Agents

Databricks and OpenAI just put frontier models inside the data plane. Here is why Agent Bricks now looks like a data-native factory, how Mosaic evals and Tecton features change the game, and what CIOs should do first.

Announced at AWS Summit New York on July 16, Amazon Bedrock AgentCore bundles Runtime, Memory, Identity, Gateway, Browser, Code Interpreter, and Observability to turn agent prototypes into scalable production systems.

Salesforce’s Agentforce 3 pairs a real Command Center with MCP-native interoperability and FedRAMP High authorization, making observability, governance, and reliability the new table stakes for enterprise AI teams.

Google’s AP2 gives AI agents a rulebook to prove intent, identity, and spend limits at checkout. See how mandates and verifiable credentials enable card, bank, and stablecoin payments you can audit today.

OpenAI has turned ChatGPT from a chat box into a doer. Agent mode opens a virtual computer that browses, runs code, edits files, and delivers finished work. Here is how to roll it out safely.

Microsoft's new Security Store is a governed marketplace for cybersecurity agents. SOC teams can build no code Security Copilot agents and deploy vetted partner agents inside their Microsoft environment.

GitHub has taken its pull request centric Copilot coding agent to general availability. Learn how it works, what guardrails it enforces, and a practical 90 day plan to pilot agent teammates without rewriting your stack.

Opera Neon reframes the browser as a runtime for AI agents, with local-first, in-page execution and new workflow primitives that let the web act. See how agentic browsing reshapes design, security, and go-to-market.

Microsoft’s Frontier rollout brings Agent Mode to Excel and Word, plus Office Agent in Copilot Chat. Learn what shipped, what is missing, and how to pilot real agent workflows in Microsoft 365 starting today.

ServiceNow’s Zurich release makes the Now Platform agent native with Build Agent, developer sandboxes, and agentic playbooks. Learn what shipped, how governance works, and a 30 day plan to ship real outcomes.

Anthropic’s Claude Sonnet 4.5 adds what enterprises lacked: sustained, reliable autonomy. With runs that stay on task for about 30 hours, agents can own outcomes across complex, multi day workflows.

Google Cloud just made agent interoperability practical for enterprises. With Agent Engine, A2A, tracing, connectors, and streaming, teams can ship multi agent workflows across vendors with real controls and speed this quarter.

Databricks unveiled Mosaic AI Agent Bricks and MLflow 3.0 with built-in evaluations, tracing, and governance. Learn why this stack changes production agents and how to build and ship across AWS, Azure, and GCP.

Cloudflare’s new Agents SDK pairs with remote MCP servers, Durable Objects and Workflows to move agents from laptop demos to governed, low-latency production. See what shipped and how to deploy it with confidence.

On September 25, 2025, Okta introduced Okta for AI Agents and Cross App Access, extending OAuth so enterprises can centrally authorize agent-to-app access. See what changed, who backs it, and how to roll it out.

Figma’s remote MCP server lets agents and IDEs query components, tokens, styles, and usage rules directly from your design system. Expect cleaner handoffs, faster reviews, and automation that follows your standards.

On July 17, 2025, OpenAI turned chat into computer use. ChatGPT Agent researches, clicks, and acts inside a virtual computer with narration and approvals. Explore strengths, limits, and how to put it to work.

GitHub Copilot’s coding agent is now GA and ready to own end to end tasks. It plans changes, runs tests in isolated runners, opens draft PRs, and works within branch protections and review gates.