Insights
Practical analysis and guides to help you buy, build, and work with AI more effectively.
Four open-source agent harnesses compared — which one a non-researcher should actually try
Agents have been hot for almost two years, but when people actually try to run one, many stall at step one. What to install. How to make it really do work instead of just chatting in a box. How to pi
ElevenAgents Architect: Tuning Voice Agents by Simply Talking
ElevenLabs added a built-in expert to its voice-agent platform called ElevenAgents Architect. The short version: without touching configuration yourself, you improve and tune a running voice agent jus
Figma Agent Is Out of Beta: An AI Coworker Inside Your Design File
Figma moved its built-in Agent out of beta and into general availability on October 6. On the surface this looks like a routine version bump, but for people who live inside Figma every day it marks a
Mistral Large 4 preview: Europe's flagship fights US frontiners on price
Europe's pricing counterattack On October 6, Mistral pushed its flagship Large 4 into public API preview. The pricing is aggressive: 1.36 dollars per million input tokens, 4.18 per million output, wi
Meta open-weights an 816B agent model you can run on two Mac Studios
An 816B open-weight chip to play On October 15, Meta released the weights of Muse Spark Open. It is a natively multimodal sparse-expert model: 816 billion total parameters, 42 billion active per toke
Claude Haiku 5.5 cuts small-model pricing to a tenth, and it actually changes how you build agents
Anthropic shipped Claude Haiku 5.5 on October 7, and at first glance it looks like the usual small-model refresh. Look closer and the interesting move is in the pricing, not the parameter count. This
Claude Opus 5.5 review: a pricing reset, not a capability leap
Opus 5.5, released September 22, looks like a flagship but is really an accounting move Anthropic shipped Claude Opus 5.5 on September 22. The pricing tells you it is not here to show off: 4 dollars p
SWE-2 review: 0.9 points behind the frontier, 64 percent cheaper
The 2026 coding-model story is not about who tops the chart It is about who gets close enough, cheaply enough. Cognition released SWE-2 on September 10, the coding model inside its Devin agent. On its
Meta Dropped a Price Bomb: It Tells 20 Speakers Apart, and 1,000 Minutes Costs $3
A 20-person meeting ends, and the notes are a mess Sales, engineering and marketing all talk at once. A remote colleague jumps in. When the meeting ends, the person writing the summary faces a tangle
Meta Open-Sources Muse Glimmer (2026): An Agent That Runs on One GPU, the Local-Deployment Turning Point
One-line positioning: an agent that runs on a single GPU, now open On August 10, Meta open-sourced Muse Glimmer. The pitch is concrete: a 30-billion-parameter model under the Apache 2.0 license that
Cloudflare Kitesurf (2026): A Browser Built for AI Agents - What 7x Less Memory Really Means
Browsers were built for humans for 30 years. This one says it isn't. On August 6, during Agents Week, Cloudflare shipped Kitesurf. Its one-line positioning is blunt: not a browser for humans, but one
Seedance 2.5 vs Wan3.0 (2026): Head-to-Head on 30-Second Video Generation - Which Fits Small Teams
A week apart, two Chinese video models both hit 30 seconds ByteDance's Seedance 2.5 and Alibaba's Wan3.0 shipped within days of each other, both generating 30-second clips in one pass. On the surface