Ericsson puts agentic AI at the core of its OSS/BSS operations

Ericsson’s architectural blueprint adds an agentic service experience layer

In sum – what we know:

• Outcome-first design – The Business Value Pathways work backward from a customer’s business result rather than starting with the technology, blending products and services from across Ericsson’s portfolio.
• Agents as first-class citizens – AI agents are built into the framework to orchestrate customer journeys, service lifecycles, and assurance in a unified way, not bolted on after the fact.
• Bridging networks and operations – The approach links AI-native networks to AI-native business operations, keeping RAN and core automation accountable to revenue and service-experience goals.

Ericsson has put agentic AI at the center of its operations and business support stack with a new set of Business Value Pathways — an approach that works backward from a customer’s business outcome rather than starting from the technology, blending products, services, and more from across the Ericsson portfolio to get there. AI agents are a first-class component of these pathways rather than a bolted-on feature. The announcement extends the AI-native OSS/BSS portfolio the company introduced in 2025.

The idea is to have AI agents orchestrate customer journeys, service lifecycles, and assurance in a unified way, rather than leaving each of those as a separate workflow stitched together by integration work. In practice, that means the framework is pitched as a bridge between AI-native networks and AI-native business operations — keeping RAN and core automation accountable to revenue and service experience goals instead of optimizing the network in isolation. It’s targeted squarely at CSP demands for faster service innovation and the ability to experiment with AI-driven experiences without ripping out legacy OSS/BSS.

Intent-driven operations and a multi-agent architecture

Intent-based automation has been Ericsson’s consistent message for a while, and the new framework leans on it. The idea is that commercial teams can use natural language to define an offer or an outcome, and the system handles the rest. State an intent like “reduce churn among high-value customers,” and the framework is meant to generate consistent catalog entries, charging logic, and provisioning workflows from that single statement rather than having product, billing, and network teams build each piece by hand. Ericsson’s own OSS/BSS blogs already describe multi-agent AI doing this for product configuration, where agents interpret commercial intent, map it to technical product structures, validate feasibility, and write out consistent configurations across catalog, charging, and provisioning systems.

Under the hood, this runs as a multi-agent system. Ericsson describes specialized agents for data ingestion, reasoning, planning, simulation, and execution, coordinated by orchestration logic that ties their actions together. Across its OSS/BSS and data and analytics portfolio, the company is already building agentic capabilities aimed at distinct domains. 

The Telco Agentic AI Studio fits into this too. Launched in 2025 on Amazon Bedrock, the Studio helps build and accelerate AI applications for OSS/BSS, including the kind of multi-agent, end-to-end service experiences these pathways call for — agents coordinating across marketing, charging, QoS, and support to satisfy an intent.

Ecosystem integration and the data pipeline

None of this works without good data, of course. The framework deeply integrates the Telco DataOps Platform as a real-time streaming data backbone for AI and automation. The point is to wire customer, service, and network data into a single pipeline so agents act on consistent data across the whole network-business stack. Ericsson’s June 2026 “From data to decisions” blog makes the case bluntly that agentic operations need disciplined pipelines that collect, clean, correlate, and contextualize data before agents act. Without that, an agent adjusting QoS based on predicted churn is acting on guesswork.

The framework also enforces closed-loop service experience management. Earlier closed loops were mostly about network performance, but here the loop extends to experience metrics. Agents directly observe KPIs like NPS, adjust service quality or offers in response, and then measure the resulting business impact. That’s the part that turns customer experience from something you watch into something you control.

Crucially, this isn’t a standalone island. The framework links with the Intelligent Automation Platform (EIAP), which Ericsson expanded to cover core network automation alongside RAN, plus new real-time data-streaming capabilities for Network Manager. It also folds in capabilities from Ericsson’s early 2026 Differentiated Support release, using agentic AI for automated multi-vendor ticket triage and root-cause analysis — the “hero” of the company’s Intelligent Support portfolio. Set against Ericsson’s broader 2026 moves, including its recent “AI in RAN” subscription that embeds telco-grade AI models directly into basebands and radios, the direction is clear enough — the company is working to close the loop between autonomous networks and autonomous operations.

This is also where the competitive contrast with operators gets interesting. Verizon has publicized using agentic AI across its 60,000-site vRAN for configuration, service assurance, and optimization, and has called for interoperability standards for agentic systems. That’s an in-house build. Ericsson’s framework offers a standardized, vendor-driven alternative — appealing to operators that don’t want to engineer their own agentic stack, though it comes with the usual tradeoff of leaning on a vendor’s roadmap and assumptions.

Cloud strategy and hyperscaler deployment

AWS is a significant part of the picture, though not the whole of it. The Telco Agentic AI Studio and the Gen-AI Lab are both built on Amazon Bedrock, and AWS is a strategic partner for Ericsson. The logic is straightforward enough. Hyperscaler AI platforms provide the elastic compute needed to scale agentic workloads, while Ericsson keeps operational telco data control and governance inside its Telco DataOps environment and within CSP boundaries. 

And Ericsson is explicit that it’s cloud agnostic. The Telco IT AI Apps run on Ericsson’s own cloud-agnostic platform, the Telco IT AI Engine, or on other hyperscaler AI and Gen-AI platforms, and the more than 20 cloud-native AI applications from the 2025 portfolio can deploy across those options. AWS is a proven and well-developed path, particularly for multi-agent workloads on Bedrock, but it isn’t a requirement for getting Ericsson’s capabilities into production.

For operators already comfortable on AWS, having Bedrock as a proven path lowers the barrier to getting agents into production. For those running private data centers or committed to a competing cloud, the practical question is how much of the most-developed tooling carries over to Ericsson’s own engine or another platform, and at what cost in capability. That’s a fair thing for CSPs to press on with specifics before they build a service experience strategy on top of it.