How I used AI to build a single-serving micro tool

A lot of the attention right now is around AI Agents, which promise an incredible future of autonomous efficiency – eliminating grunt work from our lives by giving AI the agency to just figure things out and get things done on our behalf.

But there’s another interesting way to think about AI that hasn’t gotten nearly as much attention – single-serving apps, or maybe better titled: micro tools.

Having spent a large sum of time in AI tools that offer prompt-based agents to build code, effectively creating digital products without the need for an engineer, I’ve gotten a good sense of how these tools work. I’ve honed my approach to crafting prompts that get to the desired results. And I’ve figured out where I might hit a wall with limited engineering skills.

For me, as someone who has worked in the digital services space for over 20 years, leading teams that design and build digital products and services, the magic of AI is in having my own personal engineer who just writes code based on what I ask for. Sometimes it gets it right. Sometimes it gets it wrong. But eventually, together, we get to the right place. I’m someone who’s plagued with limitless ideas, but limited time and resources to make things. Now, that’s all changed.

In using these tools, I’ve begun to understand how easy it might be to replace many of the SaaS services our agency uses – many of them mediocre products that don’t seem to evolve or improve. While these SaaS tools do a reasonable job in meeting expectations, they rarely do exactly what we want them to do. Using AI, I’ve begun to explore ways to break apart some of these services by rebuilding simple, bespoke products that do exactly what we need, exactly the way we want them to. Just like magic.

There are many of these agent oriented services entering the market. Some of the latest include Lovable, “Your superhuman full stack engineer,” and Bolt, “Prompt, run, edit, and deploy full-stack web and mobile apps.” I’ve chosen to spend most of my time recently exploring Replit – “Your own automated app developer.”

The incredible ease of product development that’s available right now allows us to reach further into domains of expertise that were once inaccessible without huge learning curves. Now, we can spend less time at the mercy of mid SaaS products, and more time thinking about interesting ways to create better services for ourselves. Where in the past, we might have a need, we’d search Google for a service, exhaust any free options before ultimately settling on something that has more features than we need. We’d then begrudgingly lock ourselves into an endless cycle of subscription fees for what ultimately might be a single use. We can now use a tool like Replit to build a single-serving app – a micro tool – to quickly meet that need.

What could be better than tax time? Every year I scramble to organize receipts and figure out what I need to send to my tax team. I’ve tried a number of consumer-grade web services and mobile apps to manage receipts, but it’s only ever ended in frustration. They do a lot, often too much, when all I need is some very basic functionality.

This year I turned to Replit to build the tool I wanted. With a single prompt describing the product in detail, its basic feature set and how I wanted it to work, I had a proof of concept in minutes. But like any good conversation, the first prompt was just the beginning. Over the next couple of hours, I continued to edit, refine, and adjust through prompts until I had a working product able to solve my needs.

My personal “receipt assistant” does exactly what I need:

• I can upload receipts in image or PDF format

• It scans the files using OCR and parses the data into a table format that includes business name, date, total cost, and taxes paid

• It displays a chart-based visualization of spending across categories and across time

• And it provides an export function so I can download my data in an XLSX doc

Sure, not groundbreaking innovation, and yes, ChatGPT can kind of handle this to an extent, but in a fraction of the time it might take for someone to manually transcribe receipts into a spreadsheet, I was able to create a simple, reusable tool to automate a tedious process and eliminate this grunt work. I might only use this tool once or twice per year – it is after all a micro tool – but even then, it allowed me to remove an annual task that I general loath and instead, use my time in a more meaningful way.

Like all things, experience helps. This entire process took about two hours from identifying the problem to using the working product. It required both clear direction in prompt writing to provide the agent with the right guidance, but also my own 20+ years of experience planning, designing, and building digital products and services, and knowing exactly how to write prompts according to product specifications. AI tools combined with deep expertise is the real unlock.

I expect I’ll keep finding new ways to create micro tools for myself and for my agency. The very quick sense of accomplishment these tools provide is hard to describe, and solving problems independently, without requiring the heavy lift from a team is incredibly rewarding. Perhaps this post has intrigued you to see how these AI coding services can help you create your own micro tools.

So, what will you build?