Working outside the bubble
I am fully aware that I live in a bubble. My wife and I both use AI heavily. The tech people I interact with are all in on AI, and my X feed is curated to show the most enthusiastic people about this amazing technology.
But the truth is, we are very early, and the vast majority of people have no idea what AI can do. That’s also true for countless small and medium businesses. And you know what? They don’t need AI.
I’d dare say many small and medium businesses are so far behind that they need simple tools and automation. Heck, some of them just need a few spreadsheets with formulas and a Google Form.
What they need is someone who can go in and review their business and their processes, spot the opportunities for improvement in both the processes and the tools and then use AI to actually implement those improvements.
The most important thing to keep in mind is that AI is non-deterministic and much more expensive than simple deterministic automation. Using AI to design and implement a tool that will cost virtually nothing to run forever is the right move.
There is a difference between software built with AI and software that needs AI to work.
An AI agent can help write a script, build a form, or connect two systems. Once that work is reviewed and tested, the result can run as ordinary software. No model needs to wake up every morning and decide whether an overdue request deserves a reminder.
Imagine a small repair business. Requests arrive by email and phone. Someone copies the details into a spreadsheet. Someone else asks whether the customer approved the quote. The owner keeps checking which jobs are waiting for parts.
It would be tempting to put an AI assistant in the middle of all that.
But I would start somewhere less exciting. A form to capture the request. A shared record with a status and a person responsible for it. A reminder when a quote has been waiting too long. A view showing jobs blocked by missing parts.
AI could help build all of those things. None of them necessarily needs a model to run.
The difficult part is deciding what those statuses mean. Does approved mean the customer said yes, or that a deposit arrived? Who updates the record when a part comes in? Should the reminder go to the customer, the technician, or the owner?
If nobody agrees on those answers, the automation will faithfully produce confusion. Faster confusion, perhaps, but still confusion.
That is where the useful work starts. Sit with the people doing the job. Follow a request from arrival to completion. Find the information they keep asking each other for. Remove a pointless handoff before building a clever way to manage it.
I wrote in Made to Measure about AI making small, specific tools worth building. This is one way that becomes useful outside our bubble: software fitted to a real workflow, rather than a business reorganized around another subscription.
Of course, building it is only the beginning and someone still needs to own the result.
Forms change. Email delivery fails. An integration stops working. The person who created the spreadsheet leaves, and everyone discovers that the reminder script runs under their account.
A small tool needs sensible access controls, recoverable data, and a way to notice when it stops doing its job. Someone needs to know how to fix it, or who to call. AI can help with that work too. It does not make the responsibility disappear.
Now, once the basics are done, then we can start putting AI inside the finished tool.
Back to our imaginary repair business: customers might send rambling descriptions, photographs, or documents in different formats. A model could help turn that material into a draft job record, summarize the history, or flag missing information for someone to check.
That is a different problem from checking whether an approval box is ticked.
Use the model where interpretation earns its place. Use explicit rules where the decision is already defined. For consequential actions, decide what a person must review and what happens when the model cannot give a useful answer.
I am enthusiastic about AI precisely because it gives us more ways to solve these problems. Sometimes the best result is a tool whose users never need to think about AI at all.
The business does not need to catch up with my feed. It needs to stop losing track of the work.
If your team is held together by copy-paste, reminders, and one person’s memory, I can help map the workflow, simplify it, and build tools someone can actually own. Here’s what I offer.