Ashlr Brief.

Inception launched Mercury 3, a diffusion language model it says runs above 1,000 tokens per second with a 1 million token context window.

A coding agent can read a key stored in .env. Phantom Secrets replaces it with a local placeholder and supplies the real credential to supported API requests.

Claude Code and Cursor change how agents run, OpenAI discloses four more incidents, Replit makes OpenAI its default, and a paper breaks agent loops.

Speech-to-text gets common words right and your names wrong. Lexicon is one file of the words you say, fixed before your agent reads them.

Jessica Wachtel’s single-run comparison found similar results on two agent tasks, with Muse Wonder using more tokens and costing more than Claude Fable 5.1.

Why read a whole file for one function? ashlr-plugin offers compact reading and search tools, with benchmarks you can repeat on your own code.

OpenAI is ending API access to GPT-4o, GPT-4o mini, GPT-4.1 and related models on 14 January 2027. The three-month window and behavioral differences in the recommended GPT-6 Luna replacement require code and agent workflow changes.

On a laptop used for several clients, the active account can be the wrong one. Locus scopes commands and agent tools to an explicitly pinned identity.

Meta introduced two models for agent tasks. Muse Wonder is available through its API and Muse Code; Muse Wonder Full is available to Muse Code subscribers. Their access routes differ.

The agent finished its plan. Now someone has to read it. Ashlr MD turns Markdown files into formatted documents that teammates can review, edit and export.

GitHub released a public preview of computer use that lets Copilot agents read, click, type, scroll, and drag inside macOS and Windows applications, including legacy GUI software without APIs.

A new app can mean a database, hosting and several more setup screens. Ashlr Stack helps configure backend services from the terminal or an AI agent.

GitHub added OpenAI's GPT-6 Sol and Luna to Copilot on September 22, and Codex CLI added both to its model picker on September 23. Access differs by plan.

Four bots give our CEO a manageable view of the work. Behind them are specialist agents, clear responsibilities and decisions that still stop with a person.

OpenAI released GPT-6 Sol and Luna with default pricing halved versus GPT-5.6 equivalents. GPT-6 Sol is $2/$10 per million input/output tokens and Luna is $0.10/$0.50, alongside improved prompt caching and more direct response style.

Writing a change is only the start. Our agents follow pull-request checks, a one-at-a-time merge rule and a separate human decision to publish packages.

Alphabet's Intrinsic released ROS-compatible control, motion and grasp planning under Apache 2.0 at ROSCon 2026. The company calls it early days, and the commercial boundary is not spelled out.

Which agent is working, and what needs approval? Phantom is our workbench; coordination for proactive agents is still being built.

xAI says Grok 4.7 is its most capable model for coding and knowledge work and kept the entry price of Grok 4.6. Access is split across the API, Cursor, Grok Build and a gradual GitHub Copilot rollout.

We set up a booth at the JMU Gilliam Center to show what we build. Here are the photos and what we put on the table.

The Independent International Scientific Panel on AI published an advance unedited brief on agents that escaped their test environment. It surveys safeguards rather than setting requirements, and says none of them guarantee safety.

ashlr-plugin 1.36.5 replaces its 57 percent savings headline with per-call figures and fixes extra overhead on tiny edits. This is an Ashlr release.