This week in agentic AI, the enterprise platform layer took a decisive step forward while the security and knowledge infrastructure struggles to keep pace. Google turned Gemini from a conversational model into a fully agentic system capable of planning, delegating and completing tasks across enterprise software. Rein Security raised a $25 million Series A, confirming that runtime security for agents is now a standalone venture category rather than a feature request. And new research from MIT Technology Review and Neo4j found that only around a third of agentic AI projects make it into production, with a lack of organisational knowledge, not model capability, cited as the primary blocker.
Four threads stood out. First, Google's agentic Gemini launch signals that the platform war is now about task completion, not model performance. Second, the security funding wave, led by Rein Security and Temporal's expansion into agent controls, confirms that trust and governance are becoming the primary buying criteria. Third, the knowledge infrastructure gap, documented in a survey of 300 enterprise technology executives, explains why so many pilots stall before production. Fourth, ancillary developments from Salesforce, GoodData and Securonix show that the agentic stack is deepening across data, analytics and threat detection layers.
The message for buyers is consistent with the past month: the capabilities are arriving faster than the controls and the data foundations. For suppliers, differentiation is shifting from what your agent can do to whether it can be governed, observed and trusted inside an enterprise environment.
1. Google turns Gemini into an agent: objectives over instructions
On 8 October 2026, at its Gemini at Work conference, Google announced that Gemini is now an agent. The update moves the platform beyond question-answering and summarisation into proactive task completion. Users can assign objectives rather than single instructions, and the agent plans the work, uses custom skills and tools, and connects to internal business systems to deliver completed outcomes.
The agent can connect to Google Workspace, Microsoft 365, Slack, Jira, Confluence, Git, BigQuery, Databricks, Postgres, Snowflake and other systems, as well as any Model Context Protocol (MCP) server inside or outside a company's network. A tasks inbox interface lets users track what the agent is doing, including its reasoning process, delegation to sub-agents, code execution and progress. The AI has its own Workspace account, email address and organisational context, meaning it knows team structures, time zones, approval chains and calendars.
Google is rolling the agent out to enterprise customers first, before a wider consumer release. Sundar Pichai noted that nearly 90% of Fortune 100 businesses already use Gemini Enterprise, and that the company wants to solve the harder problems around security, scale and performance in a controlled environment first. Early testers include On, Shopify and PayPal, and enterprise customers include BNP Paribas, Bradesco, Merck, Orange Spain and Wesfarmers.
Thomas Kurian, CEO of Google Cloud, described the shift as follows: "You give it objectives, not just instructions. It plans the work, uses custom skills and tools, connects to your systems, and brings back something finished." The company also introduced flexible spending options, including multi-model orchestration, smart routing and real-time spend caps, aimed at giving enterprises more predictable cost control.
Why it matters: Google is now the fourth major platform, after OpenAI, Salesforce and Microsoft, to ship a unified agentic layer. For buyers, this means the choice of agent platform is no longer a theoretical question. It is a procurement decision with switching costs that will compound over time. For suppliers, building on top of Gemini, OpenAI, Salesforce or Microsoft will require navigating different governance models, identity systems and billing structures. The platform market is fragmenting just as the capability market is converging.
2. Rein Security raises $25 million: runtime security for agents becomes a standalone category
On 8 October 2026, Rein Security announced a $25 million Series A round, co-led by Glilot Capital and Sienna Venture Capital, bringing its total funding to $35 million. The company provides an AI-native runtime security platform that operates at the point where agents execute actions, giving security teams visibility into every line of code an agent runs and every resource it touches.
Rein's platform is built on patented sidecar technology that works without routing company or customer data through a gateway or proxy. It provides real-time guardrails that block harmful actions, secures the agent supply chain, and addresses both internal agent misconfigurations and external AI-driven attacks. According to Gartner, the market for securing AI is projected to reach nearly $4.8 billion in 2027, up 68.7% from 2026.
The funding announcement included a case study from Black Hat USA 2026, where Rein's research team compromised the AI shopping agent of a top-five U.S. retailer. In another example, a global enterprise's onboarding agent opened a PDF containing a hidden prompt injection that directed the agent to act outside its intended role. Rein detected and blocked the action before any damage was done.
Why it matters: the emergence of dedicated, well-funded runtime security vendors for agents is a structural signal. It means that enterprises are deploying agents in production at a scale where traditional security tools are insufficient. Buyers should evaluate agent platforms with runtime security as a non-negotiable requirement. Suppliers without a clear runtime security story will face longer procurement cycles and higher risk assessments from enterprise buyers.
3. Only 34% of agentic AI projects reach production. The blocker is knowledge, not models.
A report published by MIT Technology Review Insights on 5 October 2026, in partnership with Neo4j, surveyed 300 data, AI and technology executives and found that on average only 34% of organisations' agentic AI projects make it into production. Even high-technology firms struggle with this. The primary causes are legacy data systems, security and privacy concerns, and a lack of organisational knowledge and context.
The report distinguishes between three types of knowledge: semantic knowledge (what data means in context), episodic memory (what the organisation has learned from past interactions), and procedural knowledge (how work gets done). A small group of "production leaders," where 61% of projects advance beyond pilot, showed significantly stronger capabilities in all three areas, particularly semantic knowledge. Data fragmentation was cited as a top challenge by 55% of respondents.
Executives expect the biggest impact on agent quality to come from strengthening the structural foundation between organisational data and AI agents. Investment priorities include retrieval technologies, AI-ready APIs, retrieval-augmented generation, evaluation agents and knowledge graphs.
Why it matters: the report reframes the agentic AI production problem from "the models are not good enough" to "the data and knowledge layers are not ready." For buyers, this means that agent procurement should be preceded by a data and knowledge audit. For suppliers, offering a model without a knowledge integration strategy is now a known path to pilot failure. Vendors that bundle data preparation, knowledge graphs and agent runtimes into a single stack are likely to outperform point-solution agents in enterprise sales cycles.
4. Temporal expands into agent security; GoodData targets the agent data layer
This week also brought two infrastructure developments that deepen the agentic stack. Temporal, valued at $12.55 billion after a $550 million funding round last month, hired the team from New York startup Oso to build AI agent security controls into its enterprise infrastructure platform. The move signals that workflow orchestration and security governance are converging into a single runtime requirement for production agents.
GoodData.AI announced the Agentic Serving Plane on 6 October 2026, a governed execution layer designed to serve high-concurrency, context-rich data workloads for AI agents. The product focuses on keeping metrics consistent, enforcing access controls and providing timely answers when agents turn one question into hundreds of underlying queries.
Meanwhile, Securonix introduced Advanced Behavioural Analytics on 8 October 2026, adding the ability to detect both human and AI-driven threats at machine speed. The product identifies activity that is technically permitted but operationally suspicious, a challenge that becomes acute when AI agents have broad system access.
Why it matters: the infrastructure layer for agents is now developing in parallel across at least four dimensions: model selection, workflow orchestration, data serving and security monitoring. Buyers should expect their agent stack to require integration across multiple vendors, not a single monolithic platform. Suppliers building in this space should be explicit about which layer they own and which integrations they support, because buyers are now asking those questions in procurement.
5. Salesforce unveils AIforce and the C-suite rethinks its readiness
At Dreamforce 2026, Salesforce unveiled AIforce, a new interface layer, and Headless 360, which provides access to Salesforce software through non-traditional interfaces. The company described the combination as an Enterprise AI Harness that connects data, business processes and agentic execution.
Separately, research commissioned by Teradata and conducted by Wakefield Research found that C-suite executives may be systematically overestimating their organisation's readiness for agentic AI. In the healthcare sector, 90% of leaders reported that 20% or less of their data is sufficiently described and contextualised for AI agents to act on reliably. Another 43% cited governance, security or access restrictions as a primary barrier to deployment.
Why it matters: the gap between executive confidence and actual readiness is a procurement risk. Buyers who green-light agent programs without an honest assessment of data quality, governance maturity and access controls are likely to experience the same pilot-to-production stall that the MIT/Neo4j research documented. For suppliers, the sales conversation should start with readiness assessment, not capability demonstration.
The Agentic Expo takeaway
This was the week that enterprise agentic AI moved decisively from conversation to infrastructure. Google's agentic Gemini, Salesforce's AIforce and the continued expansion of OpenAI's Dots platform are making task completion the standard feature set. Simultaneously, funding rounds like Rein Security's $25 million Series A, Temporal's expansion into agent controls and the growth of the AI security market to a projected $4.8 billion in 2027 show that trust architectures are becoming the real battleground.
For buyers, the checklist ahead of any agent deployment should include three questions that have nothing to do with model performance. Is your data sufficiently contextualised for agents to reason reliably? Do you have runtime security and incident response for autonomous actions? And does your governance framework match the autonomy level you are about to grant?
For suppliers, the market is now separating capability vendors from infrastructure vendors. The winners in the next twelve months will be those who can answer not just "what can your agent do?" but "how do we know it is safe, observable and recoverable when something goes wrong?"
The agentic economy is not coming. It is here. The question is whether the operational layer underneath it is ready.