The sensible distinction turns into clear in troubleshooting workflows. When requested to triage a ServiceNow ticket, the agent reads the ticket content material, gathers context about entities talked about from the digital twin, routinely performs path traces for connectivity points, and returns a analysis. The whole workflow stays seen to operators all through the method.
This differs from easy pure language queries the place the system interprets a query and returns a solution. The agent is constructing and executing a plan which will contain a number of information sources and evaluation steps.
Customized framework for context management
Ahead Networks constructed its personal agentic framework slightly than adopting current instruments like LangChain or CrewAI. The choice centered on sustaining exact management over context engineering, which Handigol described because the core engineering problem for agentic AI.
“We constructed our personal as a result of we needed full management over how the agent executes,” Handigol stated. “The engineering drawback largely comes right down to context engineering. How do you outline and preserve the context that’s vital for the agent, that you simply ship to the LLM, to get the precise solutions?”
The workforce defines context engineering as offering all related data with out extra noise. Too little data produces unsuitable solutions. An excessive amount of data distracts the mannequin from the right process.
Context is drawn from Ahead Community’s hierarchical information stack. On the base layer sits uncooked configuration, state, and statistics collected instantly from gadgets. The following tier normalizes this uncooked information right into a queryable mannequin displaying all the things current within the community and its configuration.
