AI-generated summary of Senior Engineers Are Using AI Wrong and Software Factories Are Dumb, Joe Maddalone. Not written or verified by the author.

Takeaways: Senior Engineers Are Using AI Wrong and Software Factories Are Dumb

2026-11-05 · Source: https://www.youtube.com/watch?v=4Oq_gdhjYUw · Joe Maddalone

Gist

A short essay video. Many software best practices exist because humans wrote, read, and maintained code by hand. AI changes those costs, so some of those practices should stop making sense. Yet vibe-coded apps still follow them. His read: we haven't dropped the old mental models, him included.

He applies that to three targets. Craft habits like DRY and hand-polished abstractions. Folder hierarchies. And the "software factory," which he calls a 19th-century industrial model applied to a cognitive shift.

Opinion piece. No data, no demo. Every claim below is his, not checked here.

Takeaways

• Metaphors become habits. The desktop, the file cabinet, the floppy-disk save icon. A metaphor shapes a mental model, the model shapes habits, and a wrong model gives wrong habits. Over time people defend 1970s and 1980s design decisions "as though they were laws of physics."
• Practices encode old costs. DRY and clean-code craft made sense when every character was typed by hand and every bug meant a human reading that code. If AI can rewrite or throw away 500 lines of boilerplate in seconds, the cost of duplication and abstraction changes. Hours spent polishing a class hierarchy "isn't craft. It's nostalgia."
• Woodworker to urban planner. The planner ignores the grain of the park bench and manages traffic flow and capacity: set boundaries, enforce constraints, direct growth. An engineer's value moves from clean syntax to "express clear intent, to evaluate tradeoffs, and to guide growth." He says the current term for this is "design engineering."
• Folders are a spatial crutch. LLMs and vector search find code by meaning, not by its path. Engineers who say they don't read code still have AI organize it into folders, which he calls a 1985 fix for human short-term memory. The 1985 date is his, unverified.
• The senior-engineer trap. The loud claim is that AI only helps seniors, who can catch its mistakes. He says seniors use it as "hyper-efficient autocomplete for their own mental models" and ship 2008-era patterns faster. The next era comes from people with no attachment to DRY or old refactoring dogma, who have nothing to unlearn. Asserted, no evidence offered.
• Factory: half right. CI/CD already works like an assembly line, and he's fine with that. The trap is turning the whole SDLC into one: a PM drops in a spec, agents output a product. That only makes sense if a developer's edge is speed. He says the edge is "taste and verification and edge case evaluation."

Rob West's four objections, as he paraphrases them

He cites The Agentic Software Factory is the Wrong Metaphor, Rob West (original) by Rob West, published 19 Sept 2026.

1. A factory assumes design is solved. Deciding what to build and how it behaves is the work. "A ticket is not a blueprint."
2. Building changes your understanding of the spec. "It's not spec in and software out."
3. An assembly line removes variation. Design benefits from it, since variation is how you compare and find better approaches.
4. A factory must be fed, which pushes teams to formalize requirements before they understand the problem. The ambiguity stays, and agents fill it with guesses and "very confidently produce the wrong thing."

Framing examples, his claims

• Marvin Minsky doubted neural networks and "pretty much single-handedly caused" the AI winter. Unverified, and "single-handedly" is his word.
• Refactoring (Fowler) is a recipe book. The Phoenix Project and The Pragmatic Programmer teach principles through parables.
• Horsepower, the desktop, and the floppy-disk save icon all borrow from older things. Some adults today know the save icon without ever seeing a floppy disk.

Patterns

• a teaching metaphor outlives the thing it described ← "the metaphor is outlived the thing"
• rules built around old costs expire when those costs change ← "changing the economics of software development"
• experts use a new tool to speed up their old method ← "hyper-efficient autocomplete for their own mental models"
• people with nothing to unlearn reinvent the method more freely ← "don't know what a floppy disk is"
• when making gets cheap, value moves to intent and judgment ← "express clear intent, to evaluate tradeoffs"
• doing the work changes the spec, so spec-in, product-out fails ← "It's not spec in and software out."
• forcing early formal input hides ambiguity until it returns as confident error ← "very confidently produce the wrong thing"
• organize by meaning once retrieval reads meaning, not location ← "evaluate code based on semantic context"