For years, the Artificial Intelligence (AI) conversation has revolved around increasingly powerful models. Companies have competed to build systems that can write, reason, analyse, code and generate content at unprecedented speed. Yet as AI moves from laboratories and demonstrations into ordinary businesses, a more difficult question is emerging: how effectively can organisations actually put these technologies to work?
The answer is exposing a significant gap between AI capability and business value. Enterprises may have access to sophisticated models and may run dozens of pilots, but experimentation does not automatically translate into measurable productivity, revenue or cost savings. TCS's decision to lay off 20,000 employees, followed by plans to hire 8,900 engineers for one specific role — Forward Deployed Engineers, not software engineers or data scientists — is among the clearest signals of this shift. The next phase of the AI economy will, therefore, be defined less by who builds the most impressive model and more by who can successfully deploy AI within the messy realities of an organisation.
The scale of investment behind this shift is telling. In the space of a few weeks, Microsoft committed $2.5 billion and roughly 6,000 experts to embed engineers inside enterprise clients, AWS followed with $1 billion for its own Forward Deployed Engineering effort, and OpenAI and Anthropic together committed more than $5 billion to deployment ventures — close to $9 billion aimed not at building better models, but at getting models already built to actually work inside real businesses. The data explains why: MIT's Project NANDA, which tracked enterprise AI deployments across hundreds of organisations in 2025, found that 95% of enterprise generative AI pilots showed zero measurable impact on profit and loss.
Real businesses rarely operate in clean technological environments. They have legacy systems, fragmented data, regulatory requirements, established workflows, security constraints and institutional knowledge accumulated over decades. Employees may understand their processes intuitively but struggle to translate them into instructions that an AI system can use effectively. A model that performs brilliantly in a controlled demonstration can, therefore, encounter entirely different challenges when introduced into a live enterprise.
{{/usCountry}}Real businesses rarely operate in clean technological environments. They have legacy systems, fragmented data, regulatory requirements, established workflows, security constraints and institutional knowledge accumulated over decades. Employees may understand their processes intuitively but struggle to translate them into instructions that an AI system can use effectively. A model that performs brilliantly in a controlled demonstration can, therefore, encounter entirely different challenges when introduced into a live enterprise.
{{/usCountry}}This is creating demand for a new kind of technology professional: The Forward Deployed Engineer. Unlike conventional software engineers who primarily build products, these professionals operate close to customers and business teams, translating real-world problems into deployable technological solutions. Their value lies not simply in writing code, but in understanding what the organisation actually needs and taking responsibility for making the technology work.
The distinction is important because technical capability alone is no longer sufficient. As AI tools increasingly automate routine coding, integration and other lower-level tasks, engineers will need to operate at a higher level of abstraction. Customer discovery, communication, problem-solving, pattern recognition and business judgement are becoming increasingly important. The strongest practitioners will be those who can enter an unfamiliar organisation, understand its constraints and identify where AI can create genuine value rather than simply adding another layer of technology.
This also changes the way organisations should think about AI talent. Hiring large numbers of engineers does not automatically create an effective deployment capability. There is a difference between building an FDE team and building an FDE system. A team can simply become another cost centre, while a system can accumulate knowledge and become more efficient with every deployment — recognising in days what once took months, because it has already encountered similar problems across industries, data environments and organisational cultures.
That distinction is easy to claim and harder to build, which is creating its own problem. Roughly 30% of Forward Deployed Engineer job postings map closer to sales engineering than to genuine builder roles, and many companies hiring for the title are, in effect, hiring expensive consultants with a better job title. The real test of an FDE system is not headcount or client logos secured, but whether each engagement makes the next one faster, cheaper and more effective than the last.
This is particularly important because the biggest obstacle to enterprise AI adoption may not be a lack of willingness. Businesses are increasingly interested in AI, but many struggle to distinguish genuine opportunities from technological noise. Successful implementation often requires an internal champion who can connect AI capabilities with a specific organisational problem, build confidence among stakeholders and ensure that promising experiments progress beyond the pilot stage.
The changing nature of AI deployment also has significant implications for education. Traditional engineering education has understandably focused on programming, software development, computer science and technical fundamentals. These remain essential, but the emerging AI economy requires an additional layer of capability. Future engineers will need to understand how businesses operate, how decisions are made, how regulatory constraints influence technology and how users interact with systems.
India has a particularly significant opportunity in this transition. Its large engineering workforce, established IT services industry, growing global capability centres and experience serving international businesses provide a strong foundation for becoming a major AI deployment hub. Yet India's GCCs recorded an AI and data analytics supply-demand gap of 36 to 40 per cent in Q1 FY27 — for every ten open GenAI roles, there is roughly one qualified engineer available. Technical supply alone will not be enough. The country needs professionals who can combine engineering competence with business understanding and real-world implementation experience.
This requires a shift from model-centric to deployment-centric learning. Students should not only learn how AI systems are built, but also how they are integrated into organisations, evaluated against business outcomes and adapted to different operational environments. Industry exposure, interdisciplinary learning and practical deployment projects could become as important as traditional technical coursework. The distinction matters because the next generation of AI engineers will increasingly be judged not by what they can build in isolation, but by what they can make work in the real world.
The broader lesson is that AI will change the definition of valuable technical talent. As machines become better at producing code and handling repetitive integration tasks, distinctly human capabilities such as judgement, communication, contextual understanding and accountability will become more important, not less.
The AI industry has spent years proving what models can do. The next challenge is proving what businesses can achieve with them. The organisations that succeed will be those that build the systems, skills and institutional knowledge needed to turn technological potential into repeatable outcomes. The model was never the hardest part — neither is the integration. What compounds, and what will separate the winners of this next phase, is judgement.
(The views expressed are personal)
This article is authored by Divyashish Jindal, managing lead, School of AI, Masters’ Union.