The last year of AI in engineering has changed in a very different way than I expected

𝐓𝐡𝐞 𝐥𝐚𝐬𝐭 𝐲𝐞𝐚𝐫 𝐨𝐟 𝐀𝐈 𝐢𝐧 𝐞𝐧𝐠𝐢𝐧𝐞𝐞𝐫𝐢𝐧𝐠 𝐡𝐚𝐬 𝐜𝐡𝐚𝐧𝐠𝐞𝐝 𝐢𝐧 𝐚 𝐯𝐞𝐫𝐲 𝐝𝐢𝐟𝐟𝐞𝐫𝐞𝐧𝐭 𝐰𝐚𝐲 𝐭𝐡𝐚𝐧 𝐈 𝐞𝐱𝐩𝐞𝐜𝐭𝐞𝐝

A year ago, I was debating with engineers about AI, could it write production-ready code? Can it be used for 100% code generation?

My position was clear, and still is:

𝐀𝐈 𝐜𝐚𝐧 𝐠𝐞𝐧𝐞𝐫𝐚𝐭𝐞 𝟏𝟎𝟎% 𝐨𝐟 𝐭𝐡𝐞 𝐜𝐨𝐝𝐞. 𝐆𝐫𝐞𝐚𝐭 𝐞𝐧𝐠𝐢𝐧𝐞𝐞𝐫𝐬 𝐮𝐬𝐞 𝐢𝐭 𝐭𝐨 𝐝𝐨 𝐞𝐱𝐜𝐞𝐩𝐭𝐢𝐨𝐧𝐚𝐥 𝐭𝐡𝐢𝐧𝐠𝐬.

But where we are isn't what I expected.

𝐀𝐝𝐨𝐩𝐭𝐢𝐨𝐧 𝐢𝐬 𝐮𝐩. 𝐐𝐮𝐚𝐥𝐢𝐭𝐲 𝐢𝐬𝐧’𝐭.

AI adoption clearly increased. That should be a win.. but I’m arguing more than ever and seeing a split:

✅ A minority are using AI well, learnt over the years, strong fundamentals, validating outputs, shaping systems.

⚠️ A growing majority are producing what I can only describe as 𝐀𝐈 𝐬𝐥𝐨𝐩.

𝐓𝐡𝐢𝐬 𝐢𝐬𝐧’𝐭 𝐚𝐛𝐨𝐮𝐭 “𝐯𝐢𝐛𝐞 𝐜𝐨𝐝𝐞𝐫𝐬” 𝐚𝐧𝐲𝐦𝐨𝐫𝐞

“AI slop” used to mean non-engineers generating code that looked right but wouldn’t survive production.

Now seeing 𝐩𝐫𝐨𝐟𝐞𝐬𝐬𝐢𝐨𝐧𝐚𝐥 𝐞𝐧𝐠𝐢𝐧𝐞𝐞𝐫𝐬:

⚠️ Raising PRs that haven’t been run locally.

⚠️ Committing code they don’t understand.

⚠️ Merging changes that break immeditealy after merge.

𝐈 𝐧𝐨 𝐥𝐨𝐧𝐠𝐞𝐫 𝐭𝐫𝐮𝐬𝐭 𝐭𝐡𝐚𝐭 𝐜𝐨𝐝𝐞 𝐡𝐚𝐬 𝐞𝐯𝐞𝐧 𝐛𝐞𝐞𝐧 𝐞𝐱𝐞𝐜𝐮𝐭𝐞𝐝 𝐛𝐞𝐟𝐨𝐫𝐞 𝐬𝐮𝐛𝐦𝐢𝐬𝐬𝐢𝐨𝐧.

That’s a regression in engineering discipline.

𝐓𝐡𝐞 𝐛𝐚𝐫𝐫𝐢𝐞𝐫 𝐭𝐨 𝐜𝐨𝐦𝐩𝐥𝐞𝐱𝐢𝐭𝐲 𝐢𝐬 𝐠𝐨𝐧𝐞

Before AI, you could only build at your level of experience. Now engineers are generating systems far beyond their understanding.

But generating does not equal understanding.

Without understanding:

❌ You can’t reason about failures.

❌ You can’t debug.

❌ You can’t make safe trade-offs.

❌ You don’t know if it’s safe to release.

At that point, the behaviour and quality is no different from a non engineer vibe coding.

This isn’t just about bad code. It’s about value.

𝐄𝐧𝐠𝐢𝐧𝐞𝐞𝐫𝐬 𝐰𝐡𝐨 𝐨𝐩𝐞𝐫𝐚𝐭𝐞 𝐥𝐢𝐤𝐞 𝐭𝐡𝐢𝐬 𝐚𝐫𝐞 𝐝𝐞𝐯𝐚𝐥𝐮𝐢𝐧𝐠 𝐭𝐡𝐞𝐦𝐬𝐞𝐥𝐯𝐞𝐬.

“If this is just copying AI output, and it doesn’t work, why am I paying engineers for it?” A valid question for businesses to ask.

They have a point. If your contribution is:

⚠️ blindly generating code with AI

⚠️ pasting errors back into the model

⚠️ committing whatever comes out next

𝐘𝐨𝐮’𝐫𝐞 𝐧𝐨𝐭 𝐚𝐝𝐝𝐢𝐧𝐠 𝐞𝐧𝐠𝐢𝐧𝐞𝐞𝐫𝐢𝐧𝐠 𝐯𝐚𝐥𝐮𝐞.

𝐓𝐡𝐞 𝐧𝐞𝐰 𝐛𝐚𝐫

AI hasn’t replaced engineers. It’s raised the bar. The ones who stand out:

✅ 𝐈𝐧𝐭𝐞𝐫𝐫𝐨𝐠𝐚𝐭𝐞 𝐀𝐈 𝐨𝐮𝐭𝐩𝐮𝐭.

✅ 𝐔𝐧𝐝𝐞𝐫𝐬𝐭𝐚𝐧𝐝 𝐭𝐡𝐞 𝐬𝐲𝐬𝐭𝐞𝐦.

✅ 𝐔𝐬𝐞 𝐀𝐈 𝐭𝐨 𝐭𝐡𝐢𝐧𝐤 𝐛𝐞𝐭𝐭𝐞𝐫 𝐧𝐨𝐭 𝐥𝐞𝐬𝐬.

✅ 𝐎𝐰𝐧 𝐨𝐮𝐭𝐜𝐨𝐦𝐞𝐬.

The best are becoming dramatically more powerful. The average is drifting into low-value output and the gap is widening with larger adoption.

I’m not worried about AI replacing engineers, I’m seeing engineers putting themselves out of a job, and making a mess, as they stop being engineers.