CTGT releases open source tool Concordance

In order to help businesses harness Generative AI and make it more consistent, the stealthy startup CTGT is releasing a new system called Concordance under the Apache 2.0 open-source license, giving enterprises free tooling to investigate why an AI application makes different decisions when asked to perform the same task.

The release includes the testing protocol, reference implementation and experimental data, according to CTGT. Businesses can use and modify the software commercially under the license; the company says it has no paid or hosted Concordance edition planned.

NCM: A New Approach to AI Language Generation

Next-Concept Modeling (NCM) is a language modeling architecture introduced by researchers from Hazy Research at Stanford University. It proposes a significant shift in how AI models generate text.

Instead of producing one word or token after another, as traditional autoregressive LLMs do, NCM constructs a hierarchical representation of the response, starting with an abstract structure and progressively refining its details.

The central idea is to generate a global representation of the response first, then refine it in parallel, rather than constructing the response sequentially, token by token.

Claude Duet: Anthropic’s Agent Runs on Opus 5.5

Anthropic has opened a research preview of Claude Duet, a new proactive AI agent that messages users, works inside connected apps, and keeps running tasks after they close the app. The move, reported by The Verge on October 21, 2026, puts Anthropic squarely into the fight for a new category of AI product: the “always-on” personal agent that Meta, OpenAI and Google have all raced to ship since late summer.

Time to Scale Up Diffusion LLMs

Inception's Mercury 3 is a diffusion LLM running at 1,000+ tokens/sec, and Mercury Decide is a decision model that can now read images. Here's how they compare, what the independent evals say, and how fast they really are.

The Greatest AI Coding Workflow Ever

AI LABS explores seven actionable strategies for building with AI agents. These methods focus on verifying project direction, maintaining security, and implementing rigorous testing cycles to ensure applications meet user needs effectively without drifting from core development goals.

Google Open Sources Harness² Framework

Google Research's Harness² is an open-source framework that improves AI agent performance by optimizing the execution environment rather than retraining the underlying model.

Harness² addresses the critical gap in AI agent development by demonstrating that performance improvements don't always require more sophisticated models - sometimes just better system configuration and execution strategies can yield significant gains.

Amazon unveils a free, fast, open source Jev killer

The race to make AI agents cheaper and faster is moving beyond the models that write their answers. In the two weeks since TypeSafe AI introduced Jev, a model that chooses among predefined options instead of generating prose, developers have released a rush of competing decision models, many with downloadable weights. 

Now, Amazon is joining them with an open model designed to sit inside an agent's workflow and decide whether a proposed action should proceed: Strands Decider 2B, a roughly 2-billion-parameter model fine-tuned from an Alibaba Qwen base that answers bounded questions with choices and probabilities in a single pass.