October’s turning point for agents in production: capability is no longer the bottleneck — cost and integration speed are
2026.10.27Source · Industry watch October 2026 is recorded by several industry trackers as a dividing month: three frontier labs shipped major updates within 18 days, two of them cutting API list prices by double-digit percentages, and the first credible Fortune 500 agent deployments crossed twelve months…
MCP is the de facto standard for agents — but the battle has moved to the gateway layer
2026.10.23Source · Industry watch The Model Context Protocol (MCP) now runs in roughly 78% of enterprise agent stacks, with more than 400 million SDK connections a month. The architecture change of 28 July — a stateless core plus two extensions, Tasks and MCP Apps — and…
The AI coding productivity paradox: 30% more code, no change in delivery
2026.10.20Source · Industry research Three reports published on the same day, 22 October, reach the same conclusion from different angles: AI coding agents raise both the volume of code and the review workload, but the value actually delivered does not rise in proportion. The numbers do…
Agent security splits into three layers, each becoming a business of its own
2026.10.15Source · Industry watch In October three companies moved at three different layers at once: runtime isolation, protocol gateway protection and automatic vulnerability repair. Locking down agents is no longer a feature — it is a category forming on its own. What each layer covers Runtime…
Treasury agents in production: one protocol connection cut syndicated credit processing time in half
2026.10.08Source · Case review In a joint disclosure on 8 October, commodity trader Centalion and trade-finance platform Komgo gave transaction-level figures: in the off-balance-sheet instrument (OBSI) syndicated credit business, processing time for instrument issuance fell by about 55% and for amendments by about 80%. Note that…
Industrial AI moves from co-pilot to autopilot: quality inspection and predictive maintenance already show ROI
2026.09.28Source · Industry watch The clearest change in industrial AI in 2026 is not stronger models but agents that start doing things: looking up maintenance manuals, matching spare parts in stock, raising work orders in the ERP and adjusting upstream equipment parameters afterwards. In two high-frequency…
E-commerce customer-service agents: from cost centre to profit centre
2026.09.22Source · Industry watch The keyword for e-commerce customer service in 2026 is no longer cost cutting but taking orders. In figures published by Alibaba’s AI shop assistant Dianxiaomi, merchants that adopted it saw the hand-off rate to humans fall by 45% on average, and AI-plus-human…
Small models on proprietary corpora: a 9B model at 98% lower cost scores near the frontier
2026.09.16Source · Industry watch HR and payroll platform Deel released two in-house models on 21 October. One is a 9B open-weights model that scored 70 on the company’s own 601-question benchmark, against 74 for the frontier model Claude Opus 5.5 — but at a median response…
Model filing and AI content labelling: compliance moves from entry ticket to inspection item
2026.08.26Source · Policy analysis As of July 2026, 988 generative AI services in China had completed filing and 598 applications or features had completed registration, with 1,509 large models released; over the same period, first-phase enforcement had handled more than 14,000 non-compliant AI products. Put the…
Turning institutional knowledge into AI-executable assets: five people achieve 2 to 10 times more
2026.08.12Source · Methodology Most companies get stuck in the same place: the model is fine, the tools are there, but the agent does not know how our company does this particular thing. The fix sounds thoroughly unglamorous — write institutional knowledge into structured documents and connect…
Older articles are still being translated. The full Chinese archive is available at 繁體中文 and 简体中文.