Stroustrup drops the mic: complaints are the badge of languages that matter
Bjarne Stroustrup just crashed another panel by refusing to pick a “best” language, then spent the next twenty minutes explaining why the ugliest one is still queen of the data center. His slide showed two buckets: languages people love to hate and languages no one uses. C++ sits in the first, right between Python’s GIL rants and JavaScript’s type chaos.
The ai honeymoon is over
While GitHub Copilot spits out boilerplate, firmware teams are waking up to 3 a.m. panics when a 200-microsecond interrupt misses its deadline. Stroustrup’s answer: you can’t wish latency away with a pip install. The same libraries that make prototyping “effortless” inject garbage-collection pauses that balloon when the model starts paging tensors. His punch line: abstraction without control is a loan with compound interest.
He traced the problem to a generational split. Bootcamps teach students to glue pre-trained models; universities still hand them pointers and cache lines. The result is two parallel labor markets: one that ships Jupyter notebooks to marketing and another that keeps the power grid from browning out. Stroustrup claims the second group quietly earns 40 % more, but recruiters keep posting “Python only” because the req was written by someone who never measured a cache miss.
Complexity is not the enemy—opacity is
Modern “simple” languages outsource their mess to 300-MB container images. Under the hood, dependency graphs look like medieval tapestries: beautiful until you try to change a thread. C++ looks monstrous because its monsters are visible. You choose the allocator, you pick the move semantics, you eat the segfault. That transparency is why SpaceX firmware, Chrome’s renderer and the Unreal Engine still compile to the same grammar invented in 1985.
Stroustrup dropped a stat: the ISO C++ committee added 52 new features in the last three years, yet the runtime memory footprint of a hello world shrank 11 %. Try that while dragging in PyTorch, NumPy and the rest of the “just pip it” stack. The room laughed, then went quiet when he flashed a slide showing Python’s import time regression—38 % slower since 3.8—tracked by Instagram’s own production logs.
The academic exile
Meanwhile, the “perfect” languages—Elm, Zig, Crystal—live in the second bucket. They top Hacker News, then vanish from Stack Overflow’s annual survey. Stroustrup quipped that if no one complains about your language on Reddit, it’s not because it’s perfect; it’s because no one’s paycheck depends on it Friday at 5 p.m.
He closed with a warning to hiring managers seduced by AI demos: when the first brownout hits, the repo you’ll be grepping at 2 a.m. won’t be written in a language that promises zero-cost abstractions. It’ll be the one that lets you dump core, patch the allocator and redeploy before the cooling fans trigger a fire suppression system. In short, the language you hate is the one keeping the lights on. Learn its edges or learn outage arithmetic.