Summary

The day's signals cluster around agents becoming the primary way software is accessed rather than an add-on. Salesforce unveiled AIforce, a headless toolkit that exposes CRM data, logic, permissions, and governance to external agents like Claude over MCP, and Google opened an MCP server for Google Home so any MCP-capable agent can read device state and control smart-home hardware — pushing MCP from developer tooling into both the enterprise system-of-record and consumer devices. Agentic AI also moved into operations as Riverbed embedded its Riverbed IQ agent and conversational Q interface into network observability at no premium. On the infrastructure side, Cloudflare shipped a trio of developer-platform upgrades (R2 Data Catalog table maintenance, Hyperdrive database connectivity for Python Workers, and RPC session spans in Workers tracing), while LiteLLM added cosign-signed images to harden the LLM-gateway supply chain and Ollama removed its built-in CLI agent, sharpening the line between model-serving runtimes and the agent layers built on top.

Key themes

  • MCP is consolidating as the default interoperability layer for agents, reaching both the enterprise system-of-record (Salesforce AIforce) and consumer hardware (Google Home) as first-party platforms expose governed data and controls to any MCP-capable agent.
  • Agentic AI is expanding beyond coding and chat into operational tooling — Riverbed embedded an agentic assistant into network observability and priced it at no premium, signaling agent-driven troubleshooting as a baseline expectation.
  • Cloudflare continued a broad developer-platform push, adding lakehouse maintenance visibility (R2 Data Catalog), Python Workers database connectivity (Hyperdrive), and cross-boundary RPC tracing (Workers observability) that together strengthen data and distributed-agent workloads.
  • Gateway and runtime governance is a growing battleground: LiteLLM added cosign image signing for supply-chain trust, while Ollama shed its built-in agent loop to refocus as a serving runtime — clarifying the split between model serving and agent orchestration.

Notable items

  • Salesforce launched AIforce, a headless toolkit exposing CRM data, workflows, permissions, and governance to external AI agents through MCPs, APIs, plug-ins, and skills, with an AgentExchange ecosystem spanning Anthropic, AWS, Google, and Microsoft (high impact).
  • Google opened early access to a Model Context Protocol server for Google Home, letting agents like Claude and ChatGPT read device state, review camera and activity summaries, and control connected devices in natural language (high impact).
  • Riverbed launched Network 360, natively embedding its Riverbed IQ agentic AI and conversational Riverbed Q interface into AppResponse and NetProfiler at no premium over current pricing.
  • Cloudflare R2 Data Catalog added a Maintenance tab with compaction and snapshot-expiration schedules, a paginated maintenance audit log, and on-demand compaction queueing, moving it closer to managed Iceberg catalogs.
  • Cloudflare Hyperdrive gained PostgreSQL and MySQL connectivity for Python Workers, bringing pooled, cached database access to Python serverless apps at parity with JavaScript Workers.
  • Cloudflare Workers observability now records JavaScript RPC session spans, so traces follow calls across Worker boundaries and into Durable Objects — useful for multi-Worker and agent service-binding designs.
  • LiteLLM began publishing cosign-signed Docker images so operators can verify image provenance before deploying, alongside fixes to team-level rate limiting and budget controls (medium impact).
  • Ollama merged a breaking change removing its built-in CLI agent feature while adding reasoning_level metadata via /api/show and GPU memory budget enforcement, refocusing on model serving over its own agent loop (low impact).

Source coverage

Source rows used: 8