Services-as-Software: Rearchitecting Global Delivery in the Age of AI
The true story here isn’t a single company closing offices – it’s the architecture of work being rewritten.
Why this matters now
I recently read the Opendoor case: a U.S. home‑buying platform that scaled operations in India and then decided to close those centers, citing a move toward smaller “AI‑native” teams and bringing operational work closer to customer markets. That decision has become shorthand for a deeper shift: AI is not merely automating tasks, it’s changing how businesses structure operational capacity and where value is created.
Rethinking enterprise design: from headcount to outcome
For two decades the playbook for many global firms was straightforward: shift repeatable, labour‑intensive processes to low‑cost locations and optimise for utilisation. AI changes the variables in that equation.
- Automation reduces marginal labour needs. The real leverage now comes from integrating generative models into end‑to‑end workflows so fewer humans can supervise larger volumes of work. That compresses the role of large back‑office teams.
- The unit economics change: it’s no longer staff cost per transaction but the cost of reliable AI inference, data pipelines, and governance. This incentivises consolidating critical operations closer to product owners – not out of nationalism, but to shorten feedback loops and manage model risk.
- Architecture must shift from “people as scale” to “software + humans as composable services.” That means productising operational flows (Services‑as‑Software), with clear APIs, SLAs, and telemetry.
Practical architectural implications for CTOs and architects
If you run or advise engineering organisations, here are the concrete consequences I expect and what to act on now:
- Design for human‑in‑the‑loop: Treat humans as exception‑handlers, not batch processors. Build orchestration layers that route ambiguous cases to specialists, with interfaces optimised for fast context switching and error correction.
- Invest in MLOps and data contracts: The bottleneck becomes clean, auditable data and stable model deployment. Versioned data contracts, end‑to‑end lineage, and reproducible training pipelines are non‑negotiable.
- Move from monoliths to composable services: Productise operational capabilities (document ingestion, valuation rules, exception resolution) as services with measurable outcomes. That reduces dependency on large regional teams and makes automation portable.
- Adopt robust governance and observability: When AI drives decisions, you must measure bias, drift, latency, and user‑impact in production. Observability for models is as critical as for infrastructure.
- Rebalance cost vs. resiliency: Firms tempted to centralise everything risk single‑point failures. Hybrid placement (edge + cloud), geo‑redundancy, and disaster recovery must be part of any AI‑native roadmap.
- Prioritise retraining and role evolution: Technical staff must move up the value chain – from execution to model supervision, feature engineering, and product design.
What this means for India (and opportunity, not only risk)
India’s Global Capability Centre (GCC) ecosystem will feel pressure on labour‑arbitrage models. But the larger opportunity is a pivot: GCCs and Indian engineering teams can become centres of excellence for AI productisation, MLOps, and domain‑specific model tuning. I’ve seen similar transitions working with startups in Assam and guiding curriculum at local universities – where the real differentiator is not low cost, but domain expertise plus engineering discipline.
Takeaways for leaders
- Measure outcomes, not headcount: Build KPIs around cycle time, error rate, and customer outcomes.
- Treat models as first‑class production systems: Invest in MLOps, lineage, and auditability now.
- Productise operational work: Convert repeatable flows to APIs and SLAs before trying to replace people with models.
- Upskill strategically: Move talent from rote processing to model supervision, prompt engineering, and product ownership.
- Plan for governance and resilience: Model risk is business risk – make it visible and manageable.
Closing thought
We are not witnessing the “end of outsourcing”; we are watching the rebirth of engineering organisations that think in services, models, and outcomes rather than desks and hours. The companies and GCCs that win will be those that design their systems – technical, organisational and human – around that reality.
About the Author: Sanjeev Sarma is the Founder Director and Chief Software Architect at Webx Technologies. With a core focus on Generative AI integration, Cloud-Native Scalability, and Enterprise Software Architecture, he has spent over two decades driving digital transformation across Northeast India and beyond. Beyond his corporate leadership, Sanjeev is deeply invested in shaping the future of the IT industry. He serves as an Industry Expert on the Board of Studies for Assam Don Bosco University’s School of Technology, advises state technology committees, and actively mentors emerging tech startups at STPI. He brings a unique, dual perspective of high-level enterprise execution and future-ready academic curriculum development.
