For most people, using AI still means opening ChatGPT, Claude or Gemini and typing something into a chat box.
Write this email.
Summarise this document.
Give me 20 ideas.
Research this topic.
That’s useful. But it’s a tiny fraction of what you can now do with AI.
The much bigger shift is building with AI.
You can have an idea for something that doesn’t exist and make it.
You can create a tool because you’re sick of doing something manually.
You can turn a messy spreadsheet into something you actually want to use.
You can build a workflow that does the boring parts of your job.
You can create an AI agent that researches something for you.
You can prototype a product idea before spending money hiring anyone.
You can build a tiny app that only you need.
And increasingly, you can do all of this without being a developer.
I know because that’s exactly how I got here.
I don’t have a traditional software engineering background. My background is business, marketing, UX and CRO. But I’ve now built and shipped more than 20 apps using AI.
The important thing I’ve learned isn’t how to code.
It’s how to build with AI.
And those are not the same thing.
What does “building with AI” actually mean?
Building with AI means using AI to help you turn an idea, problem or process into something that works.
That “something” could be:
- a personal tool
- an AI workflow
- an automation
- an AI agent
- a research system
- a dashboard
- a website
- an internal tool for your team
- a prototype
- an app
- a new product
- or something that doesn’t fit neatly into any of those categories
The important distinction is that you’re no longer just asking AI to give you an answer.
You’re working with AI to make something.
That’s the shift.
Using AI: “Can you analyse these customer reviews?”
Building with AI: “I analyse customer reviews every month. Let’s create a system where I can drop in the latest reviews and automatically get the recurring complaints, requested features, sentiment changes and opportunities.”
Or:
Using AI: “Help me plan what to pack for Japan.”
Building with AI: “I travel constantly and hate making packing lists. Let’s build a tiny app that creates my packing list based on destination, weather, trip length and what I’m doing there.”
Or:
Using AI: “Summarise these sales calls.”
Building with AI: “Every sales call contains customer intelligence we’re losing. Let’s create a workflow that turns every transcript into objections, buying triggers, product feedback and follow-up actions.”
Same AI.
Completely different way of thinking.
What can you build with AI?
This is where I think people massively underestimate what has changed.
The answer isn’t “apps.”
And it definitely isn’t “chatbots.”
You can build almost anything that involves information, decisions, repeatable processes, interfaces or code, provided the problem is appropriate for AI and you understand the limitations.
Personal tools
This is one of my favourite uses of AI because the economics have completely changed.
For years, we’ve adapted ourselves to software.
You need a CRM, so you learn how somebody else’s CRM works.
You need to track something, so you find an app that’s close enough.
You need a tiny feature, so you subscribe to a $19-a-month product containing 47 features you’ll never touch.
But what happens when building a tiny tool for yourself takes an afternoon?
You might build:
- your own content tracker
- a packing-list generator
- a personal research database
- a meal planner that works exactly the way you eat
- a dashboard for something you care about
- a family organiser
- a symptom tracker
- a podcast management system
- a tool that reformats something you constantly have to reformat
It doesn’t need a market.
It doesn’t need customers.
It doesn’t need to become a startup.
It can exist simply because you need it.
That is a profound change.
Workflows and automations
Not everything needs an interface.
Sometimes the best thing you can build is a process that quietly removes work.
Maybe every Friday you collect information from five places, analyse it and create a report.
Maybe every new enquiry needs to be categorised and routed.
Maybe you repeatedly turn long documents into the same structured output.
Maybe you’re constantly moving information between tools.
Those are building opportunities too.
A traditional automation generally follows predefined steps. An AI workflow can incorporate a model into those steps to do things that require interpretation, such as extracting information, classifying something or drafting an output.
The important question isn’t:
“How can I add AI to my business?”
It’s:
“Why am I still doing this manually?”
AI agents
Agents are currently everywhere, and that creates another problem: people assume everything should be an agent.
It shouldn’t.
An AI agent is useful when the system needs some freedom to decide what to do next in pursuit of a goal.
A workflow is more appropriate when you already know the sequence of steps.
If every invoice needs to follow the same five-step process, you probably don’t need an autonomous agent deciding what happens next.
If you want AI to research a question, decide which sources it needs, investigate them, identify gaps and continue until it has enough information, an agent might make much more sense.
More autonomy isn’t automatically better.
Build the simplest thing that solves the problem.
That’s one of the most important principles in building with AI.
Internal tools
This is where things become particularly interesting for companies.
The person who understands a problem best has traditionally not been the person capable of building the solution.
An operations manager knows exactly why a process is ridiculous.
A salesperson knows which information they constantly need but can never find.
A marketer knows which report takes four hours every Monday.
A customer service team knows which questions keep appearing.
But turning that knowledge into a tool traditionally meant getting budget, writing requirements, finding developers, joining a backlog and waiting.
AI is beginning to compress that gap.
The person with the domain knowledge can participate much more directly in creating the solution.
That doesn’t mean every employee should suddenly deploy production systems without governance or technical review.
It means the ability to prototype, test and create is moving much closer to the people who understand the problem.
That is where I think some of the most interesting AI innovation will happen.
Prototypes and experiments
You also don’t have to build the final thing.
Imagine you have an idea for a new service.
Previously, you might create a slide deck explaining it.
Now you can potentially create something people can actually click, use and react to.
That changes conversations.
Instead of:
“Imagine if the customer could…”
you can say:
“Here. Try it.”
A prototype doesn’t have to be production-ready to be enormously valuable.
It can help you validate an idea, explain something internally, get customer feedback or discover that your brilliant idea is actually terrible before you spend six months building it.
Before you build the full product, use this app idea validation process to decide what your next experiment should test.
That’s a win too.
Apps and websites
And yes, you can build apps.
This is where “vibe coding” exploded.
The term was coined by Andrej Karpathy in 2025 and generally describes building by telling an AI coding tool what you want in natural language, letting it generate the code, running the result and iterating conversationally.
Instead of manually writing:
some complicated code I don't personally want to write
you say:
“I want this screen to show the user’s entries from the last seven days. Let them tap one to edit it.”
The AI works on the implementation.
You test the result.
Then you continue.
That doesn’t magically eliminate engineering.
Security still matters.
Architecture still matters.
Testing still matters.
Privacy still matters.
And the more complex or consequential the product becomes, the more important professional technical expertise can become.
But the barrier between having an idea and having a working version of that idea has collapsed.
That’s the part I care about.
Can you really build with AI without knowing how to code?
Yes.
But I think this question is slightly wrong.
The more interesting question is:
What skills replace some of the coding knowledge you previously needed?
Because AI doesn’t remove the need to know anything.
It changes what you need to be good at.
You need to define the problem
AI is very good at producing things.
That doesn’t mean it knows what should be produced.
If you can’t explain the problem, who has it, what needs to happen and what a good outcome looks like, AI can very efficiently build the wrong thing.
If you need a starting point, here’s how to come up with an app idea people actually want.
You need to give context
A one-line prompt can produce a one-off answer.
If you want practical ChatGPT shortcuts that still force clarity, start with ChatGPT secret codes for better answers.
Building requires context.
What are we making?
Who is it for?
What already exists?
What constraints matter?
What have we decided?
What should never change?
What does success look like?
The better AI understands the environment it’s operating in, the better collaborator it becomes.
You need judgment
This may be the most important skill of all.
AI will confidently suggest things that are unnecessary, ugly, technically questionable or simply wrong.
You need to be able to say:
No.
That’s not the problem.
That’s too complicated.
Nobody will use that.
We don’t need an agent.
This screen makes no sense.
We’re solving the wrong thing.
Go back.
AI can generate.
You still need to judge.
You need to test
“I built it with AI” does not mean “it works.”
Click everything.
Try to break it.
Give it bad input.
Check the data.
Ask what happens when something fails.
If other people will rely on what you’ve built, your responsibility increases accordingly.
Building has become easier.
Responsibility hasn’t disappeared.
Building with AI is bigger than vibe coding
I use AI to build apps, so vibe coding is obviously part of my world.
But I think focusing entirely on vibe coding misses the much bigger change.
Not everything worth building requires code.
You might build an agent inside an AI platform.
You might create a workflow connecting tools you already use.
You might build an internal research system.
You might create a reusable AI skill.
You might configure an AI workspace around a particular job.
You might create a no-code automation.
You might use an AI coding agent to build a completely custom application.
These are different implementations of the same underlying skill:
See a problem or opportunity and know how to use AI to create a solution.
That’s why I prefer the broader phrase building with AI.
ChatGPT vs Claude vs Gemini: which is best for building with AI?
Wrong question.
Or at least, it’s becoming the wrong question.
AI tools are changing too quickly to build your entire capability around one brand.
Today Claude might be better for one job.
ChatGPT might be better for another.
Gemini might make more sense when you’re working inside Google’s ecosystem.
An AI coding agent might be what you need when you’re actually modifying a codebase.
Six months from now, that mix could look completely different.
The durable skill isn’t memorising where every button lives.
It’s understanding:
What am I trying to accomplish?
What does the AI need to know?
Does this require chat, deep research, a project, a workflow, an agent or code?
Which available tool is best suited to that job?
How will I know whether the result is good?
Become good at those questions and you’re much less vulnerable to whichever company releases the shiny new thing next Tuesday.
How do you start building with AI?
Don’t start with:
“I want to build an AI agent.”
Start with:
“This annoys me.”
Seriously.
Look at your own day.
What do you repeatedly do?
What takes too long?
What information do you constantly have to find?
What are you tracking in a spreadsheet because nothing quite fits?
What subscription do you resent paying?
What do customers repeatedly ask you for?
What do you wish existed?
What idea have you dismissed because you assumed you couldn’t build it?
Those are your raw materials.
Then:
1. Define the problem
Write down what is actually happening now.
Not the solution.
The problem.
2. Define the outcome
What would be different if this were solved?
3. Ask what the smallest useful version looks like
Don’t build the empire.
Build the thing that proves the idea.
4. Decide what kind of thing it needs to be
Maybe it’s a prompt.
Maybe it’s a reusable project.
Maybe it’s a workflow.
Maybe it’s an automation.
Maybe it’s an agent.
Maybe it’s a tiny custom tool.
Maybe it’s an app.
Don’t decide before understanding the problem.
5. Build it with AI
Give AI the context.
Ask it to help you plan.
Challenge the plan.
Build the smallest version.
Test.
Fix.
Repeat.
6. Use the thing
This sounds obvious, but it’s where you discover whether you’ve actually solved anything.
A technically impressive build nobody wants to use is still a bad build.
What should you build first with AI?
Something boring.
I mean that.
Don’t start by trying to build the next Airbnb.
Build the annoying thing you did three times last week.
Build the little tool you keep wishing existed.
Build something that takes 20 minutes off a repeated task.
Build something just for yourself.
Because your first project isn’t really about the project.
You’re learning a new mental model:
I don’t have to accept the tools and processes I’m given. I can make something better.
Once that clicks, you start seeing opportunities everywhere.
And that is where things get interesting.
What “built with AI” does and doesn’t mean
I think we’re going to see Built with AI attached to an enormous number of things.
But the phrase can be misleading.
AI can generate code.
AI can design interfaces.
AI can analyse data.
AI can create images.
AI can research.
AI can debug.
AI can suggest architecture.
AI can execute multi-step tasks.
But AI didn’t wake up one morning and decide your customer needed a better onboarding experience.
Someone identified the problem.
Someone chose what mattered.
Someone directed the work.
Someone evaluated the output.
Someone decided when it was good enough.
The human role is moving.
It isn’t disappearing.
For many people, the job shifts from manually producing every piece to directing, judging and assembling the outcome.
That’s a very different kind of leverage.
The real opportunity isn’t that AI can build
It’s that more people can build.
People who understand customers but couldn’t code.
People who understand operations but couldn’t create internal tools.
People with deep expertise but no technical cofounder.
Employees who have been complaining about the same broken process for five years.
Entrepreneurs who have ideas that would never justify a six-figure development budget.
People who simply want a tiny thing for themselves.
For most of computing history, creating custom technology required access to people with specialised technical skills.
That constraint hasn’t disappeared completely.
But it has moved dramatically.
And once the cost of turning an idea into something tangible drops, you don’t just get cheaper versions of the things we already had.
You get more experimentation.
More weird little tools.
More personal software.
More prototypes.
More people solving their own problems.
More ideas tested instead of discussed.
And probably a huge amount of terrible stuff too.
That’s fine.
Lowering the cost of experimentation means more failure as well as more success.
The skill I think matters now
I don’t think everyone needs to become a programmer.
I do think far more people need to become builders.
People who can look at something and think:
Why are we doing it this way?
Could AI handle part of this?
Could I make something for this?
Does this need an automation?
Could this become a tool?
Could we test this idea today instead of discussing it for three months?
Could I build this myself?
That is a completely different relationship with technology.
You’re no longer limited to choosing from what already exists.
You can participate in creating what comes next.
And that, to me, is the genuinely exciting part of AI.
Not better prompts.
Not another chatbot.
The fact that an idea in your head has a much shorter journey to becoming something real.
Frequently asked questions about building with AI
What does building with AI mean?
Building with AI means using AI to help turn an idea, problem or process into something functional. That could include workflows, automations, agents, websites, personal tools, internal tools, prototypes or apps.
Can I build with AI without coding?
Yes. Many AI tools now let people create useful tools, workflows and even applications using natural language. More complex or production-critical systems may still require professional technical expertise.
What can a beginner build with AI?
Good beginner projects include personal tools, simple automations, research workflows, dashboards, small websites, prototypes and tools that solve a repetitive problem in your own work or life.
What is the best AI for building?
There is no single best tool. The right choice depends on what you’re building. ChatGPT, Claude, Gemini, AI coding agents and specialised building tools have different strengths.
Is building with AI the same as vibe coding?
No. Vibe coding specifically refers to using natural language and AI to generate and modify code. Building with AI is broader and also includes agents, workflows, automations, research systems, no-code tools and other AI-assisted creation.
Do I need an AI agent?
Probably not automatically. Use an agent when the system genuinely needs to decide what steps or tools to use. If the process is predictable, a workflow or simple automation can be easier to control and maintain.
Can AI build a real app?
AI can now assist with planning, interface creation, coding, debugging and iteration, allowing non-developers to create functional applications. Production applications still need appropriate testing, security, privacy and technical oversight.
How do I start building with AI?
Start with a real problem rather than an AI tool. Define the problem and desired outcome, identify the smallest useful solution, choose an appropriate approach, build it with AI, test it and iterate.
What’s the difference between using AI and building with AI?
Using AI often means requesting an output. Building with AI means creating something functional, reusable or capable of performing work.
Do I need to become a developer?
No. But you do need to develop skills in problem definition, context, judgment, testing and directing AI. The goal isn’t necessarily to become a developer. It’s to become capable of turning more of your ideas into reality.
Want to learn how to build with AI?
AI Power User is my live, tool-agnostic workshop for people who are ready to go beyond chatting with AI and start building with it.
We don’t pledge allegiance to ChatGPT, Claude, Gemini or whichever tool is fashionable that week.
We learn the underlying skill:
How to take an idea or problem and use AI to make something useful.