In summary:
- ServiceNow’s Agent Fabric introduces the domain-specific Large Language Model (LLM) ‘Apriel Nemotron 15B’ in partnership with NVIDIA.
- It aims to integrate and coordinate AI agents across enterprise applications to tackle the issue of disconnected digital tools.
- ServiceNow and Sopra Steria are betting on the shift from isolated AI solutions to integrated, scalable AI systems.
The exciting announcements of Day one of Knowledge25 had me captivated. Not because of the lights or the fanfare, but because right there in the keynote was something genuinely interesting: a glimpse of how enterprise AI might finally stop being a set of disconnected toys, and start behaving like a proper team.
ServiceNow’s new Agent Fabric, in partnership with NVIDIA, has released a domain-specific Large Language Model (LLM), ‘Apriel Nemotron 15B’ It’s not just your typical AI, but AI that works together. AI that reasons. AI that could finally deliver something meaningful beyond nice demos.
What is Agent Fabric, and why should we care?
For decades, we’ve complained about silos in our organisations. Departments not talking, systems failing to integrate, and inefficiencies that never seem to go away. Is it so ingrained into us humans that we're hell bent on repeating corporate history through our AI agents? They are popping up in every enterprise application, with each vendor in a race to be the next big thing in technology.
Virtual agents that don’t talk to the scheduling bot. Human Resources (HR) AIs that can’t speak to the IT ones. Every tool is doing its own thing, solving one isolated problem, while the bigger picture stays hopelessly fragmented.
You might have a chatbot logging tickets, a scheduler working in its own little world, or an LLM-powered assistant trawling through HR policies - but they’re not collaborating. They’re digital silos. What we’ve built, ironically, is another disconnected stack - with so many more three letter acronyms (TLAs) you'll need an LLM to decipher them!
Agent Fabric is ServiceNow’s answer to these challenges. It’s the integration layer that forces these agents out of their individual echo chambers. It gives them a shared language, context awareness, and the ability to coordinate across domains and platforms. Suddenly your agents are a team, not a crowd.
This isn’t just another integration framework - it’s a rethink of how we deploy, govern and scale AI in the enterprise.
Enterprise AI’s new brain?
ServiceNow’s partnership with NVIDIA to release Apriel Nemotron 15B - a 15-billion parameter LLM for enterprise reasoning, means:
- Logic.
- Structured decision-making.
- Rule-based outcomes.
It’s been optimised for performance and cost. It can be deployed securely on NVIDIA infrastructure. And crucially, it’s open source so you can customise it without handing your data off to some unknown model in the sky.
This is where ServiceNow quietly changes the game. Not by out-chatting ChatGPT, but by building intelligence that fits the messy, layered world of enterprise workflows.
In the perfect world...
Say you’re operating a smart infrastructure hub for a digital city. Real-time feeds from traffic systems, utilities, environmental sensors, and public transport are constantly generating alerts and anomalies.
- A ServiceNow Virtual Agent built into the city’s citizen services portal surfaces a report about a power outage.
- An agent in the city’s grid management system(e.g., GE Digital or Siemens) cross-references grid telemetry and confirms a substation fault.
- A maintenance planning agent embedded in the city’s Enterprise Asset Management (EAM), like IBM Maximo or Infor, retrieves recent repair history and open work orders.
- A field resource agent checks skills, location, and Service Level Agreement windows, and schedules the right team in ServiceNow Field Service Management.
- A communications agent in the city’s Customer Relationship Management system (like Salesforce or ServiceNow Customer Service Management) updates the citizen, and another logs the entire incident in the observability platform (e.g. Dynatrace or Splunk).
Each of these agents is tied to a different enterprise application, but they’re now part of a shared context thanks to Agent Fabric.
That’s not science fiction, that’s operational tech orchestration powered by ServiceNow and it’s within reach today.
Is Agentic Fabric the next step forward or a strategic risk?
This is infrastructure built for a world that should exist, but doesn’t. At least not everywhere. Most organisations aren’t yet mature enough with their data, workflows, or AI governance to make full use of this.
But ServiceNow is betting that over the next 18 months, that’ll start to change. That the era of point-solution AI is ending, and the winners will be the ones who can coordinate, govern, and scale AI across their organisations.
And what about Sopra Steria?
At Sopra Steria we’ve already started building toward Agentic AI. Our Agentic AI studio is up and running, helping clients explore, prototype and deliver real use cases without getting lost in the hype.
Whether it’s HR automation, virtual ward rounds in healthcare, recruitment assistants, or operational tech compliance, we’re proving where Agentic AI makes a tangible difference.
As a ServiceNow Elite Partner, Sopra Steria brings deep domain and platform expertise to the table - from implementation and integration to ongoing optimisation. Our work goes beyond IT Service Management, extending into customer service, HR, field services and more.
We don’t just roll out the tech. We understand that ServiceNow is only one part of your enterprise landscape. We make sure what we deliver aligns with your processes, works across your broader application estate, and ultimately delivers value.
No silver bullets. No “AI in a box.” Just hands-on delivery grounded in what actually works today, with a clear view of what might work tomorrow.
If you’re curious, sceptical, or somewhere in between reach out to andy.thomas@soprasteria.com. We’re figuring this out together, but one thing’s clear: AI in the enterprise just got a lot more interesting.
And this time, it might just work!