Fluent in Machine
Why we've been speaking the wrong language to computers
I realized I'd become fluent in a language no human should have to speak: the precise, unforgiving syntax of a machine.
People communicate with each other just by talking. So why haven't we taught our computers to do the same? This is why large language models are going to revolutionize user interfaces.
Think about it: People have evolved over millions of years to communicate with voices, facial expressions, body movements. But filling out forms on a web page? That forces the sort of mental gymnastics that increases our discomfort and completely misses out on our natural communication patterns. There's something deeply unnatural about translating thoughts into dropdown menus.
Let's try a thought experiment
Picture yourself driving at 6:00 PM on the notoriously busy Interstate 285 in Atlanta. Your phone rings. Work calls with an emergency: "Produce the quote for Boeing in the next 30 minutes or the deal is off."
You have two choices. You can either fill out the web-based quoting form in the next few minutes, while somehow driving 50 miles per hour in stop-and-go, bumper-to-bumper traffic. Or…and here's the interesting part…you can simply tell an AI in human speech: "Generate a quote for Boeing based on our last three proposals, adjust for the new specifications, and send it to Sarah for review."
Surely keeping your eyes on the road is the best policy.
A real life example
Let me tell you about something I built recently. Consultants like me track our time, even when working on fixed-cost projects. It's proper to be able to produce regular reports of what you did for your clients, or as a reference for your own use. It's also good time management—knowing what you did, logging activity, improving in the future, studying your past performance.
There are many ways to do this. Some efficient, most clunky. Several of my colleagues do this in a spreadsheet, for example.
I track my time in a simple local database. Nothing fancy. Just a flat table with fields like client name, description of work, et cetera. Pretty straightforward really; just clean data that makes it easy to generate reports.
I interact with it through a program I named ct. just two letters. A good name, simple and certainly easy to type.
The old way worked like this: I'd type "ct start testclient" followed by a description in quotes, like "working on a security review." That starts a timer for my work until I type "ct stop."
Or, if I forgot to start my timer, I could type something more complex: "ct start testclient" followed by the description, then a bunch of optional parameters.
It's faster than using a form on a GUI, unless I mess up one of the arguments. Then it doesn't work. The computer, in its infinite literalness, spits out an irritating error message and makes me slow down, remember the exact syntax, and try again.
So, working with Claude Code, I made a simple enhancement: a version that can use a large language model to process natural language comments.
Now, instead of having to remember a bunch of command line switches, I can simply type: "today I worked on a security review for testclient. it took 3.5 hours."
Or even more casually: "reviewed security for testclient, 3.5 hours yesterday."
The machine now understands the wandering, imprecise way people actually communicate. It's more natural and freeform, and it appears to work. I'm testing it now on my real work. You can see the applicability. Imagine this translated to a cell phone app that lets you simply speak to it.
And that's my point. Large language models are not toys; they are an adaptation of computers into a much higher fidelity match for human communications. And that's powerful.
The Star Trek universal translator just appeared. It just so happens that the first aliens we can communicate with are machines of our own making. We just haven't quite realized it yet.
We're going to see this more and more. So if you are a product manager, CEO, head of marketing, or anyone building things people use,you should pay attention to this emergent capability.