Network engineering has, over the years, involved planning, long change windows and a deep risk aversion, where the network was the one piece of infrastructure nobody wanted to touch without a rollback plan and a weekend to spare. That made sense, but a badly configured push could bring down everything that depended on connectivity in the first place. However, software-defined networking, cloud-native functions, programmable APIs and now AI agents are slowly rewriting that culture, pushing networking teams towards the same continuous-deployment mindset that’s been a standard in application development for a long time.
The debate is not about whether networks are becoming programmable. They are, considering the tools, resources and adoption curves, moving in the same direction. However, the harder question that arises is whether such programmability is creating agility in a meaningful way, which can be seen by software teams as the ability to make changes quickly, safely and quite frequently, with a high level of confidence in low-cost failure recovery.
Programmability is real and spreading
Numbers help capture the magnitude of change. Gartner indicates that by 2026, 30% of enterprises will be capable of automating over 50% of network operations from the current 10% recorded in mid-2023. The majority of changes are attributed to the adoption of intelligent automation and AI-driven analytics that seek to enhance operational efficiency, resiliency and agility of infrastructure and operations teams. Furthermore, 50% of enterprises will use AI capabilities to automate ‘day two’ network operations by 2026, up from fewer than 10% in mid-2023.
In addition, the market has taken a similar course. Mordor Intelligence predicts the global intent-based networking (IBN) market will rise from about US$3.04 billion in 2026 to $8.47 billion by 2031, representing a compound annual growth rate (CAGR) of 22.74%. Enterprises using AI-driven orchestration platforms to execute business intentions through network configuration are projected to drive the market.
Telecoms operators have gone furthest. Analysys Mason and Nokia found that optical network automation cut network lifecycle management costs by up to 56%, cut service delivery labour costs by up to 81%, avoided up to 26% in capex through better resource use and shortened service order fulfillment from an average of 10 days to 24 hours. That is agility measured in hard operational terms, not just process language.
Yet Analysys Mason’s own Open Network Index, a survey of 50 Tier 1 operators, identified a gap between belief and execution. Roughly 90% of operators view open, programmable networks as critical to their survival, yet only about 20% had an open network strategy actually in place.
Where the friction lives
Although challenges are concrete, they manifest in four different ways. Firstly, in the dependencies involved between every level and the ones below; next, in the need for reliability that stands in the way of iterative innovation; thirdly, in the abilities of the teams involved in implementing change; and finally, in the extent to which AI adoption has gotten beyond being a pilot project.
- Legacy dependency chains: Networks underpin every other agile initiative running on top of them. A sprint cannot ship a feature if the transport, routing policy or security posture beneath it has not moved in step.
- Reliability that punishes iteration: Frequent small releases suit stateless applications. They work poorly for control planes carrying live voice, financial transactions or industrial traffic, where a rollback is not a quick redeploy.
- Skills, not just structure: McKinsey‘s benchmarking of over 20 telecoms operators worldwide found that operators realise full benefits of agile only when they extend the transformation beyond team structures to include portfolio management, talent management and architecture simplification. Operators that focused primarily on reorganising teams without uniformly changing operating models, processes and legacy systems struggled to achieve faster delivery and innovation.
- AI adoption stalling on ground: IDC’s 2026 AI in Networking research found that organisations already at ‘select use’ of AI in networking mostly stayed there over 18 months rather than progressing to substantial deployment, with complexity rather than appetite cited as the blocker.
- Manual approach continues to reign: Enterprise Management Associates (EMA) disclosed that 64% of enterprise networking teams use in-house software or scripts to automate the network. And 61% of those teams spend six or more hours per week on maintaining and debugging those tools, which shows how automation itself often adds extra operational overhead.
The skills gap is widening even as budgets grow. EMA’s Network Management Megatrends 2026 report found that only 31% of respondents considered their network operations strategy completely successful, down from 42% in the 2024 edition of the biennial study. This comes marked with talent shortage, tool sprawl and unmanaged AI workloads cited as recurring pressures. Spending is moving the opposite way.
In its fourth annual State of AI in Telecommunications report, NVIDIA says that 89% of those surveyed are planning on spending more on AI in the next 12 months than they did last year, which was already 65%. Without a doubt, the cash is coming in faster than the talent to spend it.
Agentic NetOps is taking shape
Gartner’s newer research frames the next phase as ‘agentic NetOps,’ where AI agents sense, reason and act on network tasks inside defined guardrails rather than merely suggesting fixes. Gartner estimates that adoption of agentic NetOps is below 1% of organisations, reflecting limited vendor capabilities and organisational readiness. The firm predicts that by 2030, about half of organisations will be using agentic NetOps with minimal human involvement, up from almost 0% in 2025.
Agentic NetOps with a network AI assistant will be the primary user interface for networking operations. That’s the economic stakes that keep vendors and consultancies pushing past pilots.
Deloitte released its Agentic AI Blueprint for Telcos, stating that ‘agentic AI could create US$150 billion for the telecoms industry in the next 5 years.’ Traditional AI typically offers incremental improvements, while agentic AI systems can reason, act and adapt across core telecom functions such as network management, service delivery and customer care.
Complexity is also coming in from the fact that network teams are not fully in control. Last year’s cloud disruptions necessitated a hard truth. According to Juniper Research, firms are reconsidering their single-cloud approach in favour of a multi-cloud approach that is resilient enough to reduce downtime and prevent interruptions regardless of any additional complexity.
While analysts say that 2026 will face challenges in terms of execution and adoption rather than technology itself, whether businesses can adopt automation and resilience approaches without taking on additional risks. For network teams, that shifts what ‘agility’ even means, now has to be designed across cloud boundaries they don’t own, not just infrastructure they directly control.
Reality of agile network engineering
Network engineering can absorb agile practice, but mainly in the layers built to tolerate iteration, such as orchestration, service provisioning and policy on programmable infrastructure. Physical layer, core routing and anything touching carrier-grade reliability will keep running on change control cycles closer to traditional release management, regardless of what the org chart calls itself.
The realistic path looks less like copying software sprints onto network teams and more like automation absorbing work that once needed a change window. Agentic tools handle routine execution under human-set guardrails and engineers are redirected toward architecture and resilience. Agile network engineering will not resemble agile software development, but appear like speed applied selectively, with humans still making the calls that carry real consequences when they go wrong.
Where agility meets reliability
The future of network engineering is unlikely to mirror software development perfectly. Rather, what is taking place is the development of a more pragmatic approach to network design, which involves integration of programmable infrastructure and automation together with the classic sciences of resilience and safety.
The question now is not whether network engineering can go agile but how agile organisations can be without jeopardising their safety. Gartner’s Agentic NetOps research offers a sense of the pace, where adoption is expected to grow from under 1% of organisations today to 50% by 2030, with agentic NetOps becoming the primary interface for network operations, marking a clear signal of where things are headed,
Network engineering is becoming more software-centric, but it’s unlikely to become software engineering outright. Organisations that pair programmable infrastructure and AI-assisted operations with operational discipline and carrier-grade networks are likely to come out ahead. Going forward, agility won’t be measured by how fast teams can change the network, but by how safely they can do it.
Anna Ribeiro