Will AI enable custom software for the non-engineer?

A friend of mine proposed a question today.
“If I spend three hours a day doing similar routine work, will an AI coder be able to pick up any repetitive patterns and streamline the process for me specifically?”

It’s a fantastic thing to picture - akin to a lightning-fast, software developer watching over your shoulder and tailoring the software just for you.

More than code

For the average person I do not think this is where we are headed - at least, not exactly. AI coding has reached a capability level that frankly, beggars belief. But the software that you use on a daily basis; whether G-Suite, Notion, your companies third party CRM, is a lot more than just the written code. It’s weeks and years (and sometimes decades) of careful planning, designing, and deliberation; compromise, trade-offs, and solution; integration & regression testing, bug management and backlog management; and carefully managed deployment via highly-managed infrastructure. To expect an GPT agent to observe a workflow and develop a ‘working’ solution is one thing. To expect it to safely & autonomously implement across the entire software development lifecycle would require more than just faith.

This is why I believe that in 10, 20, and 30 years time - there will still be software development teams and software engineers working away. They will certainly be coding less (as they already are today!) but they are experts in their field for a reason - through accumulating years of experience on how software needs to be developed in sensible, safe and manageable way.

Prototypes

There is an alternative way forward though, for my friend, that can allow them to have what they dream of - highly custom, bespoke software personally created for their use case. And while today they may need some software-development chops to get it working, it’s several orders of magnitude easier than it was just a year or two ago.

I’m talking of course about prototyping and mini-app creation. The flagship GPTs have all got to the stage where you can provide a simple list of requirements, and have an web-app created in mere minutes bug-free with a beautiful UI.

The trickier bit comes next, but is more achievable than ever. A local development environment needs to be created (probably node), with a long-term storage solution (probably sqlite or a filesystem). Then presuming the requirement integration with their daily software, the software will need an open API, SDK or MCP server, and for the necessary feature set to be exposed to these APIs. And then, given a bit of luck, they will have what they’re after - a small slice of bespoke, personal software to improve their workflow.

This is one of a few opportunities that the average white-collar worker can take advantage of improve their workflow through AI today. Exploring these opportunities is exactly what I’m aiming to do on this blog with future posts. I’m also excited to deep-dive into this example to provide a guide on how you can create a useful, working prototype to enhance your personal workflow.

It suddenly gets trickier again once you add more requirements. What if you wanted to add another feature? Okay, fair enough. What if the API changes? Tricky, but just about manageable with prompt-coding. What if you want to be able to use this workflow on more than one device? Okay, now you’re talking deployment and have security concerns. What if heaven forbid, you want a team member to access the same software? Now, that’s a true tin of worms for a non-engineer.

This is where the software engineering team once more rears its ugly head and presents its raison d'être. They’ll be enormously grateful for your proof-of-concept, and thanks to AI, be able to integrate it in your beloved software better and faster than previously possible - and your colleagues and peers will benefit from it, too.

New paradigm

A new breed of software may emerge that leans into this idea. It’s certainly an exciting enough concept that is worthy of an experiment. As long as relevant APIs are made available, new interfaces can be generated to take advantage of them within in-house software. Frontend can be served up custom on a user-by-user basis with much less security and safety concern than any backend code.

The concept still needs to be approached with caution. If not given proper constraints and guardrails, It could create weird and wonderful workflows that are impossible to handover for a holiday, create buildups of data over months with poor integrity or failing regulatory compliance, or need a specialist to unravel issues when it goes wrong.

Perhaps the sweet spot lies somewhere in between - a curated, software-native marketplace of user-generated workflows and automations, vetted and approved by the engineering team before deployment. This would allow for the creativity and personal optimisation that individual users crave, while maintaining the guardrails and quality standards that enterprise software demands. It's a model we're already seeing emerge in platforms like Notion with their template galleries, and Zapier with their workflow sharing. If software engineering teams are brave enough to build this into their software, AI could supercharge this concept exponentially.