Large enterprises have spent the past two decades modernizing their technology estates. Yet despite repeated waves of transformation — enterprise resource planning (ERP) upgrades, cloud migration, application rationalization, and platform modernization — most still run large, complex legacy IT environments that support critical processes and core data.
As a result, too significant a portion of enterprise IT budgets remains tied up in merely keeping systems running. Across the technology stack, systems demand thousands of daily interventions to maintain stability, resolve incidents, manage releases, monitor performance, and ensure compliance. Such outsize effort to run legacy technology estates is becoming a direct constraint on productivity, efficiency, and competitive speed.
In an AI-first world, negotiating these constraints is no longer tenable. But AI itself can enable the redesign of how legacy environments are operated and simplified. Rather than focusing solely on outsourcing or system replacement, organizations can now rethink how these systems are managed day to day. AI adds a new operational layer that can execute tasks, orchestrate workflows, and process operational signals at scale — reshaping the economics and operating model of run-the-business IT (or “Run IT”).
Why the current model of running legacy IT is structurally unsustainable
Traditional legacy IT operating models attempt to manage extraordinary technical complexity through layers of human coordination, which in turn creates organizational complexity. That approach can preserve continuity, but it is increasingly uneconomical to scale and, over time, introduces meaningful resilience risk. The challenge's aspects are as follows.
Legacy IT’s operating environment is inherently complex. Large enterprises operate vast, interdependent technology estates spanning hundreds or thousands of applications, layers of infrastructure, fragmented data, and distributed ownership. Even basic activities such as incident resolution or release coordination span systems and organizational boundaries, making execution difficult.
Operational signals have exploded. The volume of data, alerts, logs, tickets, changes, and vendor inputs now far exceeds what human teams can interpret and act on in real time. The constraint is no longer data access, but speed of understanding and response.
The increased adoption of AI in software development and business activities is significantly increasing the complexity of managing Run IT. The volume of new software being delivered and the level of risk are increasing steadily. Humans will not be able to move quickly enough or coordinate enough to track and scale with the increased complexity without AI at the core of running activities.
The cost of coordination has become excessive. In most enterprise IT environments, the dominant cost driver is now the organizational overhead required to move work across fragmented teams and disconnected systems. Entire layers of effort — tickets, playbooks, expert judgment — exist solely to bridge activities that the technology estate cannot coordinate autonomously. Compounding this, labor costs continue to rise while hardware costs decline at equivalent performance levels — a divergence that makes human-mediated coordination increasingly difficult to justify at scale.
The talent model is increasingly fragile. Operations depend on a small group of experienced, often older operators with undocumented, hard-to-replace knowledge. As they exit, resilience erodes.
Full replacement of legacy estates is not usually an option. These systems often support critical processes such as billing, claims processing, financial reporting, or supply chain management, and have major customizations and workflows built on top of systems of record. As a result, replacing them completely is often too costly, time-consuming, or risky, even in large-scale transformations.
How AI orchestration is redefining the legacy IT operating model
AI and digital agents enable organizations to efficiently overhaul the operations of legacy IT environments. Ultimately, AI represents an entirely new operating model. Here are the dimensions of the transformation.
- Operational work: AI can absorb large portions of the operational work that historically required human effort. It can interpret unstructured and structured signals, identify patterns across tools, generate recommendations, trigger workflows, document actions, and coordinate tasks across systems with much greater speed and consistency than traditional manual models. Digital agents extend these capabilities further by actively executing work within operational processes, rather than merely providing insights.
- System connectedness: AI serves as a connective layer to overcome the challenges of fragmented legacy environments. Instead of relying solely on human coordination and communication to connect processes across systems, organizations can use AI to provide a consolidated view of activity or incidents across tools, logs, configuration data, and historical tickets almost instantly.
- Human labor: With digital agents handling more and more operational execution, the focus of human labor shifts to supervision and decision making. Engineers review AI recommendations, intervene in complex or ambiguous situations, approve high risk actions, and refine policies.
- Organizational design: Over time, leading IT organizations will look less like collections of operational teams and more like hybrid human‑machine systems with integrated, coordinated, real-time workflows. Productivity will no longer scale with labor but through the combination of AI execution, agent orchestration, and human oversight, building operating leverage into the system itself.
AI materially improves speed, cost, and resilience in legacy IT
Embedding AI inside the flow of work and redesigning the operating model around it can produce meaningful improvements across several dimensions.
- Speed: Activities across incident, release, and support processes that currently take hours or days due to queueing and handoffs can often be compressed significantly with digital agents working continuously across data sources.
- Productivity: When AI takes over low-value parsing, synthesis, and routing workloads, human teams can focus on exceptions and broader issues such as service quality and structural improvement. As a result, output per full-time employee improves even before headcount is reduced.
- Cost: In addition to labor substitution, cost decreases markedly through lower rework, faster recovery, fewer escalations, better use of expert capacity, tighter vendor oversight, more consistent execution, and reduced service disruption.
- Resilience: Through codification and operationalization of knowledge, AI can reduce dependency on concentrated human expertise or scattered documentation, thereby improving continuity as workforce composition changes.
Top five priorities to redesign the run IT operating model
Scaling AI-enabled Run IT requires more than implementing copilots or point solutions. It demands deliberate choices across architecture, workflows, data, talent, and the overall transformation approach.
The first priority is to identify where AI can intervene directly in core operational flows. Many organizations begin with generic use cases, optimizing individual productivity with copilot-like solutions, but the greatest value comes from embedding AI into workflows such as incident and service management, releases, batch operations, application support, and vendor coordination.
The second priority is to build an orchestration layer, enabled by digital agents, that operates across fragmented systems. AI delivers the most value when connected to tickets, logs, telemetry, configuration management databases (CMDBs), runbooks, release artifacts, vendor inputs, and collaboration tools. This connective layer transforms AI from a source of information to an engine of action.
The third priority is to strengthen the data and knowledge foundation. Legacy environments rarely have clean, unified data. Treating knowledge capture, data structuring, and interface rationalization as enabling moves is essential for scale. While unglamorous, this work is critical — and AI can expedite it.
The fourth priority is to execute targeted quick wins and institutionalize learning. Organizations should fully transition specific, well‑bounded Run IT activities to AI‑enabled workflows and digital agents operating in production, with humans focused on supervision and exceptions. These scoped replacements generate immediate, visible progress while revealing where autonomy works reliably, where judgment is still required, and how controls and escalation thresholds should be set. Organizations should use AI agents to help build the future IT model, thereby shortening the path to transformation considerably.
The fifth priority is to fundamentally redesign the workforce and governance model. Organizations must be deliberate about what is owned by humans versus digital agents, what requires human approval, how operational risk is managed, and how operator roles evolve over time. Reskilling is therefore a core design variable, not a peripheral adjustment.
Governance must equally rise to the challenge of managing agents at scale. This requires a structured framework encompassing data harness and guardrails alongside explicit delegation mechanisms to version and update agents, manage token economics, and maintain auditability. The emerging architectural principle reinforces this: Rather than proliferating hundreds of point-specific agents, leading organizations are consolidating around a smaller set of deeply capitalized, domain-specific operational agents. The winning model is not human versus AI, but a deliberately engineered collaboration in which each performs what it does best.
The future of legacy IT depends on how it is operated
For CxOs, the key question is no longer if legacy IT will remain in most large enterprises — it will. The more important issue is whether the current model for operating and evolving the estate remains effective. In many cases, it is not. The future of legacy IT will not be defined by what systems organizations replace, but by how they operate intelligent existing systems.