Context Engineering: Why Less Context Makes Better Agents
Context engineering explained: why long-running AI agents suffer context rot, what 2026 research shows about pruning and compaction, plus a Python demo.
Llm
Kimi K3 Review: The 2.8T Open Model That Beats Claude on Paper
Kimi K3 review after a week on real code: Moonshot's 2.8T open-weight model tops benchmarks at a third of Claude's price, but it's slow and hallucinates.
AI Agent Memory: How LLM Agents Remember Across Sessions
AI agent memory explained: the four memory types, how the MemoryOS paper borrows from operating systems, and a runnable Python example you can extend.
How to Build an AI Agent in Go (From Scratch, No Framework)
Build an AI agent in Go from scratch: a tool-calling loop against the Anthropic API in ~120 lines, no frameworks, with real code and the gotchas I hit.
Claude Sonnet 5 Review: A Week With Anthropic's New Default
Claude Sonnet 5 review after a week: it nearly matches Opus 4.8 on coding, beats it on Terminal-Bench, and runs at just $2/$10 — but the tokenizer bites.
FastContext: Why Coding Agents Waste Half Their Tokens on Repo Search
Microsoft's FastContext shows coding agents burn 56% of tool turns on exploration. A 4B subagent cuts token use 50% and lifts SWE-bench scores 3 points.
GLM-5.2 Review: 753B Open-Weight Model That Undercuts GPT-5.5
GLM-5.2 scores 62.1 on SWE-bench Pro vs GPT-5.5's 58.6, ships under MIT, and costs $1.40/M input tokens. Benchmarks, pricing, and the China data question.
GPT-5.5 Review After Seven Weeks: Where It Beats Claude and Where It Doesn't
GPT-5.5 hits 82.7% on Terminal-Bench and uses 72% fewer tokens than Claude — but loses SWE-Bench Pro to Opus 4.7. Seven weeks of real agentic use, reviewed.
Claude Fable 5 Review: 80% SWE-Bench Pro, but Read the Fine Print
Claude Fable 5 hits 80.3% SWE-bench Pro and 29.3% FrontierCode Diamond. It also costs 2x Opus 4.8, retains your data 30 days, and silently falls back.
DeepSeek V4 Pro Review: 80% SWE-bench at 1/7th Claude's Price
DeepSeek V4 Pro scores 80.6% on SWE-bench Verified at $1.74/M input tokens — 7x cheaper than Claude Opus 4.7. Real benchmarks, costs, and safety gaps.