Does AI actually make developers faster? The answer is more complicated than vendors suggest. These three pieces separate measurement from perception.

Not All AI-Assisted Programming Is Vibe Coding

TLDR: Simon Willison draws a clear line between “vibe coding” — accepting AI output without scrutiny — and responsible AI-assisted engineering. Vibe coding is fine for prototypes and disposable scripts, dangerous for production systems.

Key Insight: Know which mode you are in and switch deliberately. Vibe coding and rigorous AI-assisted development are different activities with different risk profiles.

Read the full article ->

AI Coding Is Now Everywhere. But Not Everyone Is Convinced

TLDR: MIT Technology Review covers the METR study finding that experienced developers were 19 percent slower with AI assistance, despite perceiving themselves as faster. The gap between perception and measurement is significant.

Key Insight: Build measurement into AI adoption, not just vibes. Perceived productivity and actual productivity can diverge significantly.

Read the full article ->

AI Coding Tools May Not Speed Up Every Developer

TLDR: TechCrunch reports on the same METR study: experienced developers took 19 percent longer with AI on familiar codebases while believing they were 20 percent faster. The perception-reality gap held across task types.

Key Insight: AI may slow experienced developers on familiar codebases — the gains appear on unfamiliar territory where the developer lacks context the model can provide.

Read the full article ->

What does this mean for developers?

The speed debate reveals a measurement problem, not a tool problem. AI coding tools likely deliver genuine gains on unfamiliar code and rote tasks, but may introduce overhead on work where the developer already has deep context. Teams should measure actual outcomes rather than relying on developer self-assessment when evaluating AI tool ROI.