OpenAI enters the agent orchestration layer

Introducing OpenAI Presence — the enterprise platform for deploying voice and chat agents across customer and internal workflows. This is OpenAI staking a claim on the orchestration layer — a shot squarely at LangChain and Microsoft's Copilot Studio. "A proven enterprise AI agent platform," they call it, which is a bold adjective for a launch-day product. — openai.com

SHOOTOUT

Kimi K3 vs DeepSeek V4 Pro vs GLM-5.2 — puts three open trillion-scale MoE flagships side by side on measured intelligence, MIT vs Modified-MIT weights, and real serving cost. The trade-off worth planning around: DeepSeek's efficiency against Kimi's reasoning. Read this before you commit infra budget. — Michal Sutter

Kimi K3: second only to Fable 5 on AA-Briefcase — lands Moonshot AI in second place on the agentic-knowledge benchmark, putting it in GPT-4o / Claude 3.5 Sonnet territory for multi-step reasoning and tool use. Second place is still a good place to be when Fable 5 is the ceiling. — wertyk

Laguna S 2.1 — a 10B-parameter code model from Poolside aimed at the local-inference and IDE sweet spot. At that size it goes head-to-head with Qwen2.5-Coder and DeepSeek-V2.5 — small enough to run yourself, big enough to be useful. — rexledesma

Reverse-engineering is cheap now — argues coding agents have rewritten the ROI math on hacking undocumented device APIs. "Prior to agents, it was entirely possible to reverse-engineer home devices. The problem was the ROI," Willison notes — agents now eat the boilerplate that made it not worth it. — simonwillison.net

How to Run Claude Code Agents for 24+ Hours — walks through keeping autonomous coding agents reliable across multi-hour loops. As context windows grow, state management is the real hurdle — this targets the reliability gap that current agentic IDE tools quietly leave open. — Eivind Kjosbakken

Big tech needs to justify AI spending as investors dump stocks — public markets want ROI evidence after the capex binge. For founders, the tell is a shift from growth-at-all-costs to unit economics across the AI stack. — 1vuio0pswjnm7

10 Open-Source No-Code AI Platforms for Building LLM Apps, RAG Systems, and AI Agents — While visual builders like LangFlow and Flowise are popular, this list highlights open-source alternatives that help teams prototype RAG pipelines without vendor lock-in. — marktechpost.com