AI is not killing service integrators. It is redefining the winners

For much of the past year, Wall Street and its pundits have been writing an obituary for service integrators. The logic sounds simple. AI can generate code, resolve incidents, test applications and increasingly execute workflows. If machines can do more of the work, enterprises will need fewer people, fewer billable hours and, therefore, fewer integrators.

There is a useful truth inside that argument. AI will compress large parts of the traditional services model. Repetitive work will be automated. Productivity gains will flow through to pricing. Clients will no longer accept a straight-line relationship between headcount and value. Any provider that tries to protect the old labour pyramid should expect a difficult future.

But the death of a delivery model is not the death of an industry. A recent HFS analysis makes the more important point: AI and services need each other, because the main barrier to enterprise AI is not access to a model. It is the accumulated technology, data, process, and talent debt that prevents models from operating safely and effectively at scale. HFS estimates that the burden across the Global 2000 is $18 trillion and uses the term “Services-as-Software” for the new category emerging between traditional services and software.

AI is not a protective moat for incumbents. It is a forcing function. It will expose providers that sell capacity without differentiation. At the same time, it will reward firms that combine deep engineering, enterprise context, platforms, change leadership and operational accountability. That is a more demanding role than traditional systems integration, but also a more valuable one.

Adoption is widespread. But is it worth the cost?

The industry has spent the last couple of years measuring AI adoption. That is becoming the wrong scoreboard. Because, despite the cost of the token dropping rapidly, from $60 per million tokens to $0.06 within a span of 5 years, the bill for enterprises is increasing as agentic AI uses far more tokens than ever before. Hence, it’s not surprising that more and more companies are now either limiting AI usage or nudging their employees to use AI more “productively”.

This is before we get into the weeds of meaningfully integrating AI into an enterprise’s tech stack. A model can perform brilliantly in a controlled environment and still struggle inside a bank, manufacturer, insurer or healthcare provider. Real businesses run on decades of applications, integrations, policies, exceptions, security controls, and institutional knowledge. Data ownership is often unclear. Process steps have grown around regulatory obligations and historical workarounds. The people doing the work know where the formal process ends and the real process begins.

This is where the hardest part of AI starts. Someone has to connect the model to systems of record, establish data lineage, manage identity and permissions, redesign workflows, define human escalation, test failure modes, monitor performance and prove compliance. Someone also has to prepare the workforce and take accountability when an automated decision affects a customer, patient, payment or production line.

Gartner’s recent research on AI in infrastructure and operations illustrates the point. Only 28% of use cases fully succeeded and met ROI expectations, while 20% failed outright. The strongest success factors were not more sophisticated models. They were integration into existing workflows and systems, realistic business cases, executive support and cross-functional execution.

That work sits squarely in the territory of a modern service integrator.

AI reduces the cost of tasks, then expands the ambition of transformation

The bearish view of services assumes demand is fixed. If AI reduces the effort required to complete a task by 30%, revenue must fall by 30%. In some mature service lines, that will happen. Routine application maintenance, basic testing, level-one support, documentation, and standard configuration will face real deflation.

Yet enterprises rarely have a fixed backlog of change. They have a backlog constrained by cost, risk, skills, and management capacity. When AI makes modernisation faster, testing cheaper and operations more automated, projects that were previously uneconomic become possible. The unit cost of change falls, but the scope of change can rise.

The same dynamic is reshaping software. Gartner estimates that up to $234 billion of enterprise application spending could be exposed to “agentic arbitrage” by 2030 as agents execute work across multiple systems. Importantly, Gartner also says this creates a substantial opportunity for service providers that build agentic, cross-domain workflows, redesign work and deliver measurable outcomes. It notes that these solutions currently require heavy services engagement.

This is why the boundary between software and services is becoming less useful. The client does not ultimately want a licence, model, token or FTE. The client wants a claim settled correctly, an outage prevented, a product engineered faster, working capital released or customer churn reduced. Gartner expects more than half of enterprises to favour platforms that commit to workflow results over assistive AI by 2028, with enterprise context, permissions, policy enforcement and auditability becoming central to execution.

That future plays to the strengths of integrators, but only if we change our economics. We have to move from selling effort to pricing outcomes. We have to turn repeatable knowledge into platforms, agents, accelerators, and managed services. We have to share productivity gains with clients rather than hiding them. And we have to retain human accountability even as more execution becomes autonomous.

What the next-generation integrator must become

First, we must become foundation fixers. AI cannot compensate for fragmented data, brittle architecture or uncontrolled technical debt. Modernisation is no longer a prelude to the “real” AI programme. It is the AI programme. The winning partner will connect application modernisation, data engineering, cloud, infrastructure, cybersecurity and operations into one value agenda.

Second, we must become orchestrators of enterprise context. No single model provider, hyperscaler or software company owns the full reality of a client. Large enterprises will remain multi-cloud, multi-model and heterogeneous. The integrator’s role is to make that ecosystem work as one, while preserving portability, resilience, sovereignty and control. The World Economic Forum’s 2026 work similarly argues that value comes from end-to-end operating model redesign, human accountability, scalable talent systems, transparency and disciplined experimentation.

Third, we must redesign work with people, not simply automate around them. A recent study found that AI Leaders were far more likely than followers to have organisation-wide upskilling strategies, trusted data foundations, measurable use cases and visible leadership sponsorship. Adoption accelerates when people understand how their work will change, help shape that change and can see the value for customers and employees.

Fourth, we must accept greater commercial accountability. Outcome-based services require better baselines, stronger observability and shared clarity on which party controls each variable. Providers need the confidence to commit to performance, and clients need the discipline to share data, simplify processes and make decisions quickly. The contract cannot carry a transformation that the operating model refuses to support.

The real question for clients and providers

Enterprise leaders should not ask service partners for the same work at a lower rate simply because AI exists. They should ask which work can be eliminated, which foundations must be repaired, which workflows can be reinvented and which business outcomes the partner is prepared to own. A rate-card negotiation may reduce this year’s cost. A redesigned operating model can change the economics of the business.

Service providers should be equally honest. AI will not preserve every revenue stream, role or company. Some work will disappear. Some firms will struggle to fund the platforms, talent and intellectual property needed for non-linear growth. The market is right to discount undifferentiated labour. It is wrong, however, to equate fewer hours with less need for transformation.

AI makes individual tasks easier. At enterprise scale, it makes the system more interconnected, more dynamic and more consequential. Complexity does not vanish. It moves upward, from producing outputs to governing outcomes.

The obituary for service integrators was written too early. The winners will look less like staffing businesses and more like software-enabled operators with deep industry knowledge, engineering strength and end-to-end accountability. Our opportunity is not to defend the old model. It is to help build the new one, making AI useful, trusted and economically meaningful in the messy environments where business actually runs.

(The author is Corporate Vice President and Global Head, Digital Business Services, HCLTech. Views are personal.)