Service providers have made significant investments in network automation over the past decade. The intent has always been clear: reduce manual operations, accelerate service delivery, and build the agility to compete in a market that never slows down.

The results have been real – but incomplete.

The well-understood workflows are largely automated. Standard provisioning, compliance checks, routine change management – those are running on platforms built and refined over years of hard work. But the operations that consume the most labor, carry the most risk, and generate the most organizational friction? Those remain stubbornly manual. And at carrier scale, that gap compounds quickly.

A new independent analysis from Appledore Research, a firm focused exclusively on the telecom and service provider market, examines what’s required to close it.

Appledore’s research on Itential’s approach identifies a core architectural distinction that separates effective agentic operations from AI that simply augments existing workflows.

Most vendors in the market are layering AI onto automation frameworks that were designed for a different set of problems. The result is a system that is still fundamentally rule-bound – faster, perhaps, but not smarter. Appledore characterizes Itential’s FlowAI technology as “AI first, not AI augmented,” and frames that as a meaningful distinction, not a marketing position.

The difference lies in how the two types of work are handled. Deterministic, well-understood tasks – the ones that already run reliably on automation platforms – don’t require AI reasoning. Applying an AI layer to a task like standard VLAN provisioning adds overhead without adding value. But the complex, contextual, multi-domain operations that defeat traditional automation? Those require a system that can interpret context, weigh risk, and reason toward an outcome. Agentic AI, applied correctly, is architected for exactly those problems.

Appledore specifically calls out Itential’s MCP-native architecture as forward-thinking in the context of where the service provider market is heading.

The service provider environment is the hardest place to operate infrastructure automation at scale. That’s not an opinion – it’s the reality of what CSPs manage: layered physical and virtual infrastructure, proprietary domain controllers, BSS/OSS systems with decades of operational logic, and brownfield environments that don’t behave consistently across regions or vendors.

Traditional automation handles the predictable edge of that environment well. But the operational complexity that lives in the middle – cross-domain incidents, multi-layer changes, service assurance across hybrid infrastructure – has always required experienced engineers in the loop because no workflow could fully encode the context required to act correctly.

This is precisely the problem agentic orchestration addresses. Not by replacing the deterministic workflows that already work, but by giving the complex, open-ended problems the reasoning capability they need.

For a service provider, that distinction is not academic. Getting it wrong in a carrier environment means downtime measured in customers, not servers.

One of the themes Appledore’s analysis reinforces is that trust in agentic systems cannot be assumed – it has to be built.

This mirrors what we’ve seen in practice with service provider customers. The path to agentic operations doesn’t start with autonomous execution. It starts read-only: agents that observe, analyze, and surface insight before they’re ever trusted to take action. Role-based access controls, policy enforcement, and integration with existing change management processes aren’t constraints on AI – they’re what makes it deployable in a production network.

At Itential, we’ve built our FlowAI technology around this principle. AI agents reason and recommend within guardrails defined by the operator. Every proposed action moves through established workflows, validations, and approvals before execution. AI doesn’t bypass operations – it becomes part of them.

The operational story that best illustrates this in a service provider context is Lumen.

Lumen’s automation journey with Itential didn’t start with agentic AI – it started with a commitment to automating the well-understood workflows first, building the governed execution foundation that agentic operations require. What began as 16 workflows has grown to over 350, with the scope expanding from routine operational tasks toward the complex, multi-domain changes that previously required significant manual effort at every step.

As Greg Freeman at Lumen put it: agentic AI is “AI that takes action.” That’s the north star. Not AI that generates recommendations, but AI that can move through governed steps to deliver outcomes, with human oversight at the points where the risk profile demands it.

The service provider industry is at a specific inflection point. Pressure to modernize is intensifying – from hyperscalers entering the connectivity market, from enterprise customers demanding cloud-like service agility, and from the internal economics of running increasingly complex networks with operations models built for a simpler era.

At the same time, the technology required to address this is ready in a way it wasn’t two years ago. Standards like MCP provide the auditable interface between AI reasoning and infrastructure systems. Domain-specific agents trained on network operational knowledge are delivering the accuracy required for production environments. And the orchestration platforms that serve as the governed execution layer have matured to support agentic workflows safely.

The ones that wait will find the gap harder to close.

Independent analysis from Appledore Research positions Itential’s approach as one of the most credible paths forward for service providers navigating the transition from traditional automation to governed agentic operations.

If you want the complete breakdown, including the full examination of Itential FlowAI and how it’s redefining what’s possible for service providers and telcos, click here or below to read the Appledore report.