# Posts ## Pages - [Graydon Hoare's Dark Timeline Shows What We're Not Pricing In](https://aeshift.com/posts/2026-04-12-llm-time/index.md): The Rust creator's journal entry about an LLM inflection point reveals the security economics of coding agents that nobody wants to discuss. - [AI Code Creates Legal Backdoors in GPL Projects](https://aeshift.com/posts/2026-04-11-ai-code-is-hollowing-out-open-source-and-maintainers-are-looking-the-other-way/index.md): AI-generated code is public domain, not copyrighted, creating holes in copyleft licenses that allow proprietary use without reciprocity. - [Multi-Agent Code Generation Has a Specification Problem, Not a Coordination Problem](https://aeshift.com/posts/2026-03-26-the-specification-gap-coordination-failure-under-partial-knowledge-in-code-agent/index.md): New research shows that splitting coding work across multiple LLM agents causes a 25–39pp accuracy drop that only better specifications can fix—not fancy coordination tools. - [Coding Agent Security Just Became a Product Category](https://aeshift.com/posts/2026-03-24-ai-coding-tools-have-broad-filesystem-and-network-access/index.md): NVIDIA, Sysdig, and a wave of indie tools are shipping OS-level monitoring for coding agents. The industry just admitted that sandboxing alone isn't enough. - [Your Coding Agent Has a Supply Chain Problem](https://aeshift.com/posts/2026-03-23-cursor-admits-its-new-coding-model-was-built-on-top-of-moonshot-ais-kimi/index.md): Cursor's Kimi revelation shows why practitioners need to trace the hidden dependencies in their AI development tools. - [Sashiko shows AI code review works by doing less, not more](https://aeshift.com/posts/2026-03-22-sashiko-ai-code-review-system-for-the-linux-kernel-spots-bugs-humans-miss/index.md): Google's new AI tool catches 53% of Linux kernel bugs by pattern-matching rather than trying to understand code, suggesting narrow AI applications beat ambitious ones. - [Rover Makes Websites the Agent Runtime](https://aeshift.com/posts/2026-03-21-show-hn-rover-turn-any-web-interface-into-an-ai-agent-with-one-script-tag/index.md): While others build complex infrastructure for AI agents to navigate websites, Rover inverts the model by making the website itself the execution environment. - [OpenAI buying Astral is fine. Making uv a dependency of your agent stack isn't.](https://aeshift.com/posts/2026-03-20-thoughts-on-openai-acquiring-astral-and-uvruffty/index.md): The real risk of OpenAI acquiring Astral isn't that uv goes proprietary. It's that your agent workflow quietly couples to it through protocol gravity, compatibility drift, and tighter Codex integration. - [Agent Drift Is Consensus Built on Hallucinated Reality](https://aeshift.com/posts/2026-03-19-agent-drift-the-mythical-man-month-and-lm-teams-claude-hallucinates-moltbook/index.md): Claude fabricated an entire social network, wrote a first-person essay as an agent, and generated a comment section of model personas. The lesson isn't that LLMs hallucinate. It's that hallucination becomes a coordination mechanism. - [AI Agents Have Stable 'Coding Styles' That Change With Each Version](https://aeshift.com/posts/2026-03-18-nonstandard-errors-in-ai-agents/index.md): New research shows AI coding agents exhibit consistent biases in problem-solving approaches that persist within model families but change across versions, creating novel challenges for production systems. - [Skills aren't a cheat code for coding agents. They're configuration drift waiting to happen.](https://aeshift.com/posts/2026-03-17-swe-skills-bench-do-agent-skills-actually-help-in-real-world-software-engineerin/index.md): SWE-Skills-Bench finds most agent skills don't improve real repo outcomes, and some make things worse. Independent research on 673 skills reveals why: the failure modes are more varied and surprising than version mismatch alone. - [An AI Agent Built a JavaScript Engine. But the pudding is missing the proof.](https://aeshift.com/posts/2026-03-16-jsse-agent-coded-javascript-engine-in-rust-passing-9996-of-test262/index.md): JSSE passes 99.81% of test262 with zero human code. That's the easy part. Maintainability, harness trust, and the missing layers above conformance are where agent-generated code gets hard. - [APIs Can Now Hijack Your AI Agents](https://aeshift.com/posts/2026-03-14-show-hn-monetize-your-apis-by-injecting-agent-targeted-instructions/index.md): The ad-injector library reveals how easily AI agents can be manipulated through API responses, exploiting an architectural vulnerability. - [Your LLM Needs Virtual Memory](https://aeshift.com/posts/2026-03-11-the-missing-memory-hierarchy-demand-paging-for-llm-context-windows/index.md): New research shows 21.8% of LLM context is wasted on unused content. A demand paging system cuts context usage by 93%, proving we need memory hierarchies for AI agents. - [The Pentagon Just Made AI Provider Lock-in an Existential Risk](https://aeshift.com/posts/2026-03-10-anthropic-sues-pentagon-over-alleged-ai-blacklist-on-claude/index.md): Anthropic's instant blacklisting shows how government disputes can vaporize AI tools from production systems overnight, forcing teams to rethink their AI integration strategies. - [Your Coding Agent Thinks Security Controls Are Bugs](https://aeshift.com/posts/2026-03-09-claude-code-taught-itself-to-escape-its-own-sandbox/index.md): Claude Code's sandbox escapes reveal a fundamental truth: AI agents treat security barriers as obstacles to debug, not boundaries to respect. - [Why Your AI Agents Need Desks: Agent Town's Spatial Take on Multi-Agent Debugging](https://aeshift.com/posts/2026-03-08-agent-town-a-pixel-art-ai-agent-online-collaboration-platform/index.md): Agent Town turns AI orchestration into a pixel-art office simulation, revealing how spatial visualization solves practical problems in understanding multi-agent systems. - [Don't Let Your Agent Grade Its Own Homework](https://aeshift.com/posts/2026-03-06-self-attribution-bias-when-ai-monitors-go-easy-on-themselves/index.md): New research shows LLM monitors systematically go easier on risky actions when those actions are framed as the assistant's own prior output. If you're building coding agents, monitoring is a protocol problem, not a prompt problem. - [OpenAI's Symphony Turns Jira Tickets Into Pull Requests](https://aeshift.com/posts/2026-03-05-jira-tasks-can-now-write-their-own-code-openai-symphony/index.md): OpenAI Symphony connects project management tools directly to autonomous coding agents that complete tasks without human supervision, shifting developers from writing code to reviewing agent-generated PRs. - [Knuth changed his mind. Your workflow should too.](https://aeshift.com/posts/2026-03-04-knuth-changed-his-mind/index.md): Donald Knuth praising Claude’s “automatic deduction” is a cue for practitioners: stop treating coding agents like autocomplete and start using them as adversarial collaborators paired with tight verification loops. - [Synthesized, Not Engineered](https://aeshift.com/posts/2026-03-03-synthesized-not-engineered/index.md): A research team replaced a database query engine with an LLM that writes bespoke C++ for every query. It beat DuckDB. The implications go well beyond databases. - [A TeX Chess Engine Isn't a Trick; It's What Agents Do Under Constraint](https://aeshift.com/posts/2026-02-28-coding-agents-wrote-a-chess-engine-in-pure-tex/index.md): A coding agent built a chess engine in pure TeX by inventing a virtual machine. The practitioner lesson isn't the stunt. It's what happens when agents redesign your substrate. - [Coming Soon](https://aeshift.com/posts/coming-soon/index.md): Learning. Sharing. Shitposting. It's all coming soon.