AI for Telecom: Legacy Operations vs. the Automated Operating Model
The decision AT&T is making right now is one every large telecom operator will face within the next three years: keep running a people-intensive, hardware-anchored operating model, or move to one where AI handles the repetitive, measurable work and human teams focus on decisions that actually require judgment. AT&T has made its choice. Understanding the trade-offs clearly — rather than treating this as a binary between 'keep everything' and 'automate everything' — is what separates operators who execute well from those who announce pilots and stall.
The two models side by side
Before examining each criterion in detail, here is how the legacy telecom operating model compares to the AI-driven model AT&T is actively building toward, as reported by Wired in September 2026:
| Criterion | Legacy Operating Model | AI-Driven Operating Model |
|---|---|---|
| Customer service requests | Manual agent handling for routine requests (e.g., service disconnects) | Automated workflows handle tier-1 requests; agents handle exceptions |
| Network planning | Field surveys and manual tower-site analysis | AI identifies optimal cell tower locations and flags maintenance issues |
| Network configuration | Physical technician dispatch for hardware changes | Remote software adjustment via cloud platforms (e.g., DriveNets) |
| Weather/event response | Reactive; manual escalation chains | Real-time AI systems like AT&T's GeoModeler adjust settings automatically |
| Software development | Junior developer teams maintain legacy code | Generative AI assists with code generation; fewer junior roles needed |
| Energy profile | Copper wire infrastructure, energy-intensive | Decommissioned copper; cloud-first; measurable electricity reduction |
| Cost structure | Labor-heavy; high capex on physical assets | Shifting toward opex; fewer middle management and junior developer layers |
| Revenue per employee | Lower than peers (AT&T trailed Verizon and T-Mobile in 2025 public disclosures) | Target: materially higher by 2030, approaching ~85,000 headcount from ~131,000 |
Manual hubs vs. cloud-managed hubs
The clearest comparison point is what AT&T is doing with its central network hubs. Under the legacy model, physical hardware at thousands of locations required on-site technicians to make configuration changes — including something as routine as a customer requesting a faster internet tier. Starting in 2027, AT&T will replace that hardware with cloud software from DriveNets, an Israeli startup in which AT&T holds an investment stake. The change means a technician working remotely can push a configuration update in minutes rather than scheduling a dispatch that might take days.
The trade-off is real: cloud-managed hubs create software dependencies and require robust failover planning. A misconfigured update pushed at scale is harder to contain than a localized hardware issue. Operators choosing this path need clear rollback procedures and monitoring at the network-layer, not just application-layer, from day one.
Customer service automation vs. agent-led support
AT&T is using AI for customer service interactions, including the kind of routine request — disconnecting a phone line — that until recently still required a worker to navigate paper records. That is not a trivial detail. According to U.S. Bureau of Labor Statistics data, telecom industry jobs have been declining for 25 years, and AI is now accelerating a trend that predates the current hype cycle by two decades.
The honest comparison here: automated customer service handles volume efficiently but struggles with ambiguous requests, escalations involving regulation, or customers who have already been bounced between systems. The operators getting the best results are those who define the automation boundary precisely — which requests go fully automated, which trigger a human handoff, and how the handoff is logged. That boundary definition is an implementation decision, not a product feature.
Hardware maintenance vs. remote fault detection
AT&T's GeoModeler system adjusts network settings automatically during extreme weather events. Its AI tools also identify maintenance issues on existing cell towers without requiring a field visit to discover the problem. Compare that to the legacy approach: field reports, scheduled maintenance windows, and reactive dispatch when something fails.
The performance difference is significant. McKinsey research on predictive maintenance across industrial operators consistently shows 20–30% reductions in unplanned downtime when AI-based fault detection replaces reactive maintenance. Telecom networks, with thousands of distributed assets and weather exposure, are a strong fit for the model.
The trade-off: AI fault detection is only as good as the sensor data and signal feeds it receives. Legacy infrastructure with incomplete telemetry produces noisy models. Before deploying predictive tools, operators need an honest audit of what their infrastructure actually surfaces in real time.
Capex model vs. opex model
Copper wire infrastructure is expensive to maintain and energy-intensive to run. AT&T's plan to exit copper in more than 85% of its existing footprint by end of 2026 is as much a balance-sheet decision as a technology one. Retiring those assets reduces both maintenance cost and electricity consumption — a dimension AT&T CTO Jeremy Legg specifically cited as a benefit of the shift.
Cloud and software-defined infrastructure moves cost from capital expenditure (buying hardware) to operating expenditure (paying for software licenses and cloud capacity). For most large operators, that improves budget flexibility, but it also means that cost control shifts from procurement to vendor management and contract discipline. Opex models can compound quickly if usage is not monitored.
Revenue per employee: the metric investors are watching
According to public financial disclosures cited by Wired, AT&T generated less revenue per employee in 2025 than either Verizon or T-Mobile — both of which have also reduced headcount in recent months. That benchmark is why AT&T's stated direction makes sense from an investor relations standpoint, even if the human cost of continued workforce reduction is significant.
The comparison is worth being precise about: revenue per employee is a blunt instrument. It does not distinguish between operators who shed valuable engineers and those who successfully automate low-value work. The operators who come out ahead will be the ones who track not just headcount ratios but output quality — network uptime, customer resolution rates, deployment speed — alongside the labor numbers.
For enterprise operators running AI business process automation at scale, the measurement framework matters as much as the automation itself. Deploying AI without changing how you measure performance produces cost reductions that are difficult to attribute and improvements that are difficult to replicate.
What the AT&T example means for enterprise operators
AT&T's transformation is not a replicable blueprint in every detail — a 131,000-person legacy telecom with 150 years of infrastructure is a specific context. But the structural logic applies broadly to any large operator sitting on a combination of manual workflows, aging physical assets, and a competitive benchmark that measures revenue efficiency per employee.
The lessons that travel well are these: start automation with workflows that have clear inputs, measurable outputs, and low regulatory risk; tie AI deployment to specific cost and energy metrics from the first sprint, not as an afterthought; and treat the operating model question — who manages AI systems after deployment — as seriously as the implementation question. AT&T is using AI to generate software code, manage networks, and route customer service, but someone is setting the rules for when GeoModeler acts autonomously and when a human reviews the output.
Building that governance layer before it becomes an incident is the difference between an automation program and an automation operating model.
Verdict: which model fits your stage
Pick the legacy model if your network infrastructure still has several years of depreciation life, your customer base is stable, and you have not yet built the data infrastructure needed for AI tools to produce reliable signal. Rushing AI deployment onto incomplete data pipelines produces errors at scale, not efficiency.
Pick the AI-driven model if you are already running telemetry across your network assets, your customer service queue has identifiable repetitive patterns, and your leadership team is ready to define the human-oversight boundaries before automation goes live — not after. AT&T's path shows the upside is real. So are the implementation risks for operators who treat this as a product purchase rather than an operating model redesign.
Related reads
Martin Kuvandzhiev
Co-Founder & CEO, encorp.ai
CEO and Founder of Encorp.io with expertise in AI and business transformation
LinkedIn