If a software engineer needs to perform a task that can be automated, it has probably been automated a thousand times over. Outside the realm of software engineers though, there still exist countless problems that a simple script could fix.

There are tons of people performing simple, repetitive tasks who would never consider “Could a script automate this?” ¹ But, with the advent of LLMs I wonder if this is changing. These people may not understand automation, but they’ve started using ChatGPT. I don’t expect them to suddenly start vibe coding scripts, but I suspect they may be repeatedly returning to their LLM to have it to perform the same task over and over.

This is great for them, they’re getting their task done more quickly. But it’s extremely inefficient from a computation and energy standpoint. What could consume almost no power as a script is guzzling power as an LLM. Thankfully, I think this is a rare case where corporate and environmental interests will align. Compute resources are costly, after all.

My hope is that as the AI craze shifts from a frenzy to the status quo, we’ll see companies start to analyze how much money they’re burning. As they start to ask questions, perhaps they’ll launch an in-house project or onboard a 3rd-party tool to identify areas where a small investment in development could trim a large LLM bill. These could take the form of designing the LLM to flag “highly scriptable” requests as a part of its reasoning, or potentially they’ll start proxying all LLM traffic through a system with custom models to identify trends. From there, the tasks could be delegated to different engineering teams to build tooling.

Although maybe ultimately the development work won’t be worth it as XKCD references:

What do you think? Am I overestimating how commonly ChatGPT is being used to solve simple problems? Or am I too optimistic about companies being willing to redirect LLM budgets towards building traditional tools? Feel free to join the discussions on:

• Bluesky
• HackerNews