Respectful Use of AI in Software Engineering

Many founders are continuing LLM and AI adoption in their companies. A lot of focus has been spent on security, compliance, and cost-efficiency. No one talks about the respectful usage of AI. And what does it even mean, the respectful usage? I'll explain.

Most of the engineering managers and C-level people in software engineering orgs are pushing towards the maximum usage of AI, because it gives efficiency. So what is efficiency? In simple terms, AI is a multiplier of human capacity at the moment, so the individual software engineer's capacity has been increased multiple times by using AI. But does that really affect the team and org performance? Probably not so much at the moment, as we have a lot of bottlenecks in the process. I'll talk about the bottlenecks in a separate article, when the time comes, but for now I want to focus more on the respect part of the issue, which mostly occurs during the code review process.

Usually more senior people are doing code reviews like they had been doing in the good old days, as it's one of the last resorts of code quality and product stability that remained after the active involvement of AI and LLMs in software development. And as AI keeps writing tons of code, senior developers are now overloaded reviewing the AI generated slop. And the worst part, where the respect comes in: most of the PR creators are not even aware of the code they are submitting. The job that they should have done, to manually review first, make sure that the quality is there, and not solely rely on senior engineers. So the task that individual should do, they just delegate to a more senior person. WTF is that, if not disrespect?

The current process looks like this:

The code review process

How it usually goesHow it should go

1

Task assignedAn engineer picks up the ticket.

2

AI writes the codeTons of code, in minutes.

✕

The author reviews it firstNot even aware of the code they are submitting.Skipped

4

Pull request openedNobody has read it yet.

5Senior reviewReviewing the AI generated slop, cold.First to read it

6

Merged

Who reads the code

1stThe senior reviewerThe first human being to read this code.—The authorNever read it.

The same applies for documentation. People keep generating documentation with AI, pretending that they are doing enormous work, but they never read the documentation they shared. So who gives a shit about what you've done, when you have never read the document yourself?

The documentation process

How it usually goesHow it should go

1

A document is neededSomebody has to write it up.

2

AI writes the documentTwelve pages, in minutes.

✕

The author reads it firstNever read the documentation they shared.Skipped

4

Shared with the teamPretending that they are doing enormous work.

5Somebody opens itPastes it into an AI and asks for a summary.A machine reads it

6

Decision madeOn a summary of a summary.

Who reads the document

1stA machineSummarized it, because nobody was going to read it.—The authorNever read what they sent.

People should spend more time manually reviewing the code, the documents, and everything that AI has produced first, making sure the quality is there, before sharing it with other humans.

If a human needs to send your document to AI saying please summarize what they want, it means you have not done your homework properly. The document should be summarized enough so that a human reader doesn't need another summarization.

Let's respect other humans in our organizations and their time.

This is a fully human-written article.

P.S. This post is 100% human-written, except that I needed a spell-checker. The diagrams were built from my sketch by a machine, which is exactly the arrangement I am arguing for: I decided what they say, a tool drew them.

---

Not part of the article

I did not measure any of this. Other people did.

96%

of developers do not fully trust that AI-generated code is functionally correct. Sonar 2026

48%

always check their AI-assisted code before committing it. Sonar 2026

38%

say reviewing AI-generated code takes more effort than reviewing a human colleague's code. 27% say it takes less. Sonar 2026

42%

of all committed code is AI-generated or AI-assisted today, which developers expect to reach 65% by 2027. Sonar 2026

8x

more code blocks with five or more duplicated lines during 2024, across 211 million lines analysed. GitClear

2024

the first year in which copy-pasted lines outnumbered refactored ones. GitClear

I did not write this part. The figures were compiled by a machine from the sources linked, after the article was finished. The article ends above the line.

The module I teach on AI adoption covers this from the other side, the engineer who still reads generated code before shipping it and starts looking slow next to everyone who doesn't. That session, and the four around it, are free to read: Rolling Out AI Without Breaking Trust.