There is nowadays lots of talk about “software factories”, AI agents, more AI agents, “harnesses”, “agentic engineering”, and more AI stuff.
However, there’s an issue: I would like to focus on reducing production debugging (basically solving ops’ problems) with Ferron, not on some AI stuff! And target audience of such software doesn’t care much about whether AI was used or anything complex. In fact, the very technical audience the Ferron would have would dislike this and discard this as “AI slop”.
Why would I do “software factories”? Also, would I direct mass-producing programs? No, I want to craft a working, easily debuggable web server (one program I care a lot about) that solves one painful problem (debugging friction in my case) really well. And people may see mass-produced programs as (AI?) slop.
Also, AI tends to write rough code, and also proceeds with drafting plans that involve utter nonsense, no matter if those using software would actually care about it or if it solves a painful problem for them. So I wouldn’t trust AI to fully “produce” software as it would be in a “software factory”. I would still need to QA software myself. Though AI can also QA software, but some things are lost in the context window, and then comes rough code. So I need to be involved to do QA, and even I might miss some things, so those testing the software can also do the QA.
And then, projects where software development is completely automated with AI agents. Where’s the battle-testing? Also, security issues that result from AI cargo-culting the training data… And then AI breaking things the directors didn’t intend to (there’s even a website that makes fun of that). And then skill erosion, and then who is going to create and train AI models?
And then, the code fails in production, but how could they debug? They often don’t understand the AI-written code, and AI would get also lost in context. So nobody in the team of AI would understand what’s going on, and debugging becomes harder. As a result, the project would have issues for longer, and people would just leave.
So for all of you, sink the unvetted AI slop!
P.S.: not anti-AI dogmatist, just against mindlessly accepting AI output without review that would let slop slip in.