By Dr. Harish Kotadia, Ph.D.
In my last post titled How Agentic AI Is Restructuring the IT Services Model, I explained why the market repriced IT services. This post is about the practical question a CIO has to answer next. If the delivery pyramid is becoming a diamond, where should each kind of work actually sit?
There are four ways to get IT work done. Your own people, onsite. Your own captive center, offshore. A vendor’s people, onsite. A vendor’s people, offshore. For thirty years the choice was mostly about cost and control. Agentic AI changes the calculation. It attacks the cheap, high-volume layer the offshore models were built on, and it rewards the judgment and governance layer that sits at the top. Here is how the four models compare from where a CIO sits:
CIO view of the four IT services delivery models in the age of agentic AI
| Delivery Model | Pros (CIO view) | Cons (CIO view) | Best-Fit Work |
|---|---|---|---|
| In-house, onsite Your own staff at your location |
Full control and IP retention. Deepest business and domain context; proprietary agents and data never leave. Agentic AI lets a small team punch far above its weight, making in-sourcing newly affordable. |
Highest cost per person. Senior AI and architecture talent is scarce and expensive. Fixed cost with utilization risk; slow to ramp up or down. |
Core differentiating systems, proprietary agent orchestration, enterprise architecture, governance, and anything touching sensitive data. The top of the diamond. |
| Client captive, offshore Your own GCC in a low-cost location |
Cost arbitrage with ownership. |
High setup cost and lead time. Management, attrition and entity overhead; needs scale to justify. Distance and time-zone gaps from the business; offshore data governance. |
Sustained, high-volume engineering and operations; building and running proprietary platforms and agent pipelines. Strategic enough to own, cost-sensitive enough to offshore. |
| Vendor, onsite Vendor staff at your site, onshore |
Specialist skills and scale without hiring risk. |
Highest vendor rate tier, and the one agentic AI squeezes hardest. Clients question premium onsite rates when agents do the work. Knowledge leaves when the contract ends. |
Advisory, complex integration, transformation and change programs, regulated work needing presence, and the orchestration plus governance of agentic delivery. The vendor’s defensible niche. |
| Vendor, offshore Low-cost outsourcing |
Lowest vendor cost and largest delivery scale. |
The layer agentic AI hits hardest; the labor-arbitrage value erodes as agents absorb junior work. FTE and time-and-materials pricing breaks. Least differentiated; prone to price commoditization. |
High-volume, standardized, rules-based work such as application maintenance, testing, L1 and L2 support, and data processing. This is exactly the work agents automate, so the natural workload here is shrinking. |
Cost and control are only half the decision. The other half is what each model does to your budget, and to the vendors who depend on your spend:
How each model reshapes client IT budgets and vendor revenue
| Delivery Model | Impact on Client IT Budget | Impact on IT Services Revenue |
|---|---|---|
| In-house, onsite | Shifts from contractor opex to fixed internal headcount plus AI tooling and platform licenses. Agentic leverage lowers the team size needed, making in-sourcing viable at lower total cost than before, though tooling spend rises. | Negative for vendors. Work is pulled in-house. The clearest form of disintermediation. |
| Client captive, offshore |
Large upfront and run cost to establish, but the lowest long-run unit cost for owned capability. Converts vendor margin into internal offshore cost plus tooling. Favorable total cost of ownership once at scale. |
Strongly negative. Budgets that once funded vendors now fund captives. This is the core in-sourcing threat. |
| Vendor, onsite |
Highest hourly cost, but billable hours shrink as agents absorb work. Spend shifts from large onsite teams to a small senior advisory layer plus agent oversight, with a premium for governance and accountability. |
Volume declines but value per role holds. Revenue is protected only if the vendor moves up into outcomes and governance. Pure onsite staff augmentation shrinks fast. |
| Vendor, offshore |
Historically the cheapest, and agentic AI compresses it further. Clients expect price cuts as vendors automate, forcing a shift from FTE and time-and-materials to outcome and Service-as-Software pricing. The largest single line-item saving. |
Most negative in volume terms. The revenue-per-head model breaks here first. Vendors must reprice to outcomes or lose revenue. This is where today’s market repricing is aimed. |
The mix is the strategy
The pattern is consistent across all four. Agentic AI pushes routine, well-defined work toward whoever can run agents most cheaply. It pushes scarce human judgment toward whoever owns the outcome and the risk. The offshore body-count models feel the squeeze first. The onsite advisory, in-house and captive models hold up better, but only if they move up the value chain instead of defending headcount.
For most large enterprises the answer is not one model. It is a deliberate mix. Keep the differentiating work and the governance in-house. Use a captive for sustained, ownable scale. Bring vendors onsite for advisory, transformation and the accountability layer that makes agentic delivery safe. Send only the genuinely commoditized work offshore, and pay for it as an outcome, not as a seat.
The CIOs who redraw that map first will spend less and ship more. The vendors who see it coming will follow their clients up the diamond. The ones who do not will keep selling seats that agents have already filled.
© Dr. Harish Kotadia, 2026. All Rights Reserved.
Dr. Harish Kotadia, Ph.D., is an Enterprise AI Architect with 20+ years of IT consulting experience serving Fortune 100 clients, specializing in agentic AI systems built on Anthropic Claude, AWS Bedrock, and Google Vertex AI. He holds a Ph.D. in Marketing Management with doctoral research in marketing analytics. Follow him at @agenticaiarch on X and at AgenticAIArch.com.