RE: https:// mastodon.social/@kornel/116394 628325980197 Really good here to move away from t
| Source: Mastodon | Original article
A thread on Mastodon sparked fresh debate about the next leap in AI‑driven software development after Kornel Korneliuk posted a “Coding Black Mirror” scenario on 13 April. He asked followers to imagine large language models (LLMs) that could generate tens of thousands of tokens per second, effectively rewriting an entire codebase on every keystroke. The post quickly gathered reactions from developers, AI researchers and industry observers, who warned that such speed would turn LLMs into “sloppy devs” whose output would need exhaustive human review, while also hinting at a radical shift in how software is built and consumed.
The conversation matters because it foregrounds a tension that is already emerging: LLMs are beginning to industrialise content consumption—mass‑producing documentation, tutorials and code snippets—while the tools that developers use to apply that content risk becoming de‑industrialised, i.e., less structured and more chaotic. Kornel’s speculation builds on the performance gains announced just days earlier when Google unveiled Gemini 3.1 Pro, a model whose inference throughput is more than twice that of its predecessor. Faster inference lowers the barrier to real‑time code synthesis, making the “rewrite‑on‑type” vision technically plausible within the next year.
What to watch next is whether major AI vendors will deliberately throttle generation speed to preserve code quality, or whether new safety layers—such as Anthropic’s Claude Code, recently standardised across ARI’s engineering teams—will become the default guardrails. Industry analysts will also monitor early adopters experimenting with ultra‑fast code assistants in integrated development environments, looking for signs of productivity gains versus error proliferation. If the Mastodon discussion translates into concrete product roadmaps, the balance between speed and reliability could reshape software engineering pipelines across the Nordics and beyond.
Sources
Back to AIPULSEN