Publicis Sapient AI Cuts Modernisation From Years to Weeks

The global market for legacy system modernisation is surging as enterprises race to shed technical debt. Yet, for global financial services, scaling past deeply embedded, multi-million-line common business-oriented language (COBOL) architectures presents massive systemic constraints. 

For technical leaders, the challenge has shifted from incremental code patching to orchestrating end-to-end, AI-augmented engineering outcomes.

Publicis Sapient is a 20,000-strong digital business transformation company helping clients build AI-native enterprises.

Here, Pinak Kiran Vedalankar, Group Vice President of Technology at Publicis Sapient, discusses how his teams leverage agentic networks and platform innovation to eliminate legacy gridlock, accelerate value streams and compress multi-year transformations into weeks.

Publicis Sapient works across many industries. Can you share a recent success story?

One that stands out is a large-scale legacy modernisation we delivered for a global financial services firm using Sapient Slingshot, our AI platform.

The client had a complex COBOL-based system with over 250,000 lines of code and deeply interdependent architecture. 

In 12 weeks, we used AI to extract the underlying business logic, generate specifications at over 95% accuracy and design a modern target architecture. 

Work that would traditionally take years was completed in weeks, with lower cost, greater transparency and a clear, testable path forward for the client.

We’re seeing similar results across retail, healthcare and beyond. AI-led modernisation is consistently delivering 50 to 70% faster transformation while significantly reducing risk.

How is Sapient Slingshot transforming how teams handle tech debt and legacy gridlock?

The traditional approach to tech debt is incremental. Teams patch, refactor or rewrite in silos, and more often than not, they carry the same problems forward. Slingshot flips that model entirely.

It treats legacy systems as the source of truth, extracting business rules directly from code and converting them into structured specifications, test cases and modern architectures. 

For one client we’re currently working with, this is genuinely business critical. We’re dealing with around 20 million lines of tightly coupled legacy code spanning multiple domains.

Rather than a linear, multi-year transformation, Slingshot enables parallel modernisation across domains, reduces SME dependency by 70 to 80% by shifting experts from explaining systems to validating them and provides continuous parity testing so nothing breaks along the way.

The goal isn’t just reducing debt. It’s eliminating it in one pass, where legacy logic is properly understood, validated and rebuilt cleanly rather than carried forward indefinitely.

How are engineering skill sets evolving? What’s most critical today?

Engineering is shifting from writing code to orchestrating outcomes with AI. That’s the simplest way I can put it.

The skills that matter most right now are context engineering, which means structuring domain knowledge, data and business rules so AI systems can actually reason effectively with them. 

Alongside that, AI supervision and quality control are becoming core competencies. 

Engineers are increasingly acting as validators of AI-generated outputs, not just creators of code. And systems thinking across the full software delivery lifecycle matters more than ever, moving beyond isolated coding toward managing end-to-end automation.

At Publicis Sapient, we see engineers evolving into AI-augmented problem solvers, working in lean pods with agents embedded across design, build, test and operations. 

The job is no longer just “write the code”. It is now to define the intent, guide the system and ensure the outcome.

What are the challenges in scaling agentic networks at the enterprise level?

Scaling agentic AI isn’t primarily a technology problem. It’s a systems and governance problem, and that distinction matters.

From our work, four challenges consistently come up.

The first is context fragmentation. Agents are only as good as the context they operate in. Without a unified knowledge layer, outputs become inconsistent fast. 

That’s why we anchor everything on an enterprise context graph, a single source of truth for business logic, data and workflows.

The second is orchestration complexity. At scale, you’re not managing one agent, you’re managing hundreds across the delivery lifecycle. That requires structured orchestration to ensure the right agent is acting at the right stage.

Third is trust and validation. Enterprises cannot tolerate hallucinations or logic drift. Continuous validation against legacy behaviour and automated testing become non-negotiable at that level.

The fourth is the operating model itself. Traditional team structures don’t scale with agentic systems. You need AI-native delivery models, with parallel pods, human-in-the-loop governance and outcome-based execution built in from the start.

When it’s done well, the payoff is significant. Modernisation timelines that once stretched to four or more years can compress to under two, with better quality and lower risk throughout.