Do You Still Need a Developer to Build Your MVP?

How far a business owner can get alone with AI, and when it’s time to bring in a developer, a tech lead or a CTO.

By Matt Ainsworth, founder of The Lean CPTO, with Kale, fractional CTO · Last updated 3 October 2026 · 72-minute webinar + 11-minute read

Webinar replay. Part 2 of the series, after How much should an MVP cost? Prefer to read? The key ideas are below, with chapter links at the end.

Why do business owners get stuck?

Over the past 10 years, in Japan and in Australia, Matt has seen the same pattern. Business owners struggle to find a technical co-founder. Some give away up to half their company to someone who later walks away, sometimes ending in a legal fight. Others pay an agency $150K to $300K or more for a first version, in a few cases after refinancing their mortgage to do it.

AI tools have changed how far you can get on your own. The question is how far, and when to stop and get help.

What are the 4 most common mistakes?

Kale adds a fifth: assuming you need technology at all. You can get a long way without it, then bring in tech to scale what already works.

What makes a good MVP?

“Good enough” is decided by your customer, not by you. Find a few champions, give it to them and ask them to tell you your baby’s ugly. If they say it’s clunky, fix that and show them again.

Two things make an ugly product work early on:

• Put a person around it. If you’re helping clients through it, they’ll put up with rough edges because they’re getting an expert, not just software. If a stranger in another state uses it without ever talking to you, it needs more polish.
• Solve a big problem. If you’re saving a client a million dollars, they’ll put up with three extra clicks. Solve a small problem and people nitpick everything.

Start with one workflow

Your first MVP needs one primary workflow, in three steps: the user signs in, completes one action and gets one result. Solving one problem well is enough to get customers to tell you what the second problem is.

Kale goes further. Can you do it without software at all? He calls it the Mechanical Turk, after the famous chess-playing “robot” that was really a person hidden underneath. The customer sees a simple front end, and you do the work by hand behind it.

His reasoning: software exists to scale a process. If your product is valuable, you should be able to deliver the same result manually, even if it isn’t profitable yet. Do it by hand once and you’ll understand the process well enough to brief someone to code it. “Go be a service provider first. Validate they’ll pay you, then build from there.”

What did AI change for business owners?

You no longer need to express logic. You can describe what you want it to feel like, and AI tools make it real. Older drag-and-drop tools like Bubble or WordPress made you fit your problem into their shape. Something that used to take three months of learning and building on a low-code tool can now be built over a weekend.

Kale sees the biggest opportunity in the middle. Huge problems are still solved by teams over years, and tiny ones by simple low-code tools. The problems in between, worth $20K but costing $100K to build, are now worth solving.

What hasn’t AI changed?

AI is very good at doing what it’s told. It’s not good at what nobody told it. It can build you an app and leave the key to the front door hanging outside, giving anyone access to your database or payment system. Security, architecture and scalability still need human judgement.

Be careful treating any AI assistant as your CTO, too. They’re built to keep you talking, which means they’ll keep encouraging you to build more.

Tech debt, explained like a loan

When you start a business you might borrow $100K knowing you’ll pay back $150K, because it gets you off the ground. Technical debt works the same way: shortcuts you choose so you can build faster. That’s intentional debt, and it’s fine.

The dangerous kind is like spending on a credit card without knowing you have to pay it back. AI tools take shortcuts without realising it. The day AI saved you today can turn into a week of undoing in three or four months. A person can only make so many mistakes per minute. With AI you can make ten or a hundred times more, which is why you want an expert reviewing the output.

Can you move a Lovable app to AWS later?

Yes, with some work. Newer AI builders like Lovable mostly use tools you can move, such as Supabase, which is open source. Older low-code platforms tend to lock you in.

Kale’s advice is not to worry too much. Your prototype and MVP can be a bit locked in, because their job is to test the market. Once you’ve proven demand, you step back and redesign it properly. As he puts it, code is a liability, not an asset: every line has to be maintained, hosted and secured. What you learn, and your ability to explain your vision to customers and engineers, is worth far more than the code.

A working prototype is also the best brief a developer can get. It shows them what you want, how it looks and which features matter, far better than a document or a handover from a developer who’s left.

The 4 levels of building

Work out which level you’re at. It tells you who you actually need.

Level 1. AI builders such as Lovable, Vercel’s v0 or Replit let you show someone what you mean, the way a designer’s clickable prototype would. Most people can get here without help.

Level 2. Vibe coding can still get you far, and some people earn real money from Lovable apps. If you’re hunting for a CTO at this stage, it’s probably overkill. A freelancer who can build it in two or three weeks, an offshore agency, or a developer you’re confident managing are all options. Founder communities and startup Facebook and Slack groups are a good place to find one, and to get a second opinion on a quote.

Level 3. This is where your prototype becomes properly built software. AI-assisted coding still speeds things up, but the job now is reliability and security. Kale’s test: you have more than a handful of customers who have you on speed dial, and they expect it to be up 29 days out of 30.

Level 4. A CTO handles the second-order problems. The first-order ones are uptime, cost, security and compliance. The second-order ones are hiring, and what happens if your lead engineer leaves. The question becomes how to hand this to someone new without it falling apart.

Tech lead or CTO?

It depends on where your innovation is. If you’re innovating in how technology is applied to your industry, like using AI models to solve a problem in your sector, a hands-on tech lead is usually enough. If you’re inventing new technology, you need a CTO who can solve problems nobody has solved before. Job titles in tech are loose, so judge the person, not the title.

What can you build alone with AI?

• Custom chat apps: your expertise wrapped in an AI model with its own interface, sold as scaled consulting.
• Simple automations: a form submission that emails you, then sends the customer a qualifying questionnaire.
• Anything you’re happy to throw away. If the code needs to last, think about a developer.

The rule of thumb: AI works best early, while you’re still learning what customers want.

Where do AI coding tools hit their limit?

AI never pushes back. Ask it to make the button vomit yellow and it will. It also solves problems the quickest way, not the best way. Kale’s analogy: there’s a leak in the wall, and AI puts a bucket under it.

Those shortcuts pile up. Kale sees the same pattern with clients. Months one and two go fast. Month three slows down. In month four they call him: “every time I change something, everything breaks.” A good developer works more like a craftsman, who knows when to take code away rather than add more.

Isn’t it just about better prompts?

Partly. Kale uses AI coding tools all the time, but his prompts are precise engineering instructions: add this table, model it against this data, update the login to use it. At that point you’re doing an engineer’s job. Telling an AI “you’re an expert developer” helps, but it’s still predicting the next word, not judging the design.

He compares it to a lawyer. You can get AI to write a contract in five minutes. What you’re paying a lawyer for is what it forgot to include. In one case, Kale asked an AI to add memory to a chat app. It solved the problem by adding a second AI provider, which meant another dependency, another cost and another thing to break. It worked, but it wasn’t what he wanted.

How do you hire a good developer?

• They use AI coding tools. Without them, an MVP might take three or four months instead of about one.
• They talk in business terms. A bad developer says yes even when things are unclear, like an AI given a bad prompt.
• They’re creative. They ask why you’re solving a problem at all. In software, a problem framed the wrong way can take months, and framed the right way can take two days.
• They’re curious and happy to work outside a narrow role.

Offshore or onshore? With an offshore team, expect more of an adjustment to working across cultures. Set expectations clearly up front and keep communication open both ways.

The tools to run a dev team

Don’t overdo the tools. Jira or Notion for tasks, Slack or the free, open-source Mattermost for chat, Google Meet for check-ins, GitHub for the code. Kale has run engineering teams of 5 to 10 people on Notion, one group chat and personal to-do lists. Use whatever comes naturally to you and your team.

Hold standups two or three times a week, at least three if they’re full-time. Ask agencies for a weekly demo. Run short retrospectives on what worked. Across time zones, use written standups in a shared chat: what I did yesterday, what I’m doing today, where I’m stuck.

Which AI tools do we recommend?

If you have a developer: Claude, Codex and Gemini. They’re powerful, but they show you how the sausage is made: you need to read the code and deploy it.

If you’re building yourself: Lovable, Bolt or Replit.

For automation: Zapier, Make or n8n.

Kale’s hot take: email and spreadsheets. You can run a service as the first version of your tech product. For one large client he’s starting two projects where the first quarter is simply: email this engineer and they’ll do the task by hand. Then they build what’s proven. DoorDash started as a simple page with the founders’ phone number and flyers around campus. Levels.fyi, the tech salary site, ran on a simple website connected to Google Sheets.

AI lets business owners go further than ever. It also lets you make bigger mistakes faster. That’s why direction matters more, not less.

When is a fractional CTO worth it?

Kale’s answer: once you have product-market fit. When founders ask him to work with their startup, he asks whether they’ve talked to customers and whether anyone is paying them to do it by hand. If not, they should do that first. You don’t want to pay anyone to build in the wrong direction.

The exception is day one. Some clients pay a fractional CTO for half a day to design the system and tell them honestly: this will cost $100K, or this needs skills that don’t exist yet. In between, a fractional CTO is a second opinion, like seeing another doctor. They can review a dev quote, help you hire, check an offshore contract or sharpen the product direction.

And AI agents? Probably not anytime soon for most early products. It depends on what you’re building.

Moving from an AI prototype to a developer

The hardest part is scaling beyond yourself. Developers speak a different language, and much of their work is invisible: you see 20% of the output and none of the 80% underneath. Kale often gets asked to review a team that “doesn’t do anything” and finds they’re fixing hard bugs nobody can see, like a server that crashes randomly once a month.

People coming from AI tools often ask why developers don’t produce code as fast as their AI did. The answer is that the developers are solving the problems the AI skipped. Regular updates, clear communication and a monthly review from someone you trust make that gap manageable. Expect a period of adjustment, like any new partnership.

Who really owns your AI-built app?

Frequently asked questions

Do I need a developer to build my MVP?

Not at first. A business owner can now build a working prototype alone with AI tools like Lovable, Bolt or Replit, or run the first version by hand. Bring in a developer once paying customers depend on it, often for the last 20%: payments, integrations and security.

Can I build my MVP with AI alone?

Yes, if you’re innovating in how technology is applied to your industry rather than inventing new technology, and the code is something you’d be happy to throw away. Keep it to one workflow, and get an engineer to review anything that handles customer data or payments.

When do I need a CTO?

Usually much later than most owners think. Early on a developer is often all you need. A tech lead comes in when customers rely on you and you need reliability and security. A CTO matters when you’re scaling the team and the product, or when the technology itself is the innovation.

When is a fractional CTO worth paying for?

At two points: on day one, for half a day to design the system and tell you what a build would really cost, and once you have product-market fit and need to scale. In between, they work well as a second opinion on a quote, a hire or a code review.

What is technical debt?

It works like financial debt. Intentional debt is a shortcut you choose so you can move faster. Unintentional debt is what AI tools create when they take shortcuts without telling you: the day you saved now can cost you a week of undoing in three or four months.

Related guides

• How much should an MVP cost in 2026?
• Build. Delay. Kill. How to decide what goes in your MVP
• The Manual MVP: get paid before you write code
• Stuck at 80%: what to do when your AI-built app stops working
• Is your AI-built app safe to sell?

Video chapters

• 0:00 Do you still need a developer to build your MVP?
• 1:18 Why non-technical founders get stuck
• 3:03 Meet Kale, fractional CTO
• 6:46 4 mistakes non-technical founders make
• 8:41 What is a good MVP?
• 14:21 An AI-built MVP that raised
• 15:24 What AI made possible for founders
• 17:53 What AI hasn’t changed: security and architecture
• 19:33 Tech debt explained like financial debt
• 23:44 Moving from Lovable to AWS (and why code is a liability)
• 27:09 Sensitive data and security
• 28:15 Start with one workflow
• 29:34 The Mechanical Turk: a manual back end
• 30:06 The 4 levels of building
• 30:50 Level 1: AI prototype on your own
• 33:00 Level 2: first real product and first devs
• 35:36 Level 3: bringing in a tech lead
• 37:14 Tech lead vs CTO
• 39:52 Level 4: CTO and scaling
• 43:17 What can you build alone with AI?
• 45:57 Where AI tools hit their limit
• 48:13 Is good prompting enough?
• 52:50 A fractional CTO as a second opinion
• 53:42 The risks of going it alone
• 54:22 Hiring developers
• 56:56 Offshore vs onshore developers
• 58:10 Tools to manage a dev team
• 1:00:40 Standups and async updates
• 1:01:56 AI coding tools we recommend
• 1:02:47 Email and spreadsheets as your first product
• 1:04:43 Zapier, Make and n8n
• 1:06:11 When is a fractional CTO worth it?
• 1:07:57 When to start using AI agents
• 1:08:31 From AI prototype to working with a developer