Agents & Automation
bluemat briefs and decodes on Agents & Automation.
- Google opens a preview REST API and MCP server for Gemini Notebook
- ParetoSaddler paper: apply cost preference only at the end of agent harness search
- Pipecat Labs releases tiny CPU classifiers for tone and intent in voice agents
- Decoding Anthropic's Skill authoring guide: lean SKILL.md files that actually trigger
- IBM-attributed JevAct paper gates agent tool calls against policy before they run
- Tessl's skill-eval study: skills help, but agents often don't load them
- Google opens a preview of one MCP-based path for apps into Gemini and Search
- Glasser lets AI agents buy Ahrefs, Semrush and Apollo data one API call at a time
- Cloudflare Computer v0.7.0: replay after crashes, pause for approval mid-run
- HarnessMM paper reuses settled agent decisions to cut tokens per task
- Neuralwatt data: your coding-agent harness decides how often the prompt cache breaks
- Study: frontier models rewriting their own agent harness gain little on held-out games
- Plain RAG over raw chat history beats agentic memory layers in a new benchmark study
- Anthropic donates the Agent Skills standard to the Agentic AI Foundation
- Google's agents-cli adds agent-building and eval skills to Claude Code and Codex
- Databricks details MemCurrent, the memory design behind Agent Bricks Memory
- Self-evolving agents forget earlier tasks as work drifts, new preprint finds
- Meta's thin orchestrator: a deterministic router with 14 specialist agents, decoded
- Tabular Editor's guide to writing agent skills: encode the business, then cut
- DeepSeek opens standalone Search and Fetch APIs at $0.75 per 1,000 requests
- CLI-Anything generates agent-ready CLIs for GUI-first apps
- Perplexity open-sources SPACE, the Firecracker sandbox behind its Computer agent
- Four Gemini 4 Argon agents beat Minecraft in a 27-hour autonomous coding run
- Google Research releases Harness², code that tunes an agent's harness without retraining
- Claude Platform adds OpenTelemetry trace export for agents, API and SDKs
- ai-newtab: an open-source Claude agent that rebuilds your Chrome new tab
- New paper proposes 'agent plasticity' to measure whether self-improving agents learn
- Microsoft's Quicksand puts local coding agents inside a pip-installable QEMU VM
- Three changed prompt words broke two of four agent evals while dashboards stayed green
- AgentMail launches AgentID, an OpenID Connect sign-in for AI agents
- Meta open-sources Muse Gadget SDKs for ESP32 and Raspberry Pi builds
- AWS walks through a full Pipecat voice agent with Deepgram on SageMaker
- Epoch AI puts agent costs at about $16–18 an hour for Codex, $24–50 for Claude Code
- Databricks' Omnigent puts Claude Code, Codex, Cursor and Pi under one control layer
- Pi reverses its anti-MCP stance and builds MCP into the core via Codemode
- Fatih Arslan's markdown-plan workflow for running a fleet of coding agents
- VeriHarness paper: agent rollout consensus can hide errors, disagreement can expose fixes
- Google Research's Cogentic pairs parallel provers with adversarial verifiers
- exe.dev argues for running fewer coding agents, not more
- E2B Embed runs E2B's Firecracker sandboxes on a single machine you control
- ChatGPT's MCP Events: webhook triggers from your own MCP server, with real gaps
- Vercel report: skills.sh hits one million agent skills in seven months
- Microsoft Research's Agensh scales coding agents to 1,024 without an orchestrator
- DAIR.AI tutorial puts Jev on an agent harness's routing, tool-gating and done checks
- HarnessRouter puts Codex and Claude Code behind one Responses-compatible API
- Anthropic opens a directory submission portal for Claude plugins
- Personal AI agents upsell users they infer are wealthy, study finds
bluemat home