Inspired by Harper Reed’s blog post I want to document the way I am using AI. Specifically, large language models or LLMs.

What I do

I have to confess this first. I am unemployed at the moment, on a planned career break. So, I have all the time in the world to use any AI tool I want. I can afford to devote additional time to get familiar with any LLM-based tool. And I am very privileged to be located only 15 miles away from Cambridge, Massachusetts - which is one of the biggest AI hubs in the world.

What I use LLMs for

I use them for

• Building web applications
• Writing python scripts
• Writing blog posts
• Setting up websites
• Changing website UI
• Learning about whatever I don’t know
• Choosing which drink to order at a bar (okay I only did it once!)

The biggest use case - all things web

Building anything connected to the web requires the knowledge of multiple technologies and programming languages. I think it is impossible to keep everything I need to know in my mind. And if I need to learn every technology before I use it, I would retire before I produce anything the world can use.

So, whether it is about fixing next.js errors, or about adding database entries, or about finding out the three lines of Go that I need to add to my Hugo website, or about writing python code to turn an excel sheet into a database - I always ask Windsurf, my AI tool of choice. And I almost always get a good answer. What’s more - if the answer doesn’t work, all the AI tools can try other options. Just like a human engineer, I can iterate until I am satisfied.

Team and escalation matrix

I use AI tools like I would employ a team of humans. I usually have a free LLM for small tasks, and use a paid LLM (Claude 3.5 Sonnet) for anything that needs a little innovation. Both of these are available in my AI tool of choice. For debugging I always use the paid LLM, because debugging is twice as hard as coding. I also switch from the ‘write’ mode to the ‘chat’ mode if I am speculating about a particular approach and not sure whether to proceed or not.

I do have an ’escalation matrix’ for problems which none of the LLMs in my AI tool can handle.

First level escalation

I use perplexity and chatgpt.

Second level escalation

In case the first level doesn’t help me completely, I use claude.

Ultimate escalation

When everything fails, I consult the LLM sitting on my shoulders, a.k.a. my head. The human brain is still decades ahead of all other LLMs we know of!

Smaller use case - design

I have found that v0 is really good at design. More than once I have used it to design an UI, and then downloaded the files and incorporated the design into existing code. I am sure there are other tools out there that do this.

Another use case - learning

Just like one excellent way of learning anything is to observe someone else when they are working, I find it very helpful to ask my AI tool to do something which I want to learn. For example, I still suck at making HTML5 games, but I am trying to do so by asking LLMs to generate code for games.

Is this really worth it

It is totally worth using LLMs, even for outsourcing small tasks. My brain feels so much lighter when I don’t need to bother about the low-level details of every single thing. We should remember what a wise fictional character said long ago:

I consider that a man’s brain originally is like a little empty attic, and you have to stock it with such furniture as you choose. A fool takes in all the lumber of every sort that he comes across, so that the knowledge which might be useful to him gets crowded out, or at best is jumbled up with a lot of other things, so that he has a difficulty in laying his hands upon it. Now the skillful workman is very careful indeed as to what he takes into his brain-attic. He will have nothing but the tools which may help him in doing his work, but of these he has a large assortment, and all in the most perfect order. It is a mistake to think that this little room has elastic walls and can distend to any extent. Depend upon it, there comes a time when for any addition of knowledge, you forget something that you knew before. It is of the highest importance, therefore, not to have useless facts elbowing out the useful ones

The future

An increase in productivity, and a decrease in cognitive load are good benefits. But I think this is not where it ends .I feel that LLMs will enable us to do things which we couldn’t imagine before. It is like we have been handed an airplane, while we were only driving cars so far. Yes, it will require a different kind of training, and a different kind of map to effectively use an airplane. And not everybody will be flying planes all the time, maybe more people will ride them. But we’ll be able to hop from one continent to another in a matter of hours instead of a matter of weeks. And that means we’ll have the choice of living a better life.