When NVIDIA CEO Jensen Huang took the stage at GTC 2026 in San Jose on March 16, he didn’t just announce a product — he declared the beginning of what he calls the next industrial revolution in knowledge work. The NVIDIA Agent Toolkit, an open-source enterprise AI agent platform, launched with 17 major enterprise software partners already committed: Adobe, Salesforce, SAP, ServiceNow, Siemens, CrowdStrike, Atlassian, Cadence, Synopsys, IQVIA, Palantir, Box, Cohesity, Dassault Systèmes, Red Hat, Cisco, and Amdocs. That list covers virtually every Fortune 500 corporation on the planet. By end of 2026, Gartner projects that 40% of all business applications will incorporate AI agents — up from under 5% in 2025. NVIDIA just staked its claim on owning the platform layer of that transition.
What Happened
NVIDIA unveiled the Agent Toolkit as a fully open-source software stack designed to help enterprises and developers build, deploy, and manage autonomous AI agents at scale. The announcement came at GTC 2026, NVIDIA’s flagship developer conference, and was immediately backed by 17 of the world’s largest enterprise software companies.
The toolkit is built around three core components. The first is NVIDIA OpenShell™, an open-source runtime that enforces policy-based security and privacy guardrails for autonomous agents — individual agents are called “claws” within the system, and OpenShell acts as the governance layer that keeps them operating within defined boundaries. The second is the NVIDIA AI-Q Blueprint, a hybrid architecture where frontier models handle high-level orchestration and reasoning, while lighter open-source Nemotron models perform research and retrieval tasks — resulting in a claimed 50%+ reduction in cost per query without sacrificing accuracy. The third is the NVIDIA Nemotron Model Family, which underpins partner integrations across Adobe, Salesforce, SAP, and others.
The platform is available immediately on build.nvidia.com, with inference support across Baseten, CoreWeave, DeepInfra, and DigitalOcean, plus enterprise deployment across AWS, Google Cloud, Microsoft Azure, and Oracle Cloud Infrastructure.
Why This Matters for Enterprise Automation
The AI agent market has been fragmented and experimental for the past two years. Enterprises have been running pilots — small, isolated proof-of-concept deployments that rarely made it into production workflows. NVIDIA’s move changes that calculus in a fundamental way.
By creating a shared, open-source foundation that 17 enterprise platforms are all building on simultaneously, NVIDIA has effectively standardised the infrastructure layer of enterprise AI automation. A company using Salesforce Agentforce for sales, SAP Joule for ERP operations, and ServiceNow for IT workflows can now run all of those agents on a common runtime with consistent guardrails, shared security models, and interoperable observability tools. That interoperability has been the single biggest blocker to enterprise agent adoption — and NVIDIA just removed it.
The real-world numbers are already compelling. Automation Anywhere, one of the broader ecosystem players, reported that AI agents now auto-resolve over 80% of IT support requests, cutting IT service management costs by up to 50%. IQVIA, one of the 17 launch partners, has already deployed more than 150 agents across internal teams and client environments, including 19 of the top 20 pharmaceutical companies globally.
Global Market Context
The agentic AI market is at an inflection point. Foundational AI startup funding in Q1 2026 alone doubled the total invested in all of 2025, with OpenAI raising a record $122 billion round, Anthropic closing a $30 billion Series G, and xAI securing $20 billion in January 2026. This capital concentration at the model layer is now being matched by an equally aggressive race at the platform and tooling layer — and that is precisely where NVIDIA’s Agent Toolkit is positioned.
The global enterprise AI market is projected to exceed $500 billion by 2028, with AI agents representing the fastest-growing segment. Gartner’s latest forecast places AI agent adoption at 40% of business applications by end of 2026, while IDC estimates that enterprises deploying autonomous agents at scale will see productivity gains of 25–35% in knowledge worker roles. The workforce economics are stark: companies that successfully automate repetitive decision-making tasks can redeploy human capital toward higher-value creative and strategic work.
Geographically, North America leads in enterprise AI agent deployment, but adoption is accelerating rapidly in Europe and Asia-Pacific. The EU AI Act, which came into full effect in 2025, has actually accelerated enterprise agent adoption in Europe by creating clear compliance frameworks — and NVIDIA’s OpenShell governance layer is explicitly designed to satisfy those regulatory requirements.
Key Players and Their Positions
Each of the 17 launch partners brings a distinct enterprise surface area to the NVIDIA Agent Toolkit ecosystem.
Adobe is using the toolkit as the foundation for long-running creativity, productivity, and marketing agents built on its Firefly AI models. CEO Shantanu Narayen specifically cited the integration of CUDA libraries, 3D digital twins for marketing workflows, and Nemotron-powered agentic frameworks. For enterprise marketing teams, this signals a future where creative production pipelines are largely autonomous.
Salesforce is integrating Nemotron models into Agentforce, its AI agent platform for sales, service, and marketing. The reference architecture places Slack as the primary conversational interface for agent interactions — meaning sales and customer service agents could soon be orchestrating Salesforce workflows autonomously through Slack channels, drawing from both on-premises and cloud data sources.
SAP is deploying the toolkit through Joule Studio on SAP Business Technology Platform, enabling customers and partners to design custom agents tailored to their specific business processes. Given that SAP powers the operational backbone of tens of thousands of enterprises globally, this integration has profound implications for ERP automation.
Siemens has already launched a concrete product: the Fuse EDA AI Agent, which uses Nemotron to autonomously orchestrate workflows across its entire electronic design automation portfolio, from initial design conception through manufacturing sign-off. This is a direct example of physical AI — agents managing complex engineering workflows in the real world.
The competitive threat falls most heavily on standalone AI automation vendors — companies like UiPath, Automation Anywhere, and Blue Prism — which now face a world where the platforms their customers already use (SAP, Salesforce, ServiceNow) are shipping native AI agent capabilities on a shared, enterprise-grade foundation. The question is no longer whether to add AI agents, but which platform will own the agent orchestration layer.
What This Means for Businesses
For enterprise decision-makers evaluating their automation and AI strategies in 2026, the NVIDIA Agent Toolkit announcement has several direct implications.
- Consolidate your automation stack around Agent Toolkit-compatible platforms. If you’re already on Salesforce, SAP, or ServiceNow, your path to enterprise AI agents just got dramatically clearer. Native integrations built on a shared security and governance layer reduce the implementation complexity and risk that has stalled so many enterprise AI pilots.
- Prioritise governance and compliance infrastructure now. NVIDIA’s OpenShell is designed to satisfy EU AI Act requirements and enterprise security policies. As agent deployments scale from pilots to production, companies without clear governance frameworks will face both regulatory and operational exposure. Building that infrastructure in 2026, before deployments scale, is significantly cheaper than retrofitting it later.
- Rethink workforce planning with a 2–3 year horizon. If IQVIA can deploy 150 agents across 19 of the top 20 pharma companies, and Automation Anywhere’s agents resolve 80% of IT support tickets autonomously, the knowledge work implications are systemic. Forward-looking CHROs and COOs should be modelling which roles will be augmented, which will be transformed, and which may be structurally reduced.
- Cost-per-query is a new procurement metric. The AI-Q Blueprint’s 50% reduction in cost per query is not just a marketing claim — it’s a signal that enterprise AI cost structures are changing rapidly. Procurement teams should build cost-per-query benchmarks into AI vendor contracts now, rather than paying legacy per-seat licensing on tasks that agents can handle at a fraction of the cost.
- Watch the Nemotron Coalition for future model options. NVIDIA has simultaneously launched the Nemotron Coalition, a global collaboration including Mistral AI, Cursor, LangChain, Perplexity, Reflection AI, Sarvam, and Thinking Machines Lab. The first deliverable is a base model co-developed by Mistral AI and NVIDIA on DGX Cloud. As the coalition matures, enterprises will have a growing menu of open, enterprise-grade models to deploy within the Agent Toolkit framework.
What to Watch Next
Several near-term developments will determine how quickly the NVIDIA Agent Toolkit reshapes the enterprise automation landscape.
The first signal to watch is partner product launches. The 17 founding partners are all in various stages of integrating Agent Toolkit software into their existing product lines. Adobe’s first Firefly-powered autonomous creative agent, and Salesforce’s Slack-based Agentforce orchestration layer, are both expected to reach general availability in mid-2026. Real-world adoption metrics from these early rollouts will either validate or temper the hype around autonomous enterprise agents.
The second is the Nemotron 4 model release. The Nemotron Coalition’s first co-developed base model, built in partnership with Mistral AI, is expected to launch in the coming months. Its performance benchmarks — particularly on cost-per-query metrics in enterprise retrieval tasks — will significantly influence whether enterprises adopt open Nemotron models or continue paying for proprietary frontier model APIs.
The third is the competitive response from Microsoft. Azure AI Foundry and Microsoft Copilot Studio are direct competitors to the Agent Toolkit paradigm, and Microsoft’s deep entrenchment across enterprise software gives it significant leverage. Watch for Microsoft’s GTC counter-narrative and any announcements at Build 2026 in May regarding its own enterprise agent governance framework.
Finally, regulatory developments in the US — where the AI executive order landscape remains in flux in 2026 — could either accelerate or complicate enterprise agent deployments, particularly in regulated sectors like finance, healthcare, and government contracting.
What is the NVIDIA Agent Toolkit and how does it work?
The NVIDIA Agent Toolkit is an open-source enterprise software stack launched at GTC 2026 that enables businesses to build, deploy, and manage autonomous AI agents at scale. It consists of three components: NVIDIA OpenShell (a governance and security runtime for agents), the AI-Q Blueprint (a hybrid architecture using frontier and Nemotron models for cost-efficient orchestration), and the Nemotron model family for specific retrieval and reasoning tasks. Enterprises access it via build.nvidia.com and deploy it across major cloud providers including AWS, Azure, Google Cloud, and Oracle.
Which enterprise companies are using NVIDIA’s Agent Toolkit?
At launch, 17 enterprise software companies committed to building on the Agent Toolkit: Adobe, Salesforce, SAP, ServiceNow, Siemens, CrowdStrike, Atlassian, Cadence, Synopsys, IQVIA, Palantir, Box, Cohesity, Dassault Systèmes, Red Hat, Cisco, and Amdocs. These companies collectively serve the vast majority of Fortune 500 corporations, meaning the toolkit will rapidly become embedded in existing enterprise software stacks rather than requiring separate procurement.
How much can enterprise AI agents reduce operational costs?
Real-world deployments suggest significant cost reductions. Automation Anywhere has reported that AI agents auto-resolve over 80% of IT support requests, cutting IT service management costs by up to 50%. NVIDIA’s own AI-Q Blueprint delivers more than 50% reduction in cost per query compared to using frontier models alone. IQVIA has deployed over 150 agents across 19 of the top 20 pharmaceutical companies, indicating that large-scale, cost-effective deployment is already achievable in regulated industries.
Is the NVIDIA Agent Toolkit compliant with the EU AI Act?
NVIDIA’s OpenShell governance layer is explicitly designed to satisfy EU AI Act compliance requirements through policy-based security guardrails and auditable agent behaviour controls. The EU AI Act, which came into full force in 2025, classifies many enterprise AI agent use cases under its high-risk or limited-risk categories, requiring transparency, human oversight, and logging. OpenShell’s architecture addresses these requirements natively, which is one reason European enterprises are among the early adopters.
What is the difference between NVIDIA Agent Toolkit and Microsoft Copilot Studio?
Microsoft Copilot Studio and NVIDIA Agent Toolkit both target enterprise AI agent deployment, but from different angles. Copilot Studio is a low-code platform tightly integrated with Microsoft 365 and Azure, designed for business users to build agents within the Microsoft ecosystem. NVIDIA’s Agent Toolkit is an open-source, model-agnostic infrastructure layer that any enterprise software vendor can build on, regardless of cloud or OS. The key distinction is openness: Agent Toolkit is designed to be the shared foundation beneath platforms like Salesforce, SAP, and Adobe — not a standalone product competing with them.
The launch of NVIDIA’s Agent Toolkit at GTC 2026 marks a genuine inflection point — not just in enterprise AI, but in how the automation industry is structured. By open-sourcing the agent infrastructure layer and signing 17 of the world’s largest software companies onto a shared foundation, NVIDIA has effectively called the end of the AI pilot era. For enterprise leaders, the window to move from experimentation to deployment is narrowing fast: the companies that establish agent governance frameworks, build Nemotron-compatible workflows, and retrain their workforce for human-agent collaboration in 2026 will hold compounding advantages over those that wait. The industrial revolution in knowledge work Jensen Huang described from the GTC stage is not a distant forecast — it is already shipping.
Last Updated: April 2026