High signal articles on AI, engineering, or whatever
“”
A discussion of the limits of LLM based software engineering, specifically the gaps on long term and complex maintainability type issues.
“This may be a fascinating may to assess how actually useful a model is on some level, if not having a fawning servant is the aim.”
Only two model families score above 60% on bullshit detection: Anthropic's latest models and Qwen 3.5. Reasoning models score lower, not higher — they build around wrong premises rather than reject them.
“A clear take on where we stand in AI adoption and particular value creation as of late February. The limits of "displacement" are clear.”
Citadel Securities is an award-winning global market-maker across a broad array of fixed income and equity products.
“”
Conrete experiments in what it takes to go full Ralph Wiggam and OpenAI learnings
“Essential reading on the dangers of over-contextualizing in AI systems.”
The context trap - AI is supercharging legacy leadership assumptions about context and control.
“The edit tool is the variable that matters most for coding agents.”
Improving 15 LLMs at coding in one afternoon. Only the harness changed.
“Analysis of where AI tools are headed in 2026.”
The race is to the top right, where AI agents work autonomously and have real control over your desktop.
“Data on how Claude Code usage grew from 25 mins to 45+ mins.”
Anthropic's study of its own API usage patterns measuring AI agent autonomy in practice.
“OpenAI's Harness team produced ~1,500 merged PRs with 3 engineers.”
5-month experiment: build and ship a real product with zero manually-written code.
“A simple but fundamental shift toward file-system-based agent memory.”
Files that live alongside your AI agent. The agent can read these files just like it would when working with a codebase.
“This post captures the architectural shift from GUI to CLI-based AI agents and why leveraging linux + file system is just such a great foundation.”
Moving away from the "VS Code clone sidebar" and towards CLI agents.
“”
Meta's paper on Just-in-Time test generation — automatically generating catching tests at the point of code change to detect regressions before they merge, evaluated across Meta's production codebase.
“Passive context beats active retrieval for AI coding agents.”
A compressed 8KB docs index embedded directly in AGENTS.md achieved a 100% pass rate, while skills maxed out at 79%.
“FDEs help build incrementally more valuable products from concrete use cases.”
What FDEs actually do and how to hire the right one for your team.
“Already thinking about what this means for hiring in a world where the IDE and hand-coding is not important”
\"You're in the middle of a refactor and the model says 8% context left before auto-compaction. What do you do?\"
“The moment that announced we were all in the harness engineering phase of AI engineernig, which as of this writing we remain in.”
Creating a more effective harness for long-running agents, inspired by human engineers.
“MSR 2026 paper on the tradeoffs of AI-assisted coding.”
How Cursor AI increases short-term velocity and long-term complexity in open-source projects.
“Fascinating oral history of the AI paradigm shift.”
How LLMs upended the field of NLP in just a few years.
“Revolutionary approach to hallucination detection.”
LLM hallucinations aren't bugs - they're compression artifacts.
“80+ page survey of all prompting techniques.”
The most comprehensive study of prompting ever done - 1,500+ academic papers analyzed.
“The definitive guide to agent architecture from the team behind Claude in the early days. So many unknown unknowns.”
Best practices and patterns for building production AI agents.
“Multi-agent orchestration for terminal-based AI coding. There are many competitors now, but a very interesting early implementation more or less subsumed by coding harnesses of big providers.”
Manage multiple AI terminal agents like Claude Code, Aider, Codex, OpenCode, and Amp.
“Required reading for anyone building with LLMs professionally.”
Why AI engineering is becoming the hottest role in tech.
“The canonical reference for understanding agent design patterns.”
Comprehensive overview of agent architectures and patterns.
Copyright © 2026 Thomas Neil