What led me to this?

Today AI seems to be everywhere in software development.  I can’t go a day without receiving an email about how AI is being added to another application, service, or platform.  As one of the co-founders of a SaaS platform that lets you create web applications, I have been keeping a very close eye on AI, and the rapid pace it is evolving and growing. We have had people asking about if/when we’re going to add AI to Squirrel365, and up until recently I have been very reluctant to jump on the AI bandwagon.  Let me explain my previous reticence, why I’m open to the idea now, and how I envisage AI and Squirrel365 could work together.

18 months ago, in a customer influence meeting, several customers and partners were asking for AI to be added to Squirrel365.  However, when I asked what they wanted the AI to do, there was no clear idea or consensus on what role AI should play.  At the time, the general experience the attendees had with AI was generating images or asking ChatGPT questions.  And whilst it was tempting to add something with AI into Squirrel365, it seemed to me that any AI solution would be a gimmick. Only adding an AI capability because we could, not because there is a customer or product need for it. Since then, we have kept our eye on the AI landscape but made no serious moves to add AI to our product.

Fast forward to today. The pace at which AI companies have continued to evolve and enhance their capabilities is staggering. We are now seeing software providers embracing AI in a much more considered and integrated way.  No more is the “Now with AI” a gimmicky tagline, we are seeing practical, realistic, and tangible benefits of AI implementations in traditional products.

Vibe Coding Magic!

There has also been the advent of AI driven coding or development platforms. These “vibe coding” platforms position themselves as democratising development.  You don’t need a developer, just ask AI to magically create your application for you. But does the reality live up to the hype?

On the surface vibe coding seems to work. Describe your application, pay the AI token cost, and you get an application out the other end.  But looking at the code generated, you regularly find recurring and fundamental issues. Based on my observations, the most common pitfalls of vibe coding include:

Scope creep by hallucination

Where features that you didn’t ask for are added.

• Superficial Correctness
  
  • Where the application and code look right on the surface but there are underlying issues.

• Overconfidence Bias
  
  • Where the application or code looks right. As such, users skip deeper checks or validating against edge cases.

• Poor code quality impacting scalability and support
  
  • Generated code often lacks consistent structure, documentation, or design patterns. What works in a demo can quickly break down when the application grows or when a new developer tries to maintain it.

• Security Vulnerabilities
  
  • AI models can unknowingly produce insecure defaults (e.g., hard-coded secrets, weak authentication, unvalidated inputs). These flaws may not be obvious until they’re exploited in production.

• Technical Debt
  
  • When many applications or components are created without human involvement. Systems can become fragile or difficult to work with.  Debugging and integration becomes harder.

I saw a good example of the pitfalls of vibe coding the other day:  The developer asked a vibe coding platform to create an inventory management system for a mobile device.  They wanted the app to show a list of products and the stock levels for each.

The platform thought about it for a while, and then magically generated what appeared to be the perfect app.

It had, however, added in the ability add / delete products and stock levels without asking (Scope creep by hallucination).  On the surface, that isn’t a problem, as the developer was probably going to ask for those features later anyway.

However, when they tested logging into the application, every username and password worked and let you in. The AI had spoofed the login flow and didn’t validate the credentials (Superficial Correctness, Overconfidence Bias).  As such, anyone would have access to view, add, update and delete products (Security Vulnerability).

So, does the reality live up to the hype?  In my mind, no, vibe coding does not make developers obsolete (yet). Instead, I can see the current capabilities of AI complementing the developer.  That said, looking at the progress of AI and the speed of innovation, I can absolutely see that situation changing soon.

So, how can AI add value to an application development platform like Squirrel365?

Well, this brings me nicely to how I envisage AI and Squirrel365 working together.

I was watching a video by  John Higgins from Flutterflow.  They have just released Dreamflow, which is an AI-assisted development environment.  They have done a really good job of explaining how this vibe coding approach can fit in with an existing development workflow.

They propose that an AI agent shouldn’t be treated like a magician, where the app is created for you out of thin air.  Instead, they say you should treat the AI agent like a junior developer.  One who is fast and smart but needs your direction and review.  Ask the AI to create something for you, but in small, manageable, testable, bite size chunks.

That definition really resonated with me. Until I saw that video, I was reluctant to consider an AI / vibe coding solution for Squirrel365, as I didn’t want the poor output from AI to reflect badly on our product.

With this agentic / junior developer paradigm, I can see how AI can fit in the Squirrel365 development workflow.  I am envisaging an agent in Squirrel that can help you build your applications.  Rather than asking it to “create me an application that does xyz”, instead ask it to:

• Add a navigation bar to the bottom of the page with 4 buttons.
  • You can then test the button bar once it’s added.
• Next you could say when I click on the first button show the canvas called “page1”, the second button shows the canvas “page2” etc
  • Again, you can then test that behaviour

These are smaller, discreet requests that are tightly defined. As such scope creep can be kept to a minimum, and each feature added can be well tested.

I believe AI shouldn’t replace developers but empower them. That’s the vision we’re taking into our R&D at Squirrel365. As we continue this journey, we’ll share updates on the community site; and we’d love for you to be part of the conversation. Tell us how you’d like to see AI play a role in building the future of applications with Squirrel365.

‍