AI in production: governance, control, and the billions behind AI agents in the week of June 19, 2026
Artificial intelligence has reached a point of no return in the enterprise.
We are no longer talking about proofs of concept running in sandboxes or impressive demos at conferences. AI agents are already in production, making decisions, accessing critical systems, and orchestrating workflows in sectors like banking, manufacturing, healthcare, and government. They do not wait for human approval at every step — they act, query APIs, move data, and trigger chained processes in fractions of a second.
And that is exactly where the game changes entirely. 🎯
When an agent starts acting for real, two questions become urgent: who controls what it can do and who is accountable when something goes wrong? The week of June 19, 2026, was quite revealing about where the market is headed. Funding was significant: Arcade raised $60 million specifically to solve the authorization and governance problem for agents in production, NeuBird AI secured $19.3 million to bring agents into regulated and air-gapped environments, Google Cloud and IBM announced a multi-billion-dollar partnership, and IBM also released a concerning global study about what it called the AI Control Gap — a control gap that is keeping regulators and board members on high alert. 👀
But the week did not stop there. Companies like Cognizant, Deloitte, Dynatrace, Immuta, New Relic, and Opsera also made significant moves that reinforce the same trend: governance, observability, and control have become the new frontier for AI agents. It is no longer about what they can do. It is about ensuring they only do what they are supposed to do.
Arcade and the real problem of authorization in production
When Arcade decided to go to market with its proposition, it targeted a problem many companies have already faced but few knew how to properly name: AI agents in production need granular, traceable, and revocable permissions in real time. It is not enough to give access to a system and hope the agent uses that access with good judgment. In real corporate environments, especially in regulated sectors, every action needs context, an audit trail, and a clear control mechanism that allows you to stop, adjust, or revoke what the agent can do without bringing down the entire workflow.
The $60 million raised in a Series A round, led by SYN Ventures, brought the company’s total funding to $72 million. These numbers reflect the exact size of this market and how much organizations are willing to pay to have the authorization problem solved reliably.
Arcade’s proposition revolves around a secure action layer — a kind of authorization middleware — that sits between the agent and the systems it needs to access. Think of it this way: a customer service agent should not have the same level of access as a financial reconciliation agent. But without dedicated infrastructure for this, many companies end up granting overly broad access for operational convenience, and that is when risk grows exponentially. Arcade solves this with refined authorization policies, reliability controls, and detailed audit trails for every agent action across different systems. Permissions and agent governance are treated as centralized, explicit infrastructure rather than improvised code scattered across each application.
What makes this move even more relevant is the timing. The AI agent market is accelerating at a brutal pace, with estimates pointing to hundreds of billions of dollars in value generated over the next five years. But that value only materializes if companies can put agents into production with enough confidence to let them operate autonomously in sensitive processes. And confidence, in this context, is not a feeling — it is an architecture. Every dollar invested in Arcade is essentially a vote that the authorization layer will become critical infrastructure for any serious agent stack, just as authentication and encryption are today for any web application.
Receive the best innovation content in your email.
All the news, tips, trends, and resources you're looking for, delivered to your inbox.
By subscribing to the newsletter, you agree to receive communications from Método Viral. We are committed to always protecting and respecting your privacy.
NeuBird AI and the challenge of regulated environments
If Arcade is solving the authorization problem in general environments, NeuBird AI chose an even more challenging path: bringing AI agents into regulated and isolated environments — the well-known air-gapped environments — where internet connectivity is nonexistent or severely restricted by regulatory or security requirements. This includes critical infrastructure, defense systems, hospitals with strict HIPAA requirements, financial institutions under central bank scrutiny, and government agencies that simply cannot have data leaving their internal network.
With $19.3 million in the bank, the company added support for deployments in virtual private cloud and environments completely isolated from the public network. Combining this with its Falcon engine, which focuses on predictive risk detection and root cause analysis, NeuBird enables SRE and operations teams in regulated environments to use AI agents for incident response, reliability, and cost optimization across hybrid and multi-cloud systems.
The technical problem here is genuinely hard. Most large language models powering agents today were designed to run in the cloud, with continuous access to updates, telemetry, and distributed infrastructure. Bringing this inside an isolated environment requires not only the portability of the model itself but also the entire chain of observability, updating, and version control that ensures the agent continues to function predictably over time. NeuBird is betting there is a massive market of organizations that need the benefits of artificial intelligence but cannot compromise their isolation requirements — and that this market is underserved by the major players precisely because the technical complexity of addressing this demand is too high to be treated as an additional feature of a larger platform.
From a governance standpoint, regulated environments have even stricter requirements than the conventional corporate market. It is not enough to have audit logs — they need to be immutable, verifiable, and available to external regulators in a specific format. It is not enough for the agent to make good decisions — each decision needs to be explainable in language that a human auditor can understand and question. NeuBird is building this as a native layer, not as an add-on, which completely changes the product architecture and, consequently, the level of trust that the most demanding sectors will place in it. 💡
IBM, Google Cloud, and the AI Control Gap that worries the corporate world
The partnership announced between Google Cloud and IBM is not just another commercial deal between tech giants. It signals that both companies see a specific market opportunity at the intersection of enterprise cloud infrastructure and governance layers for AI agents. IBM brings decades of relationships with companies in regulated sectors — banks, insurers, governments, heavy manufacturing — while Google Cloud brings the scale infrastructure and the most advanced AI models on the market. The initiative combines Google’s Gemini Enterprise Agent platform and security capabilities with IBM Consulting Advantage and industry-specific agents, enabling teams to design, build, and govern AI agents more quickly across domains like banking, government, retail, and telecommunications.
But what grabbed the most attention this week was the global study released by the IBM Institute for Business Value on the AI Control Gap. The report showed that a significant portion of companies that already have agents in production lack complete visibility into what those agents are doing in real time. The executives surveyed reported challenges with data residency, model oversight, and accountability for third-party AI. They know the agent was configured with certain permissions and objectives, but they do not have a unified dashboard that shows, at any given moment, which decisions are being made, which systems are being accessed, and whether any action falls outside the expected behavioral envelope.
This is the AI Control Gap in practice, and it represents an operational, regulatory, and reputational risk that boards of directors around the world have started taking very seriously. 👀
At the same time, IBM made a series of announcements about new agentic capabilities distributed across its platforms: OpenRAG in watsonx.data for governed information retrieval, Engineering AI Hub 1.3 as an agentic automation platform for lifecycle management, Docling for watsonx to convert complex documents into AI-ready data, and monitoring of agentic AI systems in Guardium. All of this reinforces IBM’s strategy of embedding agentic AI into data, security, and engineering platforms while protecting against uncontrolled behavior.
Digital transformation has never been just about adopting new technology — it has always been about ensuring the organization can operate, audit, and continuously improve what it adopted. With AI agents, this becomes more complex because the agent’s behavior is not deterministic in the same way as traditional software. It can act in unexpected ways when facing inputs nobody tested, and without real observability, you only find out when the damage is already done. The IBM study is a direct warning to the market: having agents in production without robust control is not innovation — it is uncovered risk.
Cognizant, Deloitte, and the race for industry-specific applied solutions
While startups like Arcade and NeuBird build foundational infrastructure, large consulting firms and service companies are betting on the application layer, bringing AI agents to industry-specific contexts.
Cognizant has been emphasizing applied AI solutions tied to specific industries — financial services, healthcare, manufacturing, and retail — rather than offering only generic tools. By combining modernization projects, data platform work, and domain-specific accelerators, the company is positioning itself as the services layer that translates horizontal AI capabilities into workflows and outcomes for concrete business areas.
Deloitte, in turn, launched a dedicated practice — the Google Cloud Agentic Transformation Practice — focused on helping companies adopt Gemini Enterprise and agentic AI more systematically. This practice aims to accelerate AI-driven transformation by combining Deloitte’s industry expertise and change management capabilities with Google Cloud’s Gemini agent platform. Clients gain access to reference architectures, use case libraries, and operating models for agent-based workflows.
This move by the consulting firms reflects a practical reality: technology alone does not generate business value. You need someone who can translate the capabilities of models and agents into processes that make sense within the regulatory, operational, and cultural context of each organization.
Observability and DevOps: Dynatrace, New Relic, and Opsera enter the game
Another strong theme of the week was the convergence between observability and agentic AI — the idea that agents need to be monitored, measured, and orchestrated with the same rigor as microservices in production.
Dynatrace brought a thought-provoking perspective by arguing that, as organizations adopt an AI workforce composed of both humans and agents, they need a new operating model that connects observability, automation, and governance. In the company’s view, AI workers should be monitored and measured exactly like microservices, using AI-powered observability on platforms like Dynatrace on Google Cloud to keep complex, agent-heavy systems running reliably.
New Relic took a concrete step in this direction by announcing that its MCP server now integrates with Kiro, the AI-native and agentic development environment from AWS. This integration allows developers to pull observability context directly into their AI-powered development workflows. In practice, AI-powered tools and agents can reason about real-time telemetry while designing, testing, and operating applications, making observability data a first-class ingredient in the AI-based software lifecycle.
Opsera, meanwhile, continues pushing AI into DevOps orchestration, turning CI/CD and release telemetry into recommendations and guardrails — not just passive dashboards. By analyzing pipeline data and production signals, Opsera’s AI can suggest optimizations, flag risky changes, and help teams design safer and faster delivery workflows.
Immuta and agents as consumers of governed data
One of the most conceptual yet extremely relevant announcements in practice came from Immuta. The company launched Agentic Data Access capabilities that frame AI agents as governed data consumers — subject to the same policies and monitoring as human users.
The concept of least privilege and intent-based access might sound abstract, but it solves a very concrete problem: how do you allow an agent to query and act on sensitive data in analytical and operational workflows without opening security gaps? Immuta does this by defining restricted, context-based access and exposing usage patterns, giving data and security teams a secure way to enable agents to work with sensitive information.
This approach is complementary to what Arcade does with action authorization: while Arcade controls what the agent can do, Immuta controls what the agent can see and access in terms of data. Together, these layers are beginning to form what could become a complete governance stack for AI agents in production.
Events and the technical community watching agentic architectures
It was not just products and partnerships that defined the week. Decode-style events held throughout June continued focusing on agentic architectures, experimentation frameworks, and governance as the topics where professionals most want practical guidance. Rather than just model demos, the sessions tend to break down how teams design workflows, control planes, and safety mechanisms for AI agents. These gatherings have become essential spaces for sharing playbooks that go beyond vendor demos.
Also worth highlighting is the Google Cloud report on AI agent trends in manufacturing for 2026. The document showed how agents are changing everything from quality inspections and predictive maintenance to shop floor scheduling. The key takeaway is that agents are moving beyond pilots and into permanent operational roles, orchestrating tasks and decisions across machines, operational technology data, and enterprise systems.
What the week of June 19 reveals about the maturity of the sector
What the week of June 19, 2026, made clear is that the artificial intelligence sector has entered a new phase of maturity. The race to build more capable agents still exists and will not slow down anytime soon, but it now coexists with a parallel and equally urgent race: building the trust infrastructure that allows these agents to operate at scale, with accountability, in environments where mistakes carry real costs.
The signals are consistent across all of the week’s announcements:
- Arcade raised $60 million for centralized agent authorization
- NeuBird AI secured $19.3 million to bring agents into isolated and regulated environments
- IBM warned the market about the AI Control Gap and expanded agentic capabilities in watsonx and Guardium
- Google Cloud and IBM signed a multi-billion-dollar partnership to scale agents in production
- Deloitte created a dedicated practice for agentic transformation with Gemini Enterprise
- Dynatrace advocated for a new operating model for the AI workforce
- Immuta framed agents as governed data consumers
- New Relic integrated observability into AWS’s agentic development environment
- Opsera turned DevOps telemetry into AI-powered actionable guidance
- Cognizant deepened industry-specific applied AI solutions
Governance is no longer a compliance conversation — it has become a competitive advantage for whoever can solve it first. The market is no longer asking whether agents work. It is asking whether they work in a controlled, auditable, and reliable enough way to operate at the heart of organizations. And the companies that answer this question with real infrastructure — not roadmap promises — are the ones that will capture the value of this next phase of artificial intelligence. 🚀
