27
Ericsson’s architectural blueprint provides an agentic service expertise layer
In sum – what we all know:
- Brokers as structure – Ericsson’s new blueprint treats AI brokers as a first-class OSS/BSS element, not a Gen-AI function bolted onto present operations.
- Intent followers out – A single natural-language goal like “scale back churn amongst high-value prospects” generates catalog, charging, and provisioning modifications throughout Expertise, Income, and Community brokers.
- AWS lock-in – Probably the most developed tooling runs on Amazon Bedrock, which lowers the barrier for AWS operators however raises actual lock-in considerations for everybody else.
Ericsson has put agentic AI on the heart of its operations and enterprise assist stack, formalizing a brand new OSS/BSS architectural blueprint the place AI brokers are a first-class element moderately than a bolted-on function. The announcement extends the AI-native OSS/BSS portfolio the corporate launched in 2025, however the framing is completely different this time. The place the sooner work was about embedding Gen-AI throughout operations, that is about brokers that plan, motive, and act throughout each community operations and enterprise processes.
The headline idea is what Ericsson calls an agentic AI service expertise layer, layered over its present OSS/BSS domains. The concept is to have AI brokers orchestrate buyer journeys, service lifecycles, and assurance in a unified approach, moderately than leaving every of these as a separate workflow stitched collectively by integration work. In apply, meaning the framework is pitched as a bridge between AI-native networks and AI-native enterprise operations — maintaining RAN and core automation accountable to income and repair expertise objectives as a substitute of optimizing the community in isolation. It’s focused squarely at CSP calls for for sooner service innovation and the power to experiment with AI-driven experiences with out ripping out legacy OSS/BSS.
The announcement lands alongside Nokia’s launch of its Autonomous Networks Agent Library for IP networks, which pushes the automation battleground out of the community layer and instantly into OSS/BSS information possession. Each distributors are racing to embed AI brokers into the operational and enterprise material.
Intent-driven operations and a multi-agent structure
Intent-based automation has been Ericsson’s constant message for some time, and the brand new framework leans on it arduous. The pitch is that business groups can use pure language to outline a proposal or an final result, and the system handles the remaining. State an intent like “scale back churn amongst high-value prospects,” and the framework is supposed to generate constant catalog entries, charging logic, and provisioning workflows from that single assertion moderately than having product, billing, and community groups construct every bit by hand. Ericsson’s personal OSS/BSS blogs already describe multi-agent AI doing this for product configuration, the place brokers interpret business intent, map it to technical product buildings, validate feasibility, and write out constant configurations throughout catalog, charging, and provisioning methods.
Underneath the hood, this runs as a multi-agent system. Ericsson describes specialised brokers for information ingestion, reasoning, planning, simulation, and execution, coordinated by orchestration logic that ties their actions collectively. What’s new within the OSS/BSS framing is that a few of these brokers are uncovered natively as domain-specific roles. There are Expertise brokers centered on buyer journey and QoE, Income brokers dealing with pricing, upsell, and margin optimization, and Community brokers wired into EIAP and rApps for topology and useful resource modifications. The intent is {that a} single acknowledged goal can fan out throughout all three.
That additionally reshapes what the Telco Agentic AI Studio is for. When it launched in 2025 on Amazon Bedrock, the Studio was about constructing and accelerating AI functions for OSS/BSS. The brand new framework strikes it away from fundamental app creation towards actively composing multi-agent, end-to-end service experiences — brokers coordinating throughout advertising and marketing, charging, QoS, and assist to fulfill an intent. It’s a significant shift in ambition, although it’s price noting it’s additionally a pure extension of what the Studio was already doing moderately than a clear break.
Ecosystem integration and the information pipeline
None of this works with out good information, in fact. The framework deeply integrates the Telco DataOps Platform as a real-time streaming information spine for AI and automation. The purpose is to wire buyer, service, and community information right into a single pipeline so brokers act on constant information throughout the entire network-business stack. Ericsson’s June 2026 “From information to selections” weblog makes the case bluntly that agentic operations want disciplined pipelines that gather, clear, correlate, and contextualize information earlier than brokers act. With out that, an agent adjusting QoS primarily based on predicted churn is performing on guesswork.
The framework additionally enforces closed-loop service expertise administration. Earlier closed loops had been largely about community efficiency, however right here the loop extends to expertise metrics. Brokers instantly observe KPIs like NPS, regulate service high quality or provides in response, after which measure the ensuing enterprise influence. That’s the half that turns buyer expertise from one thing you watch into one thing you management.
Crucially, this isn’t a standalone island. The framework hyperlinks constantly with the Clever Automation Platform (EIAP), which Ericsson expanded round late June 2026 to cowl core community automation alongside RAN, plus new real-time data-streaming capabilities for Community Supervisor. It additionally folds in capabilities from Ericsson’s early 2026 Differentiated Assist launch, utilizing agentic AI for automated multi-vendor ticket triage and root-cause evaluation — the “hero” of the corporate’s Clever Assist Providers portfolio. And it interfaces with the “AI in RAN” subscription Ericsson launched earlier this month, which embeds telco-grade AI fashions instantly into basebands and radios for spectral and power effectivity with out new {hardware}. Stitched collectively, these items are what let Ericsson declare it’s closing the loop between autonomous networks and autonomous operations.
That is additionally the place the aggressive distinction with operators will get fascinating. Verizon has publicized utilizing agentic AI throughout its 60,000-site vRAN for configuration, service assurance, and optimization, and has referred to as for interoperability requirements for agentic methods. That’s an in-house construct. Ericsson’s framework provides a standardized, vendor-driven various — interesting to operators that don’t need to engineer their very own agentic stack, although it comes with the same old tradeoff of leaning on a vendor’s roadmap and assumptions.
Cloud dependency and hyperscaler deployment
The framework depends closely on AWS. The Telco Agentic AI Studio and the Gen-AI Lab are each constructed instantly on Amazon Bedrock, with some rApp aaS options hosted as SaaS on the AWS Market. The logic is simple sufficient. Hyperscaler AI platforms present the elastic compute wanted to scale agentic workloads, whereas Ericsson retains operational telco information management and governance inside its Telco DataOps atmosphere and inside CSP boundaries. That break up — hyperscaler for compute, telco platform for information — is the corporate’s reply to operators nervous about handing over their information.
Ericsson does current broad cloud-agnostic deployment choices, and the greater than 20 cloud-native AI functions from the 2025 portfolio may run on the Telco IT AI Engine or on hyperscaler platforms. However AWS is clearly positioned because the reference implementation for multi-agent workloads, and that’s the place probably the most developed tooling lives.
That’s a double-edged sweetener. For operators already comfy on AWS, having Bedrock because the confirmed path lowers the barrier to getting brokers into manufacturing. For operators operating non-public information facilities or dedicated to a competing cloud, the heavy AWS dependency raises actual lock-in questions — how a lot of the agentic layer genuinely runs elsewhere, and at what price in functionality. It’s a good query to place to Ericsson, and one CSPs will need answered with specifics earlier than they construct a service expertise technique on prime of it.