← notes
$ cat notes/how-i-plan-to-survive-ai-era.md

How I plan to survive AI era

/ 3 min read /
#ai#career#personal

While a lot of digital crafts are entering the “AI” era - I’m a part of one of them. Studied CS, then started to work as a software engineer since 2011 and have been working on many different things. And while LLMs are changing the way we work, I am jumping between states: “ai tools will replace me” and “ai tools won’t replace me”. I think or I hope it will result in something in between these 2 extremes. LLMs are a powerful tool that will improve some parts of my work, eliminate some of them fully and also bring new problems. But some parts of software engineering won’t go away and maybe will become even harder. Things like: debugging complex systems, having context about the codebase and product, wiring things together, supporting code quality for long-term maintenance.

I keep thinking: how should I adapt to these changes? I have no plans to move away from software development yet, but at the same time I should adapt: offload things to LLMs that eat up time, but improve the stuff that makes me a better engineer and part of the software development community. It’s not actually only about software engineering, it could be applied in different areas of my life.

What I decided for myself is to follow this separation:

  • Everything that has long-term implications and is important - I should learn it myself

  • Everything repetitive and time consuming - I should offload that to LLMs

Examples:

  • I keep track of my personal expenses, but I don’t use what banking apps provide as they do not show categories in a way I actually need. E.g., I buy many different things on Amazon and for my banking app all transactions are the same. But it could be toys for my kid, food for my cat, AirPods, you get the idea. I also use several different banks, which makes accounting a real task. I used to fill in all the transactions manually with respective categories, but it takes a lot of time. Now, with Claude, I made a script that parses PDF statements from my banks and prepares them in a way I need to import into the app I use for personal finance. I don’t want AI tools to process these PDFs in the cloud, so I built a tool that can do that without a cloud LLM. There is a small processing step using a local LLM via LM Studio.

  • Software engineering. Some tasks are time consuming as well, but straightforward. Most developers know that writing tests can take even more time than feature development. Now with LLMs it has become much easier to generate tests specifically, which saves a lot of time. At the same time I would like to think more about tasks itself. How to implement them, how to build the whole system.

With “ai tools” I am trying to outsource things to get myself a bit more time to do work that requires deeper thinking. I also mix building prototypes with “ai tools” and old-school think-and-type programming to train my brain.

I am honestly too afraid to outsource most of software development to “ai tools”. The brain and any skill are not much different from training muscles: you stop training - you lose muscles. Same with the brain.

And I don’t want to be fully dependent on a tool that instead of helping me will make me stupid.

$ echo "test"