Claude, GPT, and other LLMs for Filework

The major provider LLMs, such as Claude, GPT, and Gemini, have become increasingly good at handling all sorts of filework.

Claude in particular, with its recent release of Cowork, is great at working with files on your device, even writing ad-hoc scripts to automate certain tasks in a similar way to Gruntless apps.

And behind the scenes, Gruntless uses these LLMs to write the code for your apps - so why not go directly to the source? Fantastic question, I'm glad you asked.

Doing Filework with Claude

In the past I've had Claude handle some files for me. For example, I had Claude extract some data from a PDF into an excel file that already had other data in it. This is how that flow looked:

• Upload both files to Claude (these were not sensitive files, so I felt okay with this)
• Describe what data I want extracted and how
• Claude reads the files, thinks, writes some code, thinks some more, seems to have a long discussion with itself
• Claude presents the results, and they're a little off (it confused two columns with similar names)
• I changed the model to Opus, tried again
• This time it worked
• I wanted to grab another PDF to extract to the same excel - but my usage was up for that session
• A couple of hours later, tried again
• Claude made a small mistake
• I prompted it to fix it
• Done

All in all a pretty standard experience working with LLMs.

Is This So Bad?

In my mind, once something works there is no reason it shouldn't work again. But therein lies (one of) the problem with LLMs - they are by nature generative, that is - their output is by definition probabilistic.

And while the probability of success is generally quite good, it is not 100%.

On top of that, the LLM has to generate its response each time you ask it to do work for you - even if it did the exact same work just a moment ago. This can be maddening when you have recurring work to complete.

And of course, even if you do find the perfect prompt that works near-perfectly near-most of the time, you run it once, and then when you have to generate the same report next week or next month... now you're chasing an old chat thread and good luck with that.

So... Gruntless?

This is where Gruntless comes in. With Gruntless, you get an app that:

• Works 100% of the time, the exact same way every single time
• Gets your work done virtually instantly (depending on file size)
• Never abruptly ends your usage (you can run your app once a month or ten thousand times a day, it doesn't matter)
• Is organized neatly in your Grunt dashboard

And, of course, your file data never goes anywhere, as the apps run entirely in your browser. You also don't have an LLM snooping around your personal device (as with Claude Cowork) - and that level of privacy is always nice =]