This week in agentic AI brought three signals that matter more than any product launch. First, the gap between deploying agents and capturing value from them is now quantified, and it is larger than most boards would tolerate. Second, the unit economics of agentic decisions just shifted in a way that will accelerate adoption precisely where enterprises have hesitated. Third, UK regulators closed the loophole that might have let organisations blame their agents for harm.
Alongside those structural shifts, IBM and Microsoft launched a platform with 6,000 pre-built agents and forward-deployed engineers, Google shipped a purpose-built decision model at disruptive pricing, and cybersecurity M&A proved that agent-specific security is now a standalone investment category. For enterprise buyers, the message is that 2026 ended months ago in terms of capability, but it is only now beginning in terms of governance, economics and liability. For suppliers, the question is whether your roadmap covers what buyers now need, not what was exciting at the start of the year.
1. Genpact measures the autonomy gap: 79% in production, 15% tied to P&L
On 12 November 2026, Genpact published "Autonomy in Practice," a research report based on a survey of more than 2,000 enterprise executives across 17 industries. The headline figure is stark: 79% of enterprises have AI agents in production, but just 20% use them broadly and consistently across functions, and only 15% have fully tied AI investment to P&L outcomes.
The report frames the problem as operational, not technological. On average, just 49% of AI initiatives funded in the past 24 months have delivered the value expected. Only 9% of organisations have completed significant operating model transformation. Yet AI now leads or fully runs 40% of decisions, more than double a year ago, and humans still decide or audit in at least 87% of cases at every decision stage. The workers are not disappearing: 92% of respondents with agents in production kept their human workforce the same size or grew it.
Genpact's diagnosis is that enterprises have built agent capability faster than they have redesigned how they operate. The autonomy dividend, the value that comes when AI-enabled gains translate into better business performance rather than simply greater productivity, requires operating model change that most organisations have not attempted.
Why it matters: This is the first large-scale survey to separate agent production from agent value, and the gap is uncomfortable. For enterprise buyers, the implication is that buying more agents will not close the value gap if the surrounding processes, metrics and accountability structures stay the same. For suppliers, the procurement conversation is shifting from "how many tasks can your agent perform?" to "how do we redesign our operating model so those tasks drive P&L outcomes?" The platforms that can answer the second question will win the renewals that the 51% of disappointed buyers would otherwise cancel.
2. Google ships the Gemini Decisions API: faster, cheaper agent decisions
On 11 November 2026, Google launched the Gemini Decisions API, powered by the new Gemini Decision 1 model. The product is designed for the small, rapid choices that agents make thousands of times per workflow: tool selection, policy checks, model routing and context grading.
Instead of generating text token by token, Gemini Decision 1 scores candidate options in a single forward pass, returning probabilities and confidence scores with 150 to 250 milliseconds end-to-end latency. It handles up to 1,000 independent yes/no, multiple-choice or rubric-scoring questions per request across text, images, audio, video and PDFs. The pricing is $0.10 per 1 million input tokens, with zero output tokens billed. Integrations are already available with LangChain, OpenRouter, Vercel, LiteLLM and Pydantic.
On benchmarks spanning agent routing, reasoning, ranking and multimodal understanding, Google reports that Gemini Decision 1 leads the suite with the lowest calibration error. That matters because poorly calibrated confidence scores are a major source of agent failure in production: an agent that is 95% confident and wrong causes more damage than one that is 50% confident and defers to a human.
Why it matters: The cost and latency of agentic decisions have been quiet blockers on enterprise adoption. When every tool call, policy check and routing choice requires a full language model inference, the economics of high-frequency agents break down. Google's move prices those micro-decisions at a level that makes them viable for real-time workflows. For enterprise buyers, the question is now: which of our agent workflows are currently too expensive to run continuously, and could a purpose-built decision model unlock them? For suppliers, the bar is rising. Platforms that route every decision through a general-purpose model will increasingly look expensive next to those that use lightweight decision models where appropriate.
3. UK regulators: your agent, your blame
On 30 October 2026, the UK Competition and Markets Authority, Financial Conduct Authority, Information Commissioner's Office and Ofcom issued a joint statement, "Protecting people and businesses in an agentic economy." The message, published alongside revised consumer law compliance guidance, is unambiguous: "An AI agent's autonomy does not diminish the responsibility of organisations that deploy it to comply with applicable law and relevant regulation."
The statement makes clear that existing competition, consumer and data protection law apply to AI agent deployments without exception. The CMA's accompanying research paper, "Agentic AI and consumers," explores how autonomous agents making purchasing, pricing and service decisions on behalf of consumers could distort markets if not governed properly. Separately, the UK's Cyber Security and Resilience Bill, currently in Parliament, includes a clause that would allow ministers to make regulations governing incidents in which an organisation loses control of an AI agent, including authority for searches, seizures and inspections.
For UK enterprises deploying agents that interact with customers, process personal data, set prices, or make purchasing decisions, the legal framework is no longer theoretical. Regulators have explicitly rejected the argument that agent autonomy creates a liability gap.
Why it matters: This is the first joint regulatory statement in any major market to address agentic AI liability in such direct terms. For enterprise buyers in the UK, the compliance checklist for agents now includes the same consumer protection, competition and data protection obligations that apply to human employees. For suppliers, the message is equally direct: if your agent causes harm, the deploying organisation is liable, and the deploying organisation will soon have ministerial enforcement powers applied to its response. Platforms that cannot produce audit trails, explain decisions and demonstrate control will fail procurement in regulated sectors. Those that can will find the compliance conversation becomes a sales accelerator rather than a blocker.
4. IBM and Microsoft launch Imagine Enterprise with 6,000 pre-built agents
On 10 November 2026, IBM and Microsoft announced Imagine Enterprise, an agentic platform that combines more than 6,000 pre-built AI agents and assets from IBM Consulting Advantage with Microsoft's Azure, Entra, Purview, Defender, Fabric and 365 stack. Forward-deployed engineers from both companies are available for qualifying engagements to help move clients from experiment to production. SEI, the financial technology and asset management services provider, is already exploring the platform.
The architecture is designed for regulated industries, with built-in governance, observability and AI-compliance controls. IBM's Process Studio, Context Studio and Advantage Core handle agent configuration and oversight, while Microsoft Foundry, Copilot Studio and GitHub Copilot coordinate development and orchestration. The 6,000 figure includes agents, frameworks and assets rather than exclusively deployable agents.
Why it matters: The partnership signals that the enterprise AI market has moved past the single-vendor platform phase. Buyers now expect industry-specific agents, integration with existing security stacks and implementation support rather than standalone software licences. For suppliers, the competitive landscape now includes not just other agent platforms but the combined consulting, engineering and infrastructure resources of the largest enterprise technology companies. The platforms that compete in this tier will need deep industry expertise and implementation velocity, not just model performance.
5. Palo Alto Networks pursues Clutch Security: agent security becomes standalone M&A
Palo Alto Networks is in talks to acquire Clutch Security, a three-year-old Israeli startup that tracks and secures the credentials AI agents use to access enterprise systems. The reported price is $500 million in cash plus milestone-based payments. Clutch was last valued at $85 million in its January 2025 Series A, and is approaching $10 million in annual recurring revenue. If completed, the deal would value Clutch at roughly 50 times ARR and mark Palo Alto Networks' seventh acquisition in 2026.
The deal follows a pattern. Palo Alto Networks acquired Console in September for automated agent security testing. Earlier in the year it bought multiple AI security startups. The strategic thesis is consistent: the attack surface created by AI agents moving at machine speed, with their own identities and access credentials, requires security tooling that traditional perimeter defences cannot provide.
Why it matters: Agent-specific security is now a standalone investment category with real M&A valuations. For enterprise buyers, the signal is that protecting what agents can access has become at least as important as protecting what humans can access. For suppliers, the security conversation is no longer a nice-to-have. Buyers will ask whether your platform works with their existing identity and access management systems, whether agent credentials can be rotated and revoked centrally, and whether agent activity is observable in the same security dashboards that monitor human users. Platforms without credible answers will lose deals.
The Agentic Expo takeaway
Three themes defined the week. First, the value gap: deploying agents is not the same as extracting value from them, and the enterprises that close that gap fastest will be the ones that redesign operating models, not just add AI tools. Second, the cost shift: purpose-built decision models are about to make high-frequency agent workflows economically viable at scale. Third, the liability shift: regulators have closed the autonomy loophole, and the organisations deploying agents are the organisations accountable for them.
For enterprise buyers, the practical checklist now has five items. Understand where your agents are in production and whether they are mapped to business outcomes. Evaluate whether your current agent infrastructure uses the right model for each decision type, or over-pays for general-purpose inference on simple routing calls. Confirm that your legal, compliance and security teams have reviewed agent deployments against existing consumer protection, competition and data protection obligations. Document your agent inventory, permissions and activity logs with the same rigour you apply to human employee access. And treat agent security as a distinct discipline, not an extension of traditional perimeter controls.
For suppliers, the competitive questions have changed. Buyers are no longer asking "what can your agent do?" They are asking "can your agent be trusted with regulated data, governed with existing tooling, and held to account when something goes wrong?" The platforms that ship credible answers will define the next phase of enterprise agentic AI.
Sources: Genpact, "Autonomy in Practice: How agentic operations can turn AI progress into business value", 12 November 2026; Genpact via PR Newswire, "79% of Enterprises Report AI Agents in Production, but Only 15% Have Fully Tied AI Investment to P&L Outcomes", 12 November 2026; Google, "Gemini Decisions API: Fast, calibrated multimodal decisions", 11 November 2026; UK Competition and Markets Authority, "Agentic AI and consumers", 30 October 2026; UK Digital Regulation Cooperation Forum, "Protecting people and businesses in an agentic economy", 30 October 2026; IBM, "IBM and Microsoft announce a new AI agent platform designed for industry-specific transformation", 10 November 2026; Microsoft, "IBM and Microsoft announce Imagine Enterprise", 10 November 2026; TechCrunch, "Palo Alto Networks in talks to buy AI agent security startup Clutch for $500M", 5 November 2026.