This was the week that agentic AI moved from boardroom slides to employee desktops. Uber became the largest single-company agent deployment on record, putting ChatGPT Work in front of all 30,000 employees. The EU AI Office launched its first enforcement action. Dell showed that on-premises agent infrastructure is no longer a future ambition but a live product with more than 6,500 enterprise users. And SK Telecom published a model for treating AI agents as a managed workforce with employee IDs, supervisors and KPIs.

For enterprise buyers, the signal is that the infrastructure, governance and deployment patterns are all maturing at the same time. The question is shifting from "should we deploy agents?" to "how do we manage them, where do they run, and who is accountable when something goes wrong?" For suppliers, the weeks of launching an agent platform and calling it done are over. Buyers now expect identity, perimeter controls, observability and audit logs as defaults, not as roadmap items.

1. Uber deploys ChatGPT Work to all 30,000 employees

On 22 October 2026, Uber and OpenAI announced the largest single-company agentic AI rollout on record. Following a 1,000-person pilot, ChatGPT Work was extended to Uber's entire 30,000-strong workforce. The platform, powered by GPT-6.1 Sol, executes multi-step tasks across the tools employees already use, from messaging and calendars to enterprise systems.

The detail that matters most is who adopted it: roughly 80 per cent of the pilot's early adopters were non-engineers, according to Uber's VP of engineering, Sumanth Sukumar. That figure reverses the standard enterprise AI narrative in which technical teams lead and business teams follow. At Uber, the business teams adopted first, and the engineering teams are now building on top of that foundation.

Uber also confirmed that the deployment connects to the company's internal data systems, giving agents context about the business rather than generic answers. For any enterprise still treating agentic AI as a developer productivity tool, Uber's deployment pattern suggests the real return comes when agents understand the business, not just the codebase.

Why it matters: a 30,000-person deployment with non-engineer majority adoption is a reference case for every enterprise considering a company-wide rollout. The key procurement question: does the platform connect to the systems your business teams actually use, or does it require them to learn a new interface?

2. Dell takes agents on-premises with expanded AI Factory

At NVIDIA GTC Berlin this week, Dell announced a significant expansion of its AI Factory with NVIDIA, including new rack-scale systems, an air-cooled two-socket server and a workstation built on NVIDIA's GB10 platform. The offering includes validated software and deployment templates purpose-built for agentic workloads, and Dell confirmed that more than 6,500 enterprises are already using the platform as they move AI projects from experimentation toward production.

The expansion runs across deskside systems, data centres and on-premises environments, giving regulated or latency-sensitive operators a full alternative to sending every agent task to a public cloud. This is not a proof of concept. It is a shipping product with thousands of enterprise customers.

A related signal came from NVIDIA CEO Jensen Huang, who discussed the possibility of a recurring annual software licence of roughly $2,000 per GPU for the software used to run AI agents. NVIDIA has not confirmed which products would be covered or when the fee would begin, but for enterprise buyers modelling total cost of ownership, it is a reminder that agent infrastructure costs extend beyond model inference and hardware depreciation.

Why it matters: the choice for enterprise buyers is no longer "cloud or nothing." On-premises agent infrastructure now comes with reference architectures and pre-validated software stacks. The financial model question is whether per-GPU software licensing becomes a material line item, and whether alternative infrastructure providers offer comparable agent-runtime controls without the per-unit charge.

3. The EU AI Office launches first enforcement action, and regulators turn their attention to agents

In a significant milestone, the EU AI Office launched its first enforcement action this month, sending requests for information to model providers on security, external evaluation and model monitoring. Several AI Act obligations were already in application by 28 October 2026, including provisions that directly affect companies deploying autonomous agents. The first harmonised standard under the Act has now been published, giving enterprises a concrete compliance benchmark rather than a moving target.

At the same time, regulators globally have begun to turn specifically toward AI agents as a distinct category. US banking examiners are 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 call for evidence on agentic AI data protection risks.

A detailed analysis published on HackerNoon this week maps how the EU AI Act's risk classifications apply to agentic AI: the higher the agent's autonomy and the greater its impact on individuals, the stricter the obligations. Low-risk agents face transparency requirements. High-risk agents, including those used in employment, credit decisions and critical infrastructure, must meet conformity assessments, human oversight mandates and detailed documentation standards.

Why it matters: the regulatory framework for agentic AI is no longer theoretical. Enterprise buyers deploying agents in the EU or in regulated sectors must now demonstrate compliance with specific, enforceable requirements. Suppliers that build audit trails, decision logs and human oversight controls into their platforms as defaults will have a structural advantage over those that treat compliance as an afterthought.

4. SK Telecom publishes a workforce model for AI agents, and enterprise platforms add business context

SK Telecom published its first AI Governance Report this week and set out what may become the standard model for managing AI agents at scale. The South Korean carrier assigns every AI agent an employee identification number, records each agent's role and supervisor, monitors operating status, token use and work history, and can suspend or retire agents when performance reviews show they are no longer needed. Agents with greater autonomy or business impact are required to have a senior human supervisor such as a team leader or executive.

The model is significant because it treats agents not as tools but as accountable entities within an organisation's workforce. For any enterprise moving beyond pilot programmes, the practical implication is clear: every production agent needs an owner, a permission set, an activity log, an escalation path, and a documented retirement procedure.

On the platform side, SAP announced new capabilities across Joule Studio, Joule Work and SAP Business AI Platform that give agents trusted information about business entities, relationships, processes and policies through data-fabric and ontology capabilities. The insight is straightforward: enterprise agents fail most often because they lack reliable context, not because the underlying model cannot generate text. SAP's approach focuses on connecting agents to governed business data rather than optimising generic prompts.

Persona introduced supervised agents for identity operations, with its Case Review Agents gathering evidence, checking it against an organisation's policies and recommending decisions with visible reasoning. The design keeps consequential decisions under human control while automating evidence collection and routine analysis, a pattern applicable to any high-stakes workflow from compliance to claims processing.

Why it matters: the workforce model and the business-context approach represent two halves of the same requirement. Agents need to understand an organisation's actual data and policies, and they need to be managed like accountable actors, not disposable scripts. For buyers, the procurement checklist now includes both dimensions: does the platform connect to governed business data, and does it provide the audit and supervisory controls to treat agents as accountable entities?

5. Infrastructure and security: Daytona raises $60 million, Akamai launches Agent Sandbox, and Cisco builds a cybersecurity benchmark for agents

Daytona Platforms announced a $60 million Series B led by FirstMark Growth at a $1 billion valuation, bringing total funding to $91 million. The company builds cloud infrastructure purpose-built for agent workloads: short-lived, stateful sandboxed environments with GPU access across Linux, Windows and macOS. The funding validates the thesis that agent infrastructure requires a different operating model from conventional web services, and that general-purpose clouds may not be the optimal home for every agent workload.

Akamai entered the agent infrastructure market with two beta products: Akamai Inference, a managed model-serving service, and Akamai Agent Sandbox, an isolated runtime for executing agent-generated code and workflows with default outbound blocks and isolation controls. Positioned on Akamai's global edge cloud, the products target organisations that need to run models close to users and data while maintaining strict containment.

On the observability front, Cisco introduced Agentic FAITH, an extension of its FAITH evaluation framework for multi-step cybersecurity tasks in which an agent chooses tools, interprets results and acts through multiple stages. Its first integrated benchmark, VulnLoc, tests whether agents can locate vulnerable files in a code repository while avoiding false positives on patched code. The framework addresses what is becoming a universal requirement: buyers and security teams need repeatable measurements not just for what an agent concludes but for how it reached that conclusion.

Google Cloud made generally available improvements to VPC Service Controls, including folder-based perimeter membership and more granular exceptions, specifically positioning the changes as a way to scale autonomous agents while keeping data inside enterprise security boundaries. The update reflects the operational reality that agents frequently request data across services and projects, making conventional network rules increasingly difficult to manage.

Why it matters: each of these developments addresses a different part of the same infrastructure challenge. Daytona handles the execution environment. Akamai handles the network edge and sandboxing. Cisco handles evaluation. Google Cloud handles network boundaries. For enterprise buyers, the message is that the full stack for running agents in production now exists. The procurement task is to map each layer against your organisation's specific requirements and ensure no layer is left unaddressed.

The Agentic Expo takeaway

This was the week that agentic AI became an enterprise workforce reality, not just a laboratory experiment. Uber's 30,000-employee deployment, Dell's on-premises reference architecture with thousands of enterprise customers, the EU's first enforcement action and SK Telecom's workforce management model for agents each represent a milestone in a different dimension of agent maturity: adoption, infrastructure, regulation and operations.

For enterprise buyers, the practical agenda has four items. First, treat agent deployment as a workforce management exercise, with ownership, supervision, performance review and retirement criteria. Second, evaluate whether your infrastructure strategy supports the workloads your agents will actually run, including on-premises requirements and sandboxed execution. Third, ensure your compliance programme covers the specific requirements the EU AI Act now enforces, particularly around agent autonomy, transparency and human oversight. Fourth, demand observability that shows not just what agents produce but how they produce it.

For suppliers, the bar has risen. Enterprise buyers are no longer asking "can it work?" They are asking "can we run it, secure it, audit it and explain it to a regulator?" The suppliers that answer those four questions with product features rather than roadmap slides will define the procurement conversations through the end of 2026 and into Agentic Expo in March 2027.

Get Tickets More News

Sources: TechCrunch, "Uber puts AI agents on every employee's desktop with OpenAI", 22 October 2026; Yahoo Finance, "Uber Rolls ChatGPT Work to All 30,000 Employees", 22 October 2026; DataStudios, "Uber Deploys ChatGPT Work to 30,000+ Employees", 25 October 2026; AI Agent Store, "AI Agents News — Week of October 29, 2026"; Pearl Cohen, "EU AI Act's First Harmonized Standard Published as EU AI Office Takes First Enforcement Actions", October 2026; HackerNoon, "How the EU AI Act Regulates Agentic AI", October 2026; NeoTeo, "EU AI Act Rules for AI Agents", October 2026; TechCrunch, "From 8VC to Benchmark, Here's What's Hiding in October SEC Filings", 27 October 2026; Bloomberg, "Microsoft Alumni Score Funding to Build AI Agents for Finance", 27 October 2026.