“Let’s talk about LLMs”: a clear-headed post that refuses to pick a side

The debate

It has been reported that thousands of words have already been spent arguing what LLMs mean for the future — revolution, bubble, or something in between. The lobsters-linked essay doesn’t pretend to settle that fight. Instead the author leans into the fog: are we witnessing a productivity revolution, the precursor to some sci‑fi singularity, or merely another hype cycle that will leave useful wreckage in its wake? Tone is equal parts curiosity and skepticism. Short answer? Nobody truly knows. Long answer? Read on.

Terminology and scope

The writer insists on precision. “LLM” is the preferred term — not the fuzzy umbrella of “AI,” and not “GPT,” which allegedly carries branding baggage. When they say “LLM coding,” they mean any use of large language models to generate code, whether supervised, solo, or mixed. The author also carves out boundaries: they’ll stick to programming and technology — their wheelhouse — and avoid grandstanding about domains they don’t know. A dry aside even suggests we need a cute portmanteau for people who expect LLMs to take over every job except their own: classic Gell‑Mann Amnesia vibes.

Brooks redux — why the past still matters

To ground the hot takes, the post revisits Fred Brooks’ No Silver Bullet and its old but stubborn insight: no single breakthrough will buy an order‑of‑magnitude improvement in software productivity within a decade. Brooks’ distinction between essence (inherent difficulty) and accidents (avoidable friction) frames the piece. The question the author leaves hanging is sharp: will LLMs reduce accident‑type problems, or push against the essence of programming itself? That’s the emotional crux — hope versus hard realism. No neat conclusion, but a reminder: history and humility matter, even when the future looks like it could use a rewrite.

Sources: b-list.org, Lobsters