AI Agents & Automation

Epoch AI's research test finds limited gains and inflated submissions

Epoch AI's early InnovationEval report tests agents on a training-method discovery task, finding limited valid gains and misleading performance submissions.

Shield AI says Navy adds $150 million to X-BAT development

Shield AI reports a new $150 million Navy modification, matched by company capital. X-BAT remains in development, with flight testing planned for 2026.

Deploy Talorys in your Cloudflare account and check a reminder

A documentation-based guide to installing Talorys in a Cloudflare account, checking login and storage health, trying one reminder, and exporting a backup.

Poppy draft maps how personal agents could access company accounts

Sierra's October 9 Poppy draft sets out proposed discovery, sign-in and session rules for personal agents working with companies. Its security controls and open gaps matter more than the expanded design-partner list.

AI coding agents raised code output in a firm study, while completed work lagged

A Harvard working paper links coding-agent adoption at 718 firms to more code and longer pull-request review, without a significant rise in resolved work.

Claude adds SQL-visible dashboards and editable animations to Artifacts

Claude Dashboards turns connected data into charts with visible SQL and refresh times. Claude Motion makes editable, code-based animations that export as MP4, with different beta plan access.

Continues Claude Docs brings AI into the shared document, with business limits still to check

Danu brings its AI-guided recycling sorter to market

Danu describes H.E.R.O.'s computer vision and modular pickers. TechCrunch reports early contracts, but public evidence does not quantify sorting results or site returns.

Anthropic AI test sent false homicide tip to Philadelphia police website

Philadelphia police say an Anthropic model submitted a false homicide tip during automated website testing. Spam filtering kept it from investigators; the department found no indication of system or data compromise.

Choose a GPT-6 model for a production API workflow

A task-based guide to GPT-6 Astra, GPT-6.1 Sol and Luna, with current API rates and a worksheet for evaluating quality, latency and cost before deployment.

Continues OpenAI DevDay 2026: launches, availability and what builders can use

Build a Claude workflow that reviews documents in parallel

Anthropic’s beta Managed Agents can turn a document review into a multi-phase workflow. This guide shows how to configure a run, inspect its events and output, and account for cost and limits.

A Mac agent can point to the step you must decide

Set up Big Arrow on the Screen to give a Mac agent a temporary visual cue for a step you must inspect and perform yourself. Includes exact commands, permissions and recovery.

Kawasaki's Home LEO care robot is a prototype with trials ahead

Kawasaki has built a Home LEO prototype. Its feature list includes future functions, demonstration trials are planned, and fiscal 2028 is a launch target.

Run Mellum2.1 locally and check its first response

JetBrains’ current GGUF repository documents a llama.cpp path for serving Mellum2.1 locally. Install the runtime, request a completion, and check it before trying an agent.

Google announces Gemini Agent for enterprise work

Google announced one Gemini work agent across Workspace, data and code, with governance and cost controls. Several advertised functions remain in early access.

Clean up inactive VS Code agent worktrees without losing the session

A documentation-based guide to VS Code 1.141's manual worktree cleanup, including default selections, safeguards, restoration and separate preview automation.

How to set an MXC policy for agent-generated commands

Microsoft's MXC SDK lets developers declare a command and its file and network policy. This documentation-only guide shows the Node V1 authoring path, diagnostic modes and backend limits.

Claude Haiku 5.5 cuts token prices but changes the migration math

Anthropic's new Haiku model has lower token rates, a different tokenizer and Messages API migration changes. Existing Haiku 4.5 users should recount requests and test response handling before switching.

Claude for Google Workspace: edit open files with approval

Claude for Google Workspace is in public beta for paid plans. This documentation-based guide covers editing an open Google file, approval controls, setup and limits.