Category / T-CAT / BUSINESS
Business
Business / T-2026-0339
Gstack puts AI agents in your meetings as talking specialists
Gstack ships 18 open-source voice agents that join Google Meet as specialist roles. A test of agentic voice in real meetings.
SourceAI YC interview with Gstack agents
Business / T-2026-5223
Grok 4.5: SpaceXAI and Cursor build a model from agent data
SpaceXAI and Cursor launch Grok 4.5, a mixture-of-experts model trained on real developer-agent interactions, with a distributed agent system generating training environments at…
SourceGrok 4.5
Business / T-2026-4684
BrainFeed: The app that wants to be your new homepage
BrainFeed offers a personalized learning feed that redirects your scroll. A commentary on what it means for AI, attention, and the business of focus.
SourceBrainFeed
Business / T-2026-6005
Openbase returns as an AI agent voice manager, leaving its old life behind
Openbase, the Y Combinator-backed open-source discovery platform that shut down in 2023, relaunches as a voice-based AI agent management tool. The pivot reflects a broader shift…
SourceOpenbase
Business / T-2026-4925
Google's Gemini 3.6 Flash family: cheaper tokens and a bet on agentic efficiency
Google releases Gemini 3.6 Flash, 3.5 Flash-Lite, and 3.5 Flash Cyber, cutting token costs and verbosity for agent workflows.
SourceGemini 3.6 Flash Family
Business / T-2026-9048
AgentManager's 99-Agent Catalog Reveals the AI Coding Tool Gold Rush
AgentManager, an open-source CLI tool by Kevin Elliott, catalogs 99 AI coding agents across multiple package managers. The project reveals a fragmented ecosystem and a growing…
SourceAgentManager
Business / T-2026-2775
AGINE Academy turns Claude training into a role-playing game
AGINE Academy turns Claude training into a role-playing game with a UNIT digital double, priced at $25 per month. A commentary on what this means for AI skill-building.
SourceAGINE Academy
Business / T-2026-7770
CreateOS Sandbox: A Firecracker Micro-VM for Every AI Agent
CreateOS Sandbox launches hardware-isolated micro-VMs for AI agents in ~30ms. Each agent gets its own kernel, private networking, and pause-to-zero snapshots.
SourceCreateOS Sandbox
Business / T-2026-1196
Inkling: The 975B open-weight model that forces a reckoning on fine-tuning
Inkling is an open-weights 975B multimodal model from Thinking Machines Lab. It forces a strategic question: is fine-tuning still relevant in the era of trillion-parameter models?
SourceInkling
Business / T-2026-2285
BaseRT claims fastest Apple Silicon inference, beating llama.cpp by up to 56%
BaseRT, a native Metal inference engine, beats llama.cpp by up to 56% on decode and 1.81x on MoE prefill. The tradeoffs are telling.
SourceBaseRT
Business / T-2026-0887
Zro's private inference pitch: coding agents without the cloud
Zro launches a private inference engine for coding agents, challenging the cloud-based model by running LLMs entirely on-device.
SourceZro
Business / T-2026-0749
Kit For AI: The MCP memory layer that asks what RAG infrastructure is for
Kit For AI offers a memory layer for AI agents via MCP tools, removing the need to build RAG pipelines. The commentary explores what this means for the AI infrastructure market.
SourceKit For AI