fidelity
→ More precise editing
→ New Sketch tool
Available now across ChatGPT, Work, and Codex.
OpenAI Solves 90-Year-Old Math Problem
OpenAI has shared an AI-generated solution to the Navier-Stokes Millennium Prize Problem, produced by an internal model it says is significantly more capable than GPT-6 Astra.
They released the full write-up and formal Lean proof for mathematicians to verify.
AI Leaders Call to Pace Frontier AI
Anthropic CEO Dario Amodei is calling for frontier AI development to slow enough for safety work to keep up.
Sam Altman agreed, saying OpenAI will match Anthropic’s commitment to give independent evaluators employee-like access. Elon Musk also backed the proposal.
DeepSeek Releases V4.1 Flash
DeepSeek just released V4.1 Flash, its new multimodal model built for faster, cheaper agent workloads.
It is live now on the DeepSeek API as deepseek-flash.
Try it for coding, agents, or multimodal work.
OpenAI Releases GPT Live 1 API
OpenAI just brought GPT Live 1 to the API.
It can listen and speak at the same time, handle interruptions naturally, and delegate deeper work to other models and tools.
Developers can start building with it now.
Anthropic Researchers Warn AI Could Kill Humans
Former Anthropic researcher Jacob Coxon resigned last week, warning that frontier labs are racing toward self-improving AI without a clear safety solution.
Anthropic Alignment Science lead Evan Hubinger backed the concern and said his personal estimate of AI killing all humans within the next decade is above 10%.
The AI Slowdown Debate is Becoming Political
President Trump has rejected calls to slow AI development, arguing that the US cannot afford to lose its lead to China.
China and the EU also pushed back against the slowdown argument.
The AI safety debate is now colliding directly with the global AI race.
Looking Ahead
Our manually curated list of the top recent AI workflows, tools, & additional research
The Only AI Tools You Need Right Now
→ Grok Bot: personal agent swarms
→ GPT 6 Astra: heavy work
→ Sonnet/Opus: daily driver models
→ Hermes Agent: daily agent workflows
→ Gemini: Google native tasks
Agentic Research Framework
One of the best agent research frameworks on the internet.
This skill gives your AI agent access to Instagram, Facebook, LinkedIn, X, and more.
The best social data scraper for Hermes and OpenClaw.
→ https://github.com/Panniantong/Agent-Reach
15 Astra Agents (steal these)
Closing out
If you made it this far, thank you for reading, and we hope you found this week’s edition valuable.
If you enjoy reading, please forward our newsletter to someone you think would benefit from it.💙
Our promise to you: Every Wednesday, at 7 am EST, we’ll cut through the AI noise and send you human-curated AI content to make sure you stay ahead.
Content Pipeline
YouTube
Live Now: The Most Valuable Skill to Make Money in 2026: AI Agents (Step by Step)
Featuring: Ras Mic
X (Twitter)
Live Now: 8 Hermes Agent Skills (Prompts Included)
Free Skool Community
For exclusive behind-the-scenes, including AI prompts, cheatsheets, and PDF guides, join the free AI Edge Skool community here:
See you next Wednesday!
Interested in sponsorship or a partnership? Get in touch at [email protected]