Someone at work has been “building apps” in Base44, and showing what he has created to the SLT. it’s pretty impressive in that he’s built a rudimentary LMS and a custom “calculator” for an area of the business. I’m not surprised as he’s a clever bloke, but he’s not a developer, he’s not created any documentation or commented any code, and we don’t know how scalable these apps are.
I also have concerns about how secure the resulting apps are likely to be and how sustainable they are. This goes for other “vibe coded” apps which are built without proper experience and/or process and it suddenly hit me.
We’re back in the late 90s and 2000s where departments would have one “clever bloke” who would build something in Microsoft’s VBA, Access, or Excel which became core to how that department operated, and that person would then leave or move to a different team, alternatively something would be updated (the version of office, or underlying server technology) and it would stop working, at which point IT would be called in to try and make it work whether they liked it or not.
When I started my last job as CIO, I quickly discovered that one of the key applications for one side of the business had been written years ago by my predecessor and a student over pizzas. The software was badly running out of steam and the number of people it took to maintain it was insane. Replacing it with some “standard” off-the-shelf product was a major undertaking, not the least because the business users were used to it, didn’t want to learn anything new, but also didn’t appreciate the hidden cost of keeping that monstrosity running.
I totally see value in custom solutions but one needs to be mindful that something developed as a use it and toss it solution can quickly take on a (costly) life of its own.
Because of this ‘trend’, a lot of software companies’ stocks are badly hit over the last week. Everyone thinks that AI will replace software and all they need is that one or two clever ‘blokes’ to run the company.
I want to share this newsletter I read from Ben Thompson:
First, companies — particularly American ones — are very good at focusing on their core competency, and for most companies in the world, that isn’t software. There is a reason most companies pay other companies for software, and the most fundamental reason to do so won’t change with AI.
Second, writing the original app is just the beginning: there is maintenance, there are security patches, there are new features, there are changing standards — writing an app is a commitment to a never-ending journey — a journey, to return to point one, that has nothing to do with the company’s core competency.
Third, selling software isn’t just about selling code. There is support, there is compliance, there are integrations with other software, the list of what is actually valuable goes far beyond code. This is why companies don’t run purely open source software: they don’t want code, they want a product, with everything that entails.
Still, that doesn’t mean the code isn’t being written by AI: it’s the software companies themselves that will be the biggest beneficiaries of and users of AI for writing code. In other words, on this narrow question of AI-written code, I would contend that software companies are not losers, but rather winners: they will be able to write more code more efficiently and quickly.
There’s a lot of the software industry, especially the parts of it that depend on hype to generate vast investment, that is really only about building ever more impressive demos rather than a “work right, always” product which is dependable.
I’m pretty sure that vibe coding is more useful for the former (quickly making a working demo of a software idea) than the latter (making code as secure, reliable, efficient and robust) as it can be. But the split between the two approaches has been around for a long time and won’t disappear anytime soon.
Our experience (software company) matches Ben Thompson’s take exactly.
Business users will vibe code things they want. Leadership and IT will fight the battles Ben describes—and learn the lessons.
Some users will end up with personal app stacks (the new shadow IT), but most companies will keep paying for software development outside their core competencies.
Federico at MacStories said something that I agree with. He essentially said he’s using the Drafts app more because he’s able to know quickly create the Javascript bits he needs as connective tissue. My boss is happy with it because it removes some of the tedium of scripting and programming for various controllers (Lighting and more)
Apps that have Applescript hooks are now more enticing because AI should help novices slog through getting the script to do what is expected.
The Elephant in the room is performance. Rarely do people mention the performance of an application. Apples need somewhat universal measure of benchmarks (maybe these exist today but are just ignored). Well coded apps are a joy to use because of stability and performance.
Agree on most of that - with one additional point of note…
A really good middleground solution is to use AI coding for an initial prototype. Then when the concept is proven, the feature set is clarified, and all the stakeholders give buy-in have a professional coder either review/refactor the AI code in detail or start from scratch using the AI prototype as the plan.
This can be a win-win-win for everyone involved. The app is proven before spending the $$$ on a professional programmer, buy-in from mangement is easier when the outcome is clear, and the professional programmer saves the hassle of getting clear specs from the users and revising the app thereafter.