This week in agentic AI, the biggest retailer in the world opened its checkout to approved AI agents while a single threat actor used autonomous agents to breach more than 100 corporate sites in under a day. Those two events capture the defining tension in enterprise agentic AI right now: the commercial opportunity is real and accelerating, and the security risk is real and accelerating. Both are happening at the same time, and enterprises that invest in one without the other are building a business case that will not survive its first incident.
In between, Cognizant and Google Cloud committed 18,000 engineers to enterprise agent deployment, Town raised $108 million at a $1 billion valuation for cross-team AI assistants, and Microsoft made agent monitoring a default feature of its enterprise productivity suite. For buyers, the message is that the platform layer is maturing while the governance and security layers are becoming hard procurement requirements. For suppliers, the competitive bar now includes auditability, runtime containment and supply chain assurance, not just model benchmarks.
1. Walmart launches Agent Gateway as Amazon blocks AI shoppers
On 21 October 2026, Walmart CEO John Furner used the Masters of Scale Summit in San Francisco to announce what may be the most significant agentic commerce infrastructure yet deployed at retail scale. Walmart launched an Agent Gateway, an MCP server that allows approved AI agents to search inventory and make purchases on behalf of customers, with dedicated Agent Profiles rolling out next month so shoppers can authorise each agent and set spending limits.
The move is strategically sharp. Where Amazon has blocked AI agents from purchasing on behalf of users, and travel and booking sites have locked agents out, Walmart is building the infrastructure to welcome them. Customers retain control through per-agent spending caps and transaction-level approval gates. Critically, Walmart's returns policy applies even when an agent purchases incorrectly, removing one of the biggest barriers to consumer adoption.
The retailer also published specific terms for AI agents operating on its platform, including requirements that agents cannot access payment information directly and cannot act beyond the shopper's intent. Non-compliant agents will be blocked. This is essentially a terms-of-service contract between a retailer and third-party AI agents, and it is likely to become a model that other retailers copy.
Why it matters: Walmart has just defined a working architecture for agentic commerce at planetary scale: MCP-based access, customer-controlled permissions, agent-specific transaction tracking and codified terms for non-human shoppers. For enterprise buyers, the infrastructure pattern is transferable to procurement, supply chain and B2B commerce. For suppliers, the message is that agent-readiness is becoming a competitive differentiator at the platform level, and organisations that block agents may lose volume to those that integrate with them.
2. One threat actor, 23 hours, more than 100 breaches: the agentic attack playbook arrives
On 21 October 2026, Norforge published an investigation revealing that a single threat actor had used autonomous AI agents to breach John Deere, SharkNinja, Cummins and more than 100 other organisations in just 23 hours. At least one confirmed intrusion entered through a third-party supplier, exposing the supply chain as the weakest link in agentic attack campaigns.
The attack pattern was not exotic. The agents scanned for exposed credentials, connected to live systems and accessed business records using the same techniques that a human attacker would employ, but at machine speed. What would previously have taken weeks of manual reconnaissance was compressed into hours. The campaign demonstrates that agentic systems do not invent new attack vectors. They weaponise existing vulnerability surfaces at a velocity that human defenders cannot match.
Why it matters: this incident moves agent security from a theoretical concern to a documented operational reality. Supply chain exposure is the specific lesson: organisations that secure their own agents but grant third-party agents access without equivalent scrutiny are building defences around the wrong perimeter. For buyers, the practical implication is that vendor security assessments must now include agent-specific questions about runtime behaviour, credential management and containment. For suppliers, expect procurement to add agent security requirements to RFPs within the next quarter.
3. Cognizant commits 18,000 engineers and Google Cloud commits its full AI stack to enterprise agent deployment
On 19 October 2026, Cognizant and Google Cloud announced a significant expansion of their partnership, establishing a Gemini Frontier Capability Group backed by more than 18,000 dedicated Cognizant associates. The group will initially focus on legacy modernisation, developer experience, autonomous business process agentification and frontier cyber defence.
The scale signals intent. More than 55,000 Cognizant associates have completed the Build with Gemini workshop, the largest single Gemini partner workshop held to date. Cognizant is deploying Gemini to 100,000 associates this year, with plans to reach 200,000. The group will use standardised, security-first agentic architecture built on the Model Context Protocol and Google Cloud's Model Armor, producing solutions including a Catalog Support Agent for retail, a Healthcare Agentic Marketplace and multi-agent Order-to-Cash workflows for manufacturing.
Why it matters: this is the services industry's largest dedicated bet on agentic AI deployment. For enterprise buyers, it means that a major systems integrator now offers a structured path from pilot to production with pre-built agentic solutions for specific industries. The risk is that 18,000 engineers accelerating agent deployment without equivalent investment in governance and security creates more exposure points. Buyers should validate that the governance tooling matches the deployment velocity.
4. Town hits $1 billion valuation as cross-team agent coordination becomes a category
On 20 October 2026, San Francisco startup Town announced a $108 million Series B at a $1 billion valuation, co-led by Forerunner Ventures and Menlo Ventures, with participation from Andreessen Horowitz and First Round Capital. Town began as an AI tax preparation service for small businesses and has now built a cross-team AI assistant platform where each employee's Townie can ask a colleague's Townie for information, hand off tasks and share workflows.
Menlo Ventures put its entire firm on Town in June, and paying teams have grown ninefold since July. Town connects to more than 60 enterprise applications including Salesforce, Jira and GitHub, drafts messages by default rather than sending them, and claims SOC 2 Type 2 compliance with no model training on customer data by default. The core proposition is that AI has made individuals faster on their own but projects still stall on handoffs between teams, and Town's agents address that specific coordination gap.
Why it matters: agent-to-agent coordination is emerging as a distinct procurement category, separate from single-agent productivity. For enterprise buyers, the question is whether your organisation's agent architecture supports inter-agent communication, or whether each department's agents operate in isolation. For suppliers, Town's valuation at five months post-launch signals that the market is pricing coordination infrastructure as a premium opportunity, and every agent platform will eventually need to demonstrate how it works with other platforms.
5. Control and governance: Microsoft ships Guardian Agents, Opsin raises $21 million, and regulators sharpen their demands
The infrastructure for governing agentic AI at enterprise scale arrived in several significant forms this week. Microsoft expanded Project Perception with Guardian Agents, dynamic permissions and agent risk monitoring tools integrated into Microsoft 365 E7, providing independent oversight of corporate AI agents and real-time containment capabilities. VS Code 1.137 introduced an Agent Host system that runs AI coding agents within isolated Dev Containers, bringing agent sandboxing to the developer workflow.
On the funding side, Opsin secured $21 million co-led by SYN Ventures and Race Capital to build a standardised agent governance platform that maps ownership, permissions and runtime behaviour across Microsoft 365 Copilot, Claude Enterprise, ChatGPT Enterprise and Gemini. Its Risk Graph learns continuously from agent activity and can quarantine agents, block actions or alert owners with remediation guidance. The company argues that within three years, agents will interact with one another far more than with people, and that governing them requires treating them as accountable entities rather than restricting their speed.
Separately, US banking examiners have begun demanding decision logs, approval gates and named accountability owners for every AI agent deployment, with banks that cannot produce this evidence facing supervisory findings. The UK Information Commissioner's Office has opened a six-week call for evidence on agentic AI data protection risks. And the Cloud Security Alliance published guidance on a two-gate model that separates tool access authorisation from per-resource use authorisation.
Why it matters: the governance layer is no longer optional or aspirational. Regulators are writing the rules now, and the tools to comply are arriving in parallel. For enterprise buyers, the checklist is forming: decision logs, named accountability owners, agent-specific access controls, runtime containment, behavioural testing before deployment, and incident reporting for agent failures. For suppliers, each of these requirements is a product feature. The vendors that ship them first will own the procurement conversations in 2027.
The Agentic Expo takeaway
This was the week that agentic commerce became real, agentic attacks became measurable, agentic services became industrialised and agentic governance became enforceable. Walmart and the 100-plus-site breach represent opposite ends of the same reality: agentic AI is now a live, operational system interacting with money, supply chains, customer data and corporate infrastructure at machine scale.
For enterprise buyers, the practical agenda has three items. First, treat agent security as an operational discipline, not a future concern. Second, evaluate whether your agents can coordinate across teams and with external partners, because the competitors that coordinate will move faster than those that operate in departmental silos. Third, begin mapping your agent deployments against the governance requirements that regulators are already writing into examination frameworks.
For suppliers, the window to demonstrate production-grade governance, security and coordination is open now. The platforms that ship these capabilities as defaults rather than extras will set the standard for the next phase of enterprise adoption. The platforms that do not will find themselves answering procurement questions they cannot satisfy.
Sources: TechCrunch on Walmart Agent Gateway; AI Governance Weekly on agent breach campaign; Cognizant press release on Gemini Frontier Capability Group; Tech Funding News on Town Series B; FinTech Global on Opsin funding; AI Agents Directory on Microsoft Guardian Agents, VS Code 1.137, AuxoAI; AI Agent Store weekly roundup.