AI & Automation News

In a SiliconANGLE opinion column, Intellyx founder Jason Bloomberg analyzes the risks of enterprise AI agents, noting that their reliance on large language models makes them inherently nondeterministic. While this unpredictability grants them agency, it also introduces failure risks that make them ill-suited for tasks demanding absolute precision, such as payroll, or high-stakes operations where errors cause catastrophic harm. Bloomberg advises organizations to adopt an SRE-inspired error budget to verify whether projected business benefits substantially outweigh failure costs before deploying agents.

AWS has introduced Amazon Bedrock AgentCore payments, enabling artificial intelligence agents to pay for services such as model inference on demand. The service handles payments over the x402 protocol using embedded wallets and enforces budget caps directly at the infrastructure layer, ensuring agents cannot exceed spending limits even if prompts are manipulated. Incarna integrated the system to pay the BlockRun inference router on Base mainnet using USDC, completing the deployment in three days with roughly 200 lines of code. Across the beta phase, agents processed over 1,000 individual microtransactions.

A Salesforce article outlines the no-code AI landscape for small and medium-sized businesses (SMBs). The publication notes that three out of four small businesses are investing in AI, yet 88% remain in the exploration phase. It highlights no-code tools across sales, marketing, workflows, and service, emphasizing that a connected platform delivers greater returns than fragmented point solutions. Key evaluation criteria include setup speed, data connectivity, AI quality, security, and scalability.

Google has announced the launch of the Google Cloud CLI remote Model Context Protocol (MCP) server in preview, expanding its ecosystem of managed remote MCP servers. Powered by the gcloud and bq command-line tools, the server enables AI agents to execute cloud infrastructure management tasks and advanced BigQuery workflows directly in an isolated Google Cloud sandbox without requiring local binary installations. The solution incorporates enterprise-grade security features, including zero ambient credentials, IAM policy enforcement, Model Armor screening, and Cloud audit logging. The server exposes two core tools—run_gcloud_command and run_bq_command—and is available in public preview at no additional charge for the MCP server itself.

n8n has announced the launch of Agents, a new capability allowing users to build autonomous AI agents alongside existing workflows. Configured using plain language, agents independently determine the necessary execution steps while utilizing workflows, MCP servers, and built-in integrations as tools. They can be triggered via Slack, Telegram, Discord, on fixed schedules, or directly called inside existing workflows via the new Message an Agent node. Currently available in Preview on n8n Cloud and self-hosted instances, conversational turns with agents share the platform's standard execution quotas.