UK-based global technology consultancy firm Thoughtworks has seen a rise in its customer base seeking advice on how to implement AI efficiently across businesses, without unintended consequences.
The firm, which helps its customers with enterprise modernisation and has a strong DNA in writing custom software for clients, has seen a shift in the way customers approach AI adoption as a problem statement.
The company employs over 3,800 people in India and has an ‘AI Labs’ in the country to test AI adoption for development work. Globally, Thoughtworks employs nearly 10,500 people and has a presence across 19 countries.
Since taking over as CEO and director of Thoughtworks in 2024, Mike Sutcliff said in an interview with Forbes India that his tenure has been focused on “managing the inflection point on how software is written and tested,” with AI.
He further added that as companies roll out new AI products, building trustworthy solutions is the first hurdle and a necessity for adoption.
“A lot of people building pilots and prototypes are trying to prove whether or not it works. How we help them is by telling them not only what it can do, but what it should do and how it should do it, so that when you put it out in the market, people have the confidence to use it,” said Sutcliff.
He added, “We are big believers in tech for good, and we are making sure that clients are not building early versions of different AI-enabled solutions that will not pass the early hurdle on trust.”
Edited excerpts:Q. How has Thoughtworks' India operations grown since 2001?
A (Sutcliff): We operate across different countries and have always worked at a country level, where we engage with clients locally while supporting people on a global scale. India is part of the global delivery network, and we have expanded across seven cities in India since 2001.
For our portfolio in India, we service multiple clients from the US, Europe, Latin America, and Asia-Pacific. We also have a thriving local market business, including work with major banks. We have collaborated with the government on Digital Public Infrastructure initiatives as well.
India is also the hub for Thoughtworks University, where all graduates joining the company fly to Pune from across the world to spend a month. It gives them an idea of the thriving IT community here.
Q. Since the change in leadership, how has Thoughtworks redrawn its strategy?
A: Thoughtworks has a long history of being one of the best custom software development teams in the world, and our focus has been on how to write good custom software. We also coined the term data mesh a few years ago, applying the principles of modern software engineering to data engineering, thereby creating an analytical data architecture and operating model.
Since I joined the company, we have re-focused to ask two key questions. The first is how to manage what is probably the largest inflection point in how we write software. For this, we have leaned heavily into testing new tools, models, and techniques, and teaching our teams how to use them with AI for software development.We recognise that this is a rapidly changing space. With our AI Labs team in India, we test different tools and techniques to solve challenges faced by others in implementing or adopting AI in software development.
The second big question is how we can help our clients truly understand what is happening in the world of AI models—what they should and should not do in pre-training, post-training, pruning, and optimising inference engines, and how they can use new tools to implement AI effectively.
A widely quoted MIT study says that 95 percent of generative AI projects in business are failing to show returns on investment. This is primarily due to change management and decision-making around how to apply AI, not because AI models are failing.
Also Read: We’re not chasing the AI hype: Zoho CEO Mani Vembu
Q. What are the different functions or fields where you are helping customers adopt AI and derive maximum value?
A: My observation is that what most of our clients are trying to do with AI falls into three categories. The first is using it to increase revenue—through better digital marketing, adding AI product features, and optimising customer experience.
The second group of clients use AI to improve business processes and operating practices for example, optimising supply chains or improving financial forecasting. And the third area is engineering—building the next generation of their technology.
Now in the third category, most clients across industries are frustrated that they spend around 80 to 90 percent of their IT budgets maintaining or modernising legacy technology. We see a huge opportunity to support these companies through our reverse engineering techniques—understanding existing systems and then forward-engineering them into modular, microservice-based, high-performance systems.
We work with many clients looking to retire mainframes—often dealing with systems containing 20, 30, or 40 million lines of complex code—and replacing them with modern, efficient alternatives.
While modernisation remains a big area within engineering, there’s also growing interest in how to use AI to create new experiences, interfaces, and features. We work across the board on these challenges.
Q. Talking about adoption, what stage is agentic architecture at in the enterprise adoption cycle?
A: I would say agentic architecture is still at a very early stage. Probably only 5 to 10 percent of our clients are experimenting with it, and nobody has yet put it into production.
That’s because the standards and protocols are not mature. The first thing we have to do as technology evolves is to understand how it works, where it works, and what the boundary conditions are—what is safe and what isn’t. So, agentic AI is still in its early phase; we are learning how it works before the industry can fully agree on best practices.
Q. How do you ensure customers are aware of the cybersecurity landscape when deploying new AI models?
A: Every one of our developers is trained in and aware of the DevSecOps framework. When we introduce new models, we also introduce new attack vectors. For instance, prompts can be written that cause models to do things that were not intended.
If you are deploying technology using AI as part of a product feature, you must understand that you’ve just introduced a probabilistic system into your infrastructure—one that can now be attacked using a different set of techniques.
We spend a lot of time talking to clients about putting guardrails around technology and ensuring it is deployed responsibly as they develop new product features. One of the biggest reasons companies are hesitant to push large-scale production is cybersecurity concerns, alongside data security in the context of sovereign AI.
First Published: Nov 10, 2025, 15:58
Subscribe Now