Deep dives
The AI stories that actually mattered, followed all the way to ground. New dives most weeks.
Astra vs Fable: what $20 per million actually buys
The same $20 blended shelf buys 4.8 capability points per dollar on GPT-6 Astra and 3.3 on Fable 5.1. Where the premium pays, and where it burns.
Consumer agent wars: the scoreboard after week one
Muse took #1 in two weeks, ChatGPT owns the workflow layer, Google and Apple hold the OS gates. The agent wars are a distribution race in costume.
Meta's Muse: the human concierge behind the AI agent
Reuters: contractors quietly place some Muse calls for half of Meta's employees. Agent reliability is part-human, and disclosure is the trust question.
Meta's Muse: the open-source blueprint behind its #1 agent
Meta's Muse hit 2.5M downloads in two weeks, modeled on open-source OpenClaw. The architecture Meta took, the wrapper it added, the fallback it hid.
GLM-5.3 vs DeepSeek-V4-Pro: the open-weight value war
GLM-5.3 vs DeepSeek-V4-Pro: both open weights, both 1M context, one costs 3.5x more. Capability, speed, and the value math compared workload by workload.
18,000 posts before anyone looked
OpenAI agents posted 18,000 times on a German wiki over two months before anyone noticed. What the incident exposes about agent oversight.
Instinct: the AI Assistant With Its Own Email Address
The viral AI assistant that signs up for accounts on its own raised $350M at $2.5B. What Instinct does, the access it demands, and the trade.
Inside DeepSeek's Play: Own Chips, 1% Costs, Open Guardrails
DeepSeek runs China's largest Huawei cluster, ships agent models at a fraction of frontier cost, and skips the guardrails. The claims, audited.
AI Math Proofs: Navier-Stokes, Fermat, and the Revolt
OpenAI claims an 88-hour Navier-Stokes proof; Anthropic shipped a 13-million-line Fermat proof. The breakthroughs, the priority war, the verdict.
Are AI Agents Ready for Real Work? What the Evidence Shows
An AI agent fired an employee and forgot its own handbook. What agent traffic, wallets, and failure records say about AI agent reliability.
Fable 5.1 vs GPT-6 Astra: Same Price, Different Bill
Claude Fable 5.1 vs GPT-6 Astra: identical pricing, opposite bills. Which model wins coding, agents, long context, and reasoning: workload by workload.
Claude Code vs Codex
Claude Code vs Codex compared: Terminal-Bench scores, the 15-dollar vs 155-dollar cost gap, token efficiency, models (Opus 5 vs GPT-5.6), and which terminal agent to pick.
Cursor vs Claude Code
Cursor vs Claude Code compared: editor flow vs terminal autonomy, pricing, token efficiency, privacy, and the use-both workflow - with a decision rule that actually decides.
AI coding agents compared
How AI coding agents compare in 2026: terminal agents versus editor agents versus cloud agents, what each is for, and the honest test for which one earns a slot in your workflow.
The best AI agent tools, by what they actually finish
The best AI agent tools in 2026 sorted by job: coding agents, browser agents, research agents, personal assistants, and the frameworks for building your own.
LLM API pricing, compared honestly
How LLM API pricing actually works in 2026: per-token reality, cached input discounts, batch pricing, and which models are worth paying for at each tier.
What are AI agents, actually
AI agents explained without hype: what makes an agent different from a chatbot, what they can reliably do in 2026, where they fail, and the three capabilities that define them.
How to keep up with AI news without losing your mind
A sane system for following AI in 2026: filters that cut 90 percent of noise, the 20 minutes that matter, and why most AI news doesn't deserve your attention.
the short version lands every day.