The conversation about agentic AI has moved decisively from understanding the technology to determining how it can be deployed safely and effectively at scale.
As we close this series, one point stands out. The organisations positioned to lead over the next 18 months will not necessarily be those that adopted agentic AI first. They will be those that move deliberately, connect investment to business outcomes and establish the operational foundations required for sustainable scale.
The vendor ecosystem is consolidating around enterprise agent platforms. Interoperability standards are developing rapidly. Governance requirements are becoming more demanding, and the skills required to lead AI transformation are expanding beyond technology alone.
Here is where the market is heading and what IT decision-makers need to do next.
The vendor ecosystem is converging and accelerating
The major technology platforms have moved beyond isolated agent experiments. They are committing product roadmaps, engineering resources, partner ecosystems and investment capital to enterprise agent platforms.
At Google Cloud Next ’26, Google introduced Gemini Enterprise Agent Platform. The platform brings together model selection, model development, agent building, integration, orchestration, DevOps, security and governance capabilities within an evolution of the Vertex AI environment.
Google also announced a US$750 million partner fund for agentic development across consulting firms, software providers and channel partners. This indicates that Google sees agents not as an individual product feature, but as a platform layer supported by a broader delivery ecosystem.
Microsoft used Build 2026 to advance a similarly integrated approach.
Microsoft IQ became generally available across GitHub Copilot, Microsoft Foundry and Copilot Studio. It is designed to provide agents with organisational and external context through capabilities including Work IQ, Fabric IQ, Foundry IQ and Web IQ. Some individual components remained in preview at the time of the announcement, but the architectural direction is clear. Microsoft is creating a shared intelligence layer that can ground agents in workplace activity, enterprise data, organisational knowledge and information from the web.
Microsoft Agent Framework also reached version 1.0 general availability in April 2026. The framework combines the enterprise foundations of Semantic Kernel with the orchestration capabilities developed through AutoGen, giving developers a supported foundation for building individual agents and multi-agent workflows.
Salesforce is providing another indicator of commercial momentum.
In the third quarter of its 2026 financial year, Salesforce reported that Agentforce and Data 360 had reached nearly US$1.4 billion in annual recurring revenue, representing 114 per cent year-on-year growth.¹ By the end of the financial year, combined annual recurring revenue had exceeded US$2.9 billion. Salesforce also reported more than 29,000 Agentforce deals since launch.²
These figures include different products and, in the full-year result, Informatica Cloud ARR. They should therefore not be interpreted as pure Agentforce revenue. Even with that qualification, they demonstrate that enterprise spending is moving beyond isolated proof-of-concept activity.
Interoperability is becoming a strategic battleground
As organisations introduce more agents, the ability to connect models, tools, data sources and agents from different providers becomes increasingly important.
Two open protocols have emerged as significant parts of this interoperability layer.
The Model Context Protocol, or MCP, was introduced by Anthropic in late 2024. It provides a standardised way for AI applications and agents to connect with tools, systems and data sources.
Agent2Agent, or A2A, was introduced by Google to support communication and collaboration between agents built by different vendors or with different frameworks. At launch, Google reported support and contributions from more than 50 technology partners and service providers.³ Google later transferred the A2A project to the Linux Foundation, where it is supported by organisations including AWS, Cisco, Microsoft, Salesforce, SAP and ServiceNow.
The two protocols address related but different requirements. MCP helps an agent access tools and context. A2A helps independently developed agents communicate, coordinate and transfer work.
This distinction matters for enterprise architects.
An organisation may use MCP to give an agent controlled access to a document repository, workflow platform or application programming interface. It may then use A2A to allow that agent to collaborate with another specialised agent operating in a separate platform or business domain.
Open protocols do not eliminate vendor lock-in. Organisations can still become dependent on proprietary models, cloud infrastructure, data platforms, security controls and commercial licensing arrangements. However, standards such as MCP and A2A may reduce custom integration work and make heterogeneous agent environments more practical.
IT leaders should therefore treat protocol support as an architectural consideration rather than a product checkbox. They should examine authentication, authorisation, data handling, observability and lifecycle management before connecting external tools or agents to production environments.
The governance gap is growing
Gartner projects that 40 per cent of enterprise applications will include task-specific AI agents by the end of 2026, up from less than 5 per cent in 2025.⁴
That rate of adoption creates a significant governance challenge.
Traditional identity and access management controls were primarily designed for people, applications and relatively predictable service accounts. Autonomous and semi-autonomous agents introduce additional questions.
Who authorised an agent to act?
Which user, team or business process is it representing?
What information can it access?
Can it delegate work to another agent?
How long should its permissions remain active?
Who is accountable when an action is incorrect?
The risks include excessive privileges, unapproved agents, unclear delegation chains, data leakage, prompt injection and actions taken without sufficient human oversight.
These risks can become more difficult to identify when several agents, tools and external platforms participate in the same workflow. A seemingly low-risk permission at one stage may enable a much more consequential action when combined with permissions available elsewhere in the chain.
Regulatory complexity is also increasing. Gartner predicts that fragmented AI regulation will cover 50 per cent of the world’s economies by 2027 and drive US$5 billion in compliance investment.⁵ The EU AI Act is one example of the broader movement towards formal obligations involving transparency, risk management, documentation, human oversight and accountability.
Governance can no longer be added after an agent has moved into production. It needs to be part of architecture, procurement, development and operational management from the beginning.
The skills organisations need to develop now
Scaling agentic AI will require more than expertise in models and software development.
Salesforce’s 2026 C-suite research found that 94 per cent of CIOs said scaling AI was forcing them to expand their skill sets. The leading capabilities they were developing included leadership at 61 per cent, storytelling and narrative building at 57 per cent, and change management and communication at 55 per cent.⁶
This reflects the nature of the challenge.
Agentic AI changes workflows, decision rights, accountability structures and the relationship between people and technology. CIOs must explain why the change matters, establish confidence in the controls and help business leaders redesign work around new capabilities.
Technical teams also need to move from an emphasis on individual prompts or code components towards whole-system thinking.
Three areas deserve particular attention.
Agent orchestration and workflow design
Teams need to understand how to divide work between specialised agents, establish clear roles and manage dependencies across multi-step workflows.
This includes identifying where human approval is required, defining acceptable failure conditions and preventing an agent from expanding its responsibilities beyond the original business purpose.
Good orchestration design should make the workflow understandable. A reviewer should be able to determine which agent performed an action, which information it used and why the workflow progressed to the next stage.
Agent identity and security
Every production agent should be treated as an identity with explicitly defined privileges.
Permissions should be limited to the minimum access required for the task. They should also be reviewed, monitored and revoked when the agent or workflow is retired.
Organisations will need controls for agent registration, ownership, credential management, delegation, audit trails and emergency shutdown. Security teams should also examine how permissions behave when one agent calls another agent or invokes an external tool.
Data readiness and grounding
An agent’s usefulness depends heavily on the information it can retrieve and the quality of the context it receives.
Poorly governed data can lead to inaccurate answers, inappropriate actions and inconsistent decisions. Agent deployments therefore increase the importance of data ownership, metadata, classification, access controls, retention policies and semantic consistency.
Trusted data is no longer simply a reporting or analytics requirement. It is part of the operational infrastructure for enterprise AI.
Microsoft’s Agentic AI Business Solutions Architect certification, associated with exam AB-100, reflects this broader direction. Its scope includes architecture, security, scalability, application lifecycle management, environment strategy and organisational adoption rather than focusing solely on model development.
The emergence of credentials like this suggests that enterprise AI roles are expanding across business architecture, governance and transformation.
Where IT leaders should focus next
IT decision-makers do not need to deploy agents in every process immediately.
They need a repeatable method for deciding where agents can create value and where the organisation can manage the associated risk.
A practical near-term agenda should include:
- Connecting every agent initiative to a measurable business outcome
- Selecting workflows with clear boundaries, sufficient data and accountable owners
- Establishing minimum requirements for identity, security, monitoring and human oversight
- Evaluating MCP, A2A and other protocols as part of an enterprise interoperability strategy
- Creating an inventory of approved agents, tools, models and data connections
- Developing multidisciplinary skills across technology, security, data, risk and change management
- Measuring outcomes such as quality, cycle time, cost, customer experience and error rates
- Defining how unsuccessful pilots will be stopped, redesigned or retired
The organisations gaining meaningful advantage are treating agentic AI as an operating model change rather than a collection of technology experiments.
They are redesigning how work is assigned, completed, reviewed and improved. They are also recognising that greater autonomy must be matched by stronger accountability.
The changing role of service partners
The agentic era is also changing what organisations should expect from technology service providers.
IDC has observed that agentic AI is narrowing the distance between enterprise IT teams and service partners. CIOs increasingly expect partners to support transformation agendas, while providers are embedding their own intellectual property and automation within engagements.⁷
This shifts the relationship away from supplying licences, labour or isolated implementation services.
Organisations need partners that can help connect platform capabilities with operating processes, architecture, data, security and measurable outcomes. They also need clarity about who owns the intellectual property, who carries operational responsibility and how value will be measured over time.
That is where Insentra operates.
As the vendor ecosystem develops and the pace of capability releases increases, organisations need a way to separate useful advances from market noise. They also need practical support translating new functionality into secure architectures, governance frameworks and business outcomes.
The objective is not to deploy the greatest possible number of agents. It is to identify where agents can improve the organisation and then build the controls needed to operate them responsibly.
Your next step
This series has followed the journey from understanding agentic AI to deploying it responsibly and preparing for what comes next.
The near-term direction is becoming clearer.
Agents will become more capable. Platforms will become more integrated. Open protocols will make cross-platform interaction more practical. Regulatory and governance expectations will rise. The gap between organisations with strong operational foundations and those running disconnected experiments will widen.
The next step is to translate these developments into a concrete roadmap for your organisation.
Explore Insentra’s AI practice hub, AI Momentum, for practical guidance on building that roadmap.
The next chapter of agentic AI is already being written. The priority now is to ensure your organisation can participate securely, deliberately and with a clear connection to business value.
Sources
