AI Can Code a Quiz. But Can It Build an Order Management System?
Howie ·
Over the weekend, I did something that would have taken a team of developers a couple of weeks to achieve just a few years ago. I "vibe coded" a fully functioning, highly polished web app quiz.
If you want to try it out (and see if you can beat it), you can give it a spin here: Slot or Slop Quiz.
I prompted an LLM (Large Language Model) with a rough description of how the quiz should work, what it should look like, and my preferred tech stack. From there, the AI fetched real quotes, wrote fictitious ones in the perfect tone of voice, built the logic, and designed the layout. Unprompted, it even baked in web accessibility and baseline SEO. A few prompts later, we had Open Graph sharing, analytics tracking, and a deployment plan.
It got me 95% of the way there in a fraction of the time.
As someone who has been building software for over 25 years, it made me stop and think: If AI can build an entire application from a few prompts, where does that leave the traditional software development agency?
The Reality of Modern Development
Let’s be honest: about 90% of pure coding can be repetitive. Writing boilerplate code, or configuring standard APIs isn't where the magic happens. I use LLMs every single day to handle that heavy lifting. They are faster than me, they don't get tired, and they don't complain about doing the boring stuff.
But there is a massive gulf between "vibe coding" a short-lived promotional quiz and building software that runs a business.
The quiz was a fun, isolated ecosystem. If it crashed, nobody lost revenue. If the data structure wasn't perfectly optimized, it didn't disrupt a supply chain.
When we talk about bespoke business systems - like connecting a warehouse stock system to an e-commerce platform, integrating real-time chat into a medical API, or building automated quotation tools - the code itself is not the hard(est) part.
What the AI Doesn't Do
An LLM is an execution tool. It is an incredibly powerful assistant, but it operates entirely without context. What it cannot do is:
Conduct Project Discovery: It won’t sit down with your operations team to figure out exactly why a manual spreadsheet process is costing you 10 hours a week, or uncover the hidden bottlenecks in your workflow.
Push Back on Scope Creep: It will build exactly what you ask it to build, even if what you are asking for is a terrible idea that will break your business processes or overcomplicate the user journey.
Understand Human Relationships: It doesn’t negotiate, prioritise features based on your immediate ROI, or adapt dynamically to shifting business goals.
The "human in the loop" isn't a bottleneck anymore; they are the filter, the architect, and the strategist.
The New Choice for Businesses
The democratisation of coding leaves businesses at an interesting crossroads. If you just need a quick, short-lived digital promotion, you might not need a traditional full-service agency anymore. With a bit of patience, you might even try to prompt it yourself.
But if you are building software to streamline your business, eliminate manual mistakes, and scale your operations, your requirements haven't changed - they’ve evolved.
You don’t need to pay an agency to spend weeks manually typing out boilerplate code. You need a senior partner who understands business architecture, who can direct these AI tools with precision, and who knows how to make disparate systems talk to one another securely and reliably.
AI has made the execution of code incredibly fast. But it has made the strategy, discovery, and architecture behind that code more critical than ever.
We aren't worried about our careers at How. In fact, we're excited. Because when the boring stuff gets automated, we get to spend 100% of our time focusing on what we actually love: solving real business problems for our clients.
Want to talk about how we use modern workflows to build smarter, faster business integrations? Get in touch.