The main front of 2026. The reliability threshold for long tasks has finally started to give, and this is where autonomy first turns into money.
About this eventAnthropic donated the tool-connection protocol to the Agentic AI Foundation; its SDKs had passed 97M monthly downloads.
Donating MCP to a Linux Foundation fund separated the protocol from a single vendor and established it as a shared layer for tools and data. The standard does not make an agent reliable by itself, but it lowers integration cost and allows one tool to move between different models and applications.
Source: AnthropicAbout this eventThe agent-to-agent protocol reached a stable release with signed agent cards.
About this eventA flagship aimed at agentic work in code, biology and cybersecurity.
About this eventMETR measures a rising horizon for software tasks frontier agents complete at a given reliability; estimates beyond 16 hours remain uncertain.
About this eventThe agent works a mouse and keyboard where no API exists — the broadest and most fragile way to act.
About this eventGartner's projection for enterprise applications containing task-specific agents.
About this eventWho pays when an agent wires money to the wrong place. Without an answer, agents stay away from real accounts.
About this eventSeveral specialised agents under one lead — works in demos, falls apart on long tasks.
About this eventNot a tool inside someone's app but a participant with its own access, budget and reporting.
Primary material behind this dossier: papers, lab publications and official reports.