Applied AI Signals

Agents & Automation

The agent-framework war already consolidated and most contenders lost without noticing. What the surviving stacks actually share, and where each one stops.

Multi-Agent Orchestration

Supervisors, handoffs, swarms and parallel fan-out: when more than one agent genuinely pays for itself and when it stops.

AI Coding

The phrase 'AI coding' covers at least four distinct products, and the conflation is why most tool discussions go in circles.

Coding Agents

Terminal and IDE coding agents - Claude Code, Codex, Aider, OpenHands and peers: context handling, edit strategies, and review discipline.

LLMs & Reasoning

The defining change in this field is that 'reasoning' stopped being a model category and became a metered dimension of every model.

Computer Vision

The defining change in this field is not any single model.

Audio, TTS, Voice

The voice stack finally has a permissive open layer. What that changes for latency, turn-taking and licensing when you ship a real-time interface.

Image Models

The licence page decides more image-model adoptions than the leaderboard. What each family permits commercially, and what that costs you in quality.

Video Models

Video generation in 2026 is two markets wearing one label.

Robotics

Everything called 'robotics' in 2026 splits into two overlapping but differently-shaped disciplines, and most bad adoption decisions come from evaluating one with the other's criteria.

3D & Gaussian Splatting

Gaussian splatting is a scene representation: millions of small, coloured, anisotropic 3D ellipsoids, each with an opacity, that together reconstruct what a camera saw.

Agent Frameworks

LangGraph, CrewAI, AutoGen, Pydantic AI, Mastra and the rest - what each actually gives you over a plain loop, and where each stops.

Agent Harnesses

The layer between an agent and the system it acts on: permission boundaries, tool exposure, sandboxing, approval gates, and recovery.

Local AI Hardware

Apple Silicon, consumer GPUs, DGX Spark and Strix Halo: VRAM and bandwidth maths, and what a given model genuinely needs to run.

Local Inference Runtimes

llama.cpp, vLLM, Ollama, MLX, SGLang and LM Studio - throughput, memory behaviour, and which runtime suits which job.

Low-Cost Robot Platforms

The new accessible wave - Reachy Mini, Microduck, Open Duck Mini, K-Scale and peers: what they can do and what they cost to run.

Open-Source Robotics

LeRobot, ROS 2, SO-100/SO-101 arms and the open stack turning robotics into something you can actually reproduce.

Open-Weight Models

Llama, Qwen, DeepSeek, Mistral, GLM and the open frontier: capability, licence terms, and what they can actually displace.

The Token Economy

Pricing shapes, context windows, prompt caching, routing and quantisation - cost as an engineering constraint rather than a line item.