[AI Burst] 16 Software Giants Got Agent-Ready. Almost Nobody Is Using It Yet

NVIDIA just handed the enterprise world an open-source platform for building autonomous AI agents. It's called the Agent Toolkit — it includes a runtime called OpenShell, a family of models called Nemotron, and a set of prebuilt agent blueprints. And the list of companies already committed to building with it is genuinely staggering: Adobe, Atlassian, Box, Cisco, CrowdStrike, SAP, Salesforce, ServiceNow, Siemens, and six others.

Jensen Huang called the opportunity a $1 trillion economy at GTC this month. The global agentic AI market is at $3.1 billion today and projected to hit $47.2 billion by 2035.

Here's the part nobody is saying loudly: 85% of enterprise customers are already running agent pilots. And only 5% have moved any of those agents into production. The technology is there. The software is ready. The adoption is stalled — and security concerns are the single biggest reason why.

❓Why It Matters
Think about what your software stack looks like after this. Your ServiceNow ticketing system could start triaging and resolving issues before a human touches them. Your Salesforce CRM could update records, draft follow-ups, and route leads on its own. Your Atlassian Jira could close duplicates, assign stories, and summarize sprint risks automatically.

That's what "agent-ready" means in practice. Not a new AI chatbot. Your existing software doing things on its own.

The problem is the 95% number. When only 5% of enterprises have agents actually running, it tells you something important: being ready to deploy and being ready to govern are two completely different things. Who owns the decision when the agent gets it wrong? What's the audit trail? What happens when the agent has read access to systems a human would need approval to touch?

Those questions don't get answered in a GTC keynote. They get answered — slowly — in the security and legal teams, after months of internal debate. That's the 95%.

Pick one agent use case in your organization and run a proper pilot — not just a demo. Get the IT, legal, and security teams involved from day one, not after the fact. Define upfront which decisions the agent can make autonomously, which ones require human review, and what the rollback process looks like. The enterprises that get agents into production in 2026 are the ones who answered those questions early. The other 95% are still debating them.

🛠️ THE TOOLKIT

The high-leverage GenAI stack you need to know this week.

Your 30-second scan of the AI landscape.

In 1979, a software error at Three Mile Island caused automated safety systems to give operators contradictory readings and lock out human override. The partial meltdown that followed led to a 20-year halt in US nuclear plant construction — not because nuclear power was inherently unsafe, but because automation had outpaced the protocols for governing it.

The lesson wasn't "stop automating." It was that autonomous systems making consequential decisions require governance protocols designed before deployment, not after the incident. That's exactly where enterprise AI agents are right now.

"Move fast and break things only works if you own the things you're breaking." — paraphrase of a now-common critique of Silicon Valley's early philosophy

Till next time,

For deep-dive analysis on cybersecurity and AI, check out my popular newsletter, The Cybervizer Newsletter

Get the AI & tech news that actually matters and stay ahead of updates with one clear, five-minute newsletter.

Forward Future is read by builders, operators, and leaders from NVIDIA, Microsoft, and Salesforce who want signal over noise and context over headlines.

And you get it all for free, every day.

Sign Up. Stay Ahead.