Can AI fix legacy IT? The ‘economics don’t hold up,’ says former IBM strategist

Only 10% of surveyed IT chiefs said they use agentic AI for modernization, while 18% reported limited or unclear value from modernization efforts and nearly half said projects are behind schedule or over budget. Kyndryl says agents show early promise in mapping dependencies, generating code and documentation, but an IT expert cautions that AI could add unpredictable compute costs; three in 10 leaders already worry about that risk.

Analysis

Investment read-through: the opportunity is more likely to be an extension of enterprise IT spend than a near-term replacement cycle. If agents reduce the manual work of mapping dependencies and documenting systems, they could unlock projects that stalled on execution capacity—supportive for systems integrators and modernization services in the next 1–3 months only if bookings and backlog begin to reflect it. Kyndryl’s survey is directional evidence, not proof of realized customer ROI; low adoption and persistent overruns argue against capitalizing broad productivity gains today.

The less obvious beneficiary may be cloud and infrastructure providers: agents that inspect and orchestrate hybrid estates can increase compute and data-transfer demand before any legacy systems are retired. That creates a two-sided outcome—more workloads and services revenue, but also customer scrutiny of unpredictable bills, potentially slowing deployment or shifting workloads toward private infrastructure. For integrators, automation may lift delivery capacity initially, yet eventually pressure labor-based revenue if contracts do not shift toward outcome pricing.

Timing: immediate market impact should be limited; over 1–3 months, watch enterprise bookings, modernization backlog, and cloud-spend commentary. Over 6–18 months, the thesis depends on measurable decommissioning and lower project overruns, not agent demos. Contrarian point: AI may make modernization more feasible while making the estate larger and more complex, creating tomorrow’s legacy debt. Falsify the constructive case if adoption rises but services backlog or completed migration milestones do not, or if customers cite rising AI infrastructure bills and defer projects.

Key Decisions for Investors

• No broad AI or IT-services trade on this survey alone. Treat it as an adoption watch item; require reported contract wins, backlog conversion, or shorter delivery cycles before paying for a modernization acceleration thesis.
• Monitor Kyndryl and other systems integrators’ next earnings for modernization pipeline conversion, utilization, and project overruns. A rising pipeline without completed milestones is a negative signal, not evidence of monetization.
• Track cloud-provider commentary for AI-related infrastructure consumption alongside customer cost-control or workload-repatriation signals. Rising usage without durable customer ROI could ultimately constrain deployment rather than support a sustained spend cycle.
• Over the next 6–18 months, favor evidence of legacy-system retirement and outcome-based contract growth over agent announcements. Reassess if customers report lower delivery costs but no increase in modernization budgets, which would imply productivity benefits accrue to buyers rather than vendors.