Graydon Hoare, the creator of the Rust programming language and a veteran of decades in systems software, published a journal entry on March 14, 2026 documenting what he calls "the fastest and most violent change to working conditions and assumptions I've witnessed in my career." Writing without conclusions or predictions — and with comments disabled — Hoare describes a capability threshold crossed in late 2025 and early 2026, driven by a new generation of frontier models. Within months, colleagues moved from casual experimentation to abandoning hand-written code entirely, while managers began demanding 100x productivity gains or issuing layoffs. Hoare himself acknowledges writing less code by hand, reserving his effort for API sketches, subtle logic, and what he calls "LLM-supervisory bits."
What Hoare finds most alarming is not coding capability but the discovery that LLMs are, in his view, even more effective at vulnerability hunting than at writing code. Because attackers need only occasional success while defenders must succeed consistently, this asymmetry has dramatically shifted the economics of software exploitation. Hoare's own team is now consumed responding to a flood of new security vulnerabilities — an arms race he describes not as a future risk but as an ongoing reality.
Hoare spent years at Mozilla Research designing Rust specifically to eliminate entire classes of memory-safety exploits by construction. That work targeted a well-defined layer: memory management, ownership, lifetimes. <a href="/news/2026-03-14-agile-manifesto-ai-addendum-prioritizing-shared-understanding-over-shipping">LLMs generating plausible-but-wrong logic, subtle authentication bypasses, and semantic errors</a> operate somewhere else entirely — a layer Rust's type system was never built to constrain. The same productivity gains LLMs provide also accelerate the rate at which that attack surface expands.
The post documents deep social fractures across the software community: maintainers quitting, <a href="/news/2026-03-14-redox-os-adopts-no-llm-contribution-policy-amid-growing-oss-ai-governance-debate">contributors banned, issue trackers closed to AI-generated submissions</a>, forks emerging, dependencies severed, and license debates reigniting with an intensity Hoare compares to the GPLv3 era. These fractures hit the Rust community with particular force. Its identity was built through years of ideologically cohesive governance — RFC processes, working groups, mentorship programs — structures that were designed around human contributors and are now straining visibly. Hoare describes professional and personal relationships ending, programmers mourning their identities, and a level of community anger unlike anything he has previously encountered.
Hoare closes without resolution. His final accounting is a single sentence: "The unit economics of making and breaking software in 2026 are completely different than they were in 2025." He offers no forecast for what comes next, and makes no attempt to adjudicate the community's conflicts. For anyone tracking the AI agent ecosystem, the post's value is straightforward: a primary-source account from one of open-source software's most credible technical voices, written from inside the disruption rather than from a vantage point above it.