Ericsson and Telstra show autonomous networks need intent and knowledge, not just more bots
Key Takeaways
- Treat knowledge representation as core infrastructure, not documentation, before adding autonomous agents.
- Define intent as desired outcomes, then connect it to validated operational context and governance.
- Watch Ericsson and Telstra’s working group for reusable taxonomies, exchange patterns and agency models.
Why it matters
- ProductProduct leaders can use this as a template for moving from task automation to outcome-driven AI systems.
- InvestorsThe value may accrue to platforms that encode operational knowledge, trust and governance across complex infrastructure.
Their collaboration points to a less flashy telco AI lesson: encode what the network should know before asking it to act.
Network automation has always had a whiff of haunted spreadsheet: scripts everywhere, dashboards blinking, humans still coordinating the important bits like air traffic controllers with worse coffee. Ericsson and Telstra are now pushing a less glamorous, more useful thesis: autonomy is not mainly about adding another AI agent with a tiny hard hat. It is about giving the network a shared model of intent and operational knowledge, so actions do not depend on tribal memory and heroic Slack archaeology.
That sounds obvious until you try to build it. Most AI automation projects start by asking, “What task can we automate?” Ericsson and Telstra are asking the more annoying question: “What does the system need to know, trust and govern before it is allowed to do anything?” Annoying questions, tragically, are where the engineering lives.
Ericsson says the hard layer is knowledge
According to Ericsson’s announcement, Telstra and Ericsson are expanding their autonomous networks collaboration, with the next phase focused on common approaches to network knowledge. Ericsson frames that work around the knowledge, trust and governance foundations needed for networks that can operate across domains and technology environments. Telstra’s Mark Sanders put it bluntly, saying autonomous networks require “more than the automation of individual tasks.”
Ericsson says the companies will co-lead a joint working group and publish its outcomes. The work will examine three areas: common knowledge taxonomy, knowledge exchange and composition, and knowledge-grounded agency. In human terms, that means agreeing on the vocabulary, moving and combining the relevant context, then letting software agents act on it without turning operations into a very expensive improv troupe.
Intent needs knowledge
- Intentoutcome to achieve
- Knowledge Planenetwork representation
- Agentic systemplan validate execute
That is the useful bit for builders outside telecom too. If your system cannot represent the world it is changing, what good looks like, and how confident it is in its own context, “autonomous” is just “automated, but with better conference lighting.”
RCR Wireless says this is not just more AI sprinkled on OSS
RCR Wireless News reports that the collaboration is not simply about adding more AI. It describes the shift as architectural and organizational, moving from human-defined workflows toward intent-based systems that can plan, validate and execute network changes safely. That is an important distinction because a workflow engine with an LLM bolted on is still a workflow engine, just now it occasionally writes poetry about packet loss.
RCR quotes Ericsson’s Mats Karlsson saying intent and knowledge are the two core ideas behind autonomous networking. Karlsson also contrasts conventional automation with autonomy: existing systems still rely on human-defined flows and configuration, while intent-based networking asks the operator to specify the desired outcome. Or, as Karlsson put it, the network needs to know “what outcome” it should achieve.
That distinction matters because telecom networks are not tidy demo apps. RCR reports that the partnership, announced in 2025, is meant to create, validate and industrialize business-driven use cases while addressing fragmented data, the gap between business intent and network execution, multi-vendor complexity and AI trustworthiness. Translation: the hard part is not making a model suggest a change. The hard part is making sure the model knows which change is safe, explainable and compatible with the mess already running in production.
The Knowledge Plane is the interesting architecture bet
Ericsson describes the Knowledge Plane as a formal representation of what the network is, what it should be and what it is expected to do. That is a big clue about where telco AI may be heading: away from isolated copilots and toward shared machine-readable context. In ML terms, this is less “chat with the network” and more “give agents a structured operating model so they do not wander around the data center like raccoons in a lab coat.”
RCR Wireless News reports Karlsson’s view that knowledge is about relationships encoded into data so machines can reason over them and act. That makes this a knowledge representation problem as much as an automation problem. The AI model is not the whole product. The ontology, confidence model, governance rules and data plumbing are the product, even if they do not get the keynote applause.
For developers building agentic systems in any domain, the lesson is portable. Before adding another planner, tool caller or agent framework, ask whether your system has a shared representation of goals, constraints, dependencies and trust. Otherwise you are not building autonomy. You are building a very confident intern with root access, which is how incident reports become literature.
What to watch next
Ericsson says the joint working group will publish outcomes and identify follow-on research themes, so the next useful signal will be whether this work becomes concrete enough for broader industry alignment. Watch for definitions that operators and vendors can actually share: taxonomies, exchange formats, governance patterns and examples of knowledge-grounded agency that survive contact with multi-vendor reality. The press release vocabulary is promising, but production networks are where adjectives go to be audited.
For readers building AI systems, the takeaway is not “go copy a telco architecture.” It is smaller and sharper: intent without knowledge is vibes, and knowledge without governance is a haunted database. Ericsson and Telstra are reminding the AI crowd that autonomy, annoyingly, still starts with knowing what you meant.
Sources2 sources
The reporting, announcements and research the AI editor worked from. Links open the original publisher.
