Sooooo much FUD (I think the term has fallen out of favor, so: Fear, Uncertainty, Doubt) being spread these days about AI-assisted software development. I see a lot of posts on here of engineers claiming that AI can’t do what they can do. I’m sorry — it can. And it can do more of it and faster than you ever could. I’m sorry that you love curly braces and semicolons. I’m sorry that you love a finely crafted list comprehension or recursive function. Those skills are no longer necessary. I’m sure there were many engineers in the past that loved assembly and loved moving data between registers to squeeze every microsecond of performance gains they could. Compilers ended up being better than most and good enough for just about every problem. AI engineering is the next (and bigger) revolution that’s doing the same thing. And it’s time for everyone to adapt.
No, you cannot expect a 1-paragraph prompt to produce a well-functioning and scalable application. You still need software architecture and design skills and good understanding of the functional requirements of the system you are trying to build. You need planning and design documents (written in cooperation with an agent). You need tests, tests, and more tests. You need… backpressure.
I’m sure this terminology is not new, but I read this article and I love the analogy.
When AI wants to take us off the rails, we constantly apply this backpressure through prompts, skills, agents and tests that push back on the AI’s natural ignorance about our project and our way of doing things. But to think you can tell an AI to just “do what I would do” is not recognizing how software development is changing. HAS CHANGED
A senior engineer / architect has a new role now. Your job is to carefully craft, maintain, and utilize all of these forms of backpressure for your project. Your other job is to mentor the junior engineers around you. Teach them why you wrote what you wrote in the skills and prompts. Mentorship is more important than ever now. It will take a few years for universities to transition how they train software engineers to focus more on systems design and architecture and less on algorithms, languages and syntax (I said less, not zero).
PR reviews is a topic for another post, but I think that is also changing. Code is now written by AI, not to make it easier for humans to read, but to make it easier for AI to read. Large PRs are not necessarily a bad thing anymore — they are not necessarily a sign of scope creep or lack of discipline. They might just be an artifact of the new world order of software engineering. You can do more now in less time. And you can guard these large changes with backpressure.
Everything about software engineering is changing and its time to start adapting. The engineers that are not afraid to push the boundaries and challenge old notions of what makes good software are the success stories of tomorrow.