There is a particular way software engineers create value for the company and simplify the life of other engineers around them.

When working on a task, one can go and read the code, find high friction bits of it, optimise them so the next engineer who is going to look at it is not feeling the burden of complexity.
This behaviour is, of course, often driven by egoism and the changes are made for future self rather than the group, but the value for the group is still there.

Value for the company is there too. One engineer essentially saves other engineers from struggling more on a task, reduces the time to complete that task, which in turn, on a big scale, keeps companies moving forward faster without being overwhelmed by complexity.

This fundamental, sort of ecosystem, has been threatened recently by generative AI.

There is no incentive to optimise or remove the code because engineers don’t struggle reading it anymore, until something is significantly broken or when the fix doesn’t make sense. And something will be significantly broken. It always does. And when it happens you will need to understand the code, if not with intent to fix it yourself but with intent to prompt your agent well.

Unfortunately as of now, many bugs are getting fixed by introducing more code instead of removing it. In many cases, engineers point their agents to a bug report, say “fix it” and verify the result (at best).

Engineers who find mental capacity to explore the issue deeper and who experience more friction almost inevitably find that the bug was introduced because of irresponsible use of generative AI from one of their colleagues.

And this is very much the opposite of the dynamic I mentioned in the beginning. The egoism fueled desire to reduce friction for ourselves is now only reducing it for ourselves, thanks to agents.

And from a company perspective, more friction is added and more time is spent on a task by its engineers. LLM usage is currently hiding the slowdown by boosting engineers to overcome complexities that we have created at pre-LLM times but what else will be there to boost us against complexity we’re generating now?

I’m sure if we simply think about reducing frictions for each other instead of reducing it for ourselves, we will improve as a group of software engineers.

Reduce the amount of code in your PRs, go optimise some logic that is difficult to read, find and update documentation that is difficult to follow. Even if not for your fellow humans, do it for your agents.

By doing this we will become stronger and more connected with each other. And this will in turn help us to overcome any industry crisis, like the one we see today.