For a long time, growth in our industry followed a fairly predictable pattern. If demand increased, we hired. If delivery expanded, we added capacity. It wasn’t something we questioned too deeply. It was simply how the system worked.
AI disrupted that assumption in a way that felt less like gradual change and more like a quiet rupture. We began to notice that tasks which once required time, coordination, and effort were now being completed faster, often with fewer dependencies. At first, it seemed like an efficiency gain. Over time, it became harder to ignore that it was something more structural.
The change did not fully register until we saw its impact on our own business when revenue dropped by 22%. As parts of execution became easier to automate, the relationship between effort, output, and value began to change. Some of the work we had built our delivery model around was no longer scarce in the same way. And that was not a comfortable realisation.
It forced us to pause and rethink something more fundamental about how work itself should be designed in this new context. If execution is no longer the constraint, then what is?
That question did not lead to an immediate answer. But it did change the direction of our thinking, and eventually, how we chose to respond when AI went from being an interesting capability to becoming a business model problem.
Why We Treated AI as a Work Design Challenge, Not a Tool
The obvious response at that point would have been to adopt AI tools aggressively. Many organisations did exactly that, and for good reason. The capabilities were visible, the gains seemed immediate, and the pressure to act was real.
While we acknowledged AI’s potential, we decided to take a pause because we were not convinced we understood the problem correctly. AI adoption, as it turns out, is rarely limited by technology. It is limited by how work is structured around it. This aligns with broader industry findings. Research from Deloitte shows that while AI adoption is rising, only 6% of organisations have made meaningful progress in designing effective human+AI interactions. Adding a powerful tool into a workflow that was designed for a different reality tends to produce uneven results.
So we created space to experiment. In 2023, we set up an internal AI Lab. Not as a centre of excellence in the traditional sense, but as a place to test AI within real workflows, under real constraints. Some experiments worked. Many did not. That was useful in itself.
Over time, patterns began to emerge. AI performed well in execution-heavy, repeatable tasks. It was less reliable when context, judgment, or accountability were required. That distinction is easy to state, but harder to apply consistently when delivery pressures are high and timelines are tight.
At some point during this phase, our framing changed. We stopped thinking of AI as a tool that people use. We began to see it as a participant in the work itself. Not quite a colleague, and certainly not a replacement, but something that needed to be assigned responsibility with some care.
Designing the Human + AI Workforce Model
Redefining roles between humans and AI
Once we began to treat AI as part of the system of work, the next step was to define how that system should function realistically in day-to-day execution.
We arrived at a model that was simple in structure, but not trivial in application. AI would focus on execution, scale, and speed. Humans would define intent, design workflows, review outputs, and remain accountable for outcomes. The work did not disappear, but was redistributed.
This change is not unique to us. A study by MIT Sloan Management Review and BCG found that 66% of AI-agentic leaders expected a redefinition of roles and responsibilities, and 45% expected a reduction in middle management layers.
For us, this redistribution had consequences for how roles were defined. Employees were no longer expected to execute every task themselves. Instead, they were expected to orchestrate systems of work. This included deciding what should be done, how it should be done, and what needed closer attention. In practice, this often meant managing multiple streams of work at once, each moving at a different pace.
From task execution to system orchestration
We also introduced a “first right of refusal” model for AI. If a task was repeatable and execution-driven, AI would attempt it first. Humans would step in where judgment, refinement, or context became important. It sounds like a neat rule, and occasionally it even works like one. Most of the time, it simply provides a useful starting point.
The idea of an “AI employee” helped make this more tangible. These systems take on entry-level execution tasks such as research, first drafts, data preparation, and coordination. They do not get tired, and they do not get bored, which makes them particularly suited to certain types of work.
This is often where the conversation turns to job loss. In our experience, that framing is incomplete. What we saw instead was a shift from doing tasks to designing and managing them. The work did not reduce, but the nature of the effort changed.
From Experimentation to Organisation-Wide Adoption
Creating the foundation through experimentation and enablement
In 2023, the focus remained on learning. The AI Lab allowed us to test use cases across actual delivery scenarios, without the pressure to immediately standardise or optimise. This phase helped us understand not just what worked, but where things broke down. While experimentation created clarity, scaling required intent.
In 2024, we moved towards distributed ownership. We introduced AI Champions across departments, with the idea that AI adoption should not depend on a central team. Each function needed to interpret its relevance and apply it within its own workflows. This reduced dependency and, perhaps more importantly, increased ownership.
Alongside this, we introduced organisation-wide AI training and certification. This was mandatory for all employees, including non-technical teams such as HR, Finance, and Operations. The goal was not to create specialists, but to ensure a shared baseline. Without that, adoption tends to remain uneven and, at times, unnecessarily intimidating, especially for non-technical staff.
Driving behavioural shift and scaling adoption
By 2025, the nature of adoption began to change. What started as a leadership-led push gradually became an employee-driven pull. Teams began identifying opportunities for automation on their own. Some of the most practical ideas came from areas that were initially cautious. It turns out that scepticism, when addressed properly, often produces better solutions.
This transition from top-down push to bottom-up pull is often where organisations get stuck. While 85% of leaders say adaptability is critical, only 7% believe they are good at it. The gap is not in intent, but in the ability to translate that intent into operational change.
We also began experimenting with connected workflows, or what is now described as agentic orchestration. Instead of isolated tasks, we started linking processes so that outputs from one stage fed into the next. This allowed multiple activities to run in parallel, with human oversight at key points.
In 2026, we are in the process of redesigning all SOPs through a Human+AI lens. This involves defining clear intervention points, accountability structures, and guardrails. It is detailed work, occasionally repetitive, and not particularly glamorous. It is also where most of the actual transformation happens.
What Changed: Scale, Roles, and the Way Work Happens
The most visible change has been in how work moves.
Today, we run end-to-end AI-first workflows that span research, ideation, content production, and optimisation. AI supports each stage, while humans intervene where decisions are required or where outputs need refinement. The sequence remains familiar, but the pace and structure are different.
This has allowed us to scale significantly. For one of our clients, we delivered over 70,000 content assets using AI-enabled workflows. That level of throughput would have been difficult to sustain using earlier models, at least without introducing other trade-offs.
Work also happens more in parallel than before. AI agents can handle multiple tasks simultaneously, which changes how time is used. Employees are not waiting for one stage to complete before moving to the next. Instead, they are coordinating several streams of work at once. It is efficient, although it occasionally requires a reminder that just because everything can run at once does not mean it always should.
Roles have evolved accordingly. Execution remains important, but it is no longer the centre of gravity. Employees spend more time on briefing, reviewing, and making decisions. They are also expected to troubleshoot when outputs are inconsistent, which requires a different kind of attention.
There is a cultural dimension to this shift as well. As AI becomes part of everyday work, hesitation tends to reduce. Confidence builds through use rather than instruction. That said, discipline becomes more important. The ability to question outputs, rather than accept them quickly, turns out to be a critical skill.
What Made This Work and Where Most Organizations Struggle
Looking back, a few factors made this transition more effective.
-
Inclusion was one of them. AI adoption was not limited to specific teams. Every function participated, which helped distribute capability and reduce friction. When everyone has a baseline understanding, collaboration becomes easier.
-
We also placed equal emphasis on guardrails. Employees are encouraged to question AI outputs and not accept them at face value. Research suggests that many users tend to over-trust AI unless trained otherwise. Left unchecked, that tendency can lead to average or incorrect outcomes at scale.
-
Another important factor was treating AI as a cost, not free capacity. Each AI-based activity carries a cost, and we are building dashboards to track usage and outcomes. It changes behaviour in subtle ways. Re-running a task repeatedly starts to feel less like exploration and more like something that should be fixed at the source.
-
We also linked AI adoption to business outcomes. Productivity gains were measured and reinvested, creating a self-sustaining model. This helped ensure that adoption remained grounded in value rather than novelty.
-
Finally, culture played a role that is difficult to quantify but easy to observe. Psychological safety, openness to experimentation, and a willingness to question decisions all become more important as the pace of change increases.
There is also a broader shift underway in how organisations think about structure. As Joanne Chen, General Partner at Foundation Capital, pointed out in her article, future organisations are likely to rely less on rigid hierarchies and more on smaller, high-leverage teams coordinating systems of work. That is not a distant idea. It is already visible in how work is evolving.
The Future: Smaller Teams, Higher Leverage
The future workforce will not be defined by how much AI is used, but by how effectively work is structured around it.
I suspect we'll see smaller teams in many areas, but not because there is less to do. The volume of work is only increasing. The difference is that a lot of execution can now be handled by AI systems working together, allowing people to focus more on judgment, context, creativity, and ownership.
The line between what humans do and what AI does will keep moving, and organisations will need to adjust along the way.
The lesson we have learned is that beyond adopting tools, AI is about redesigning work. Organisations that recognise this tend to move with more clarity. Those that do not often end up with impressive tools and limited impact.
And, at least for now, AI remains very good at execution. It is still not particularly good at deciding what is worth doing in the first place. That responsibility is still ours.
References
About the Author
Ganapathy Sankarabaaham is the founder and CEO of Vajra Global and XITE Create. Vajra Global leverages its strength of combining technology and marketing to provide MarTech solutions, while XITE Create provides Generative AI services. An engineer with an MBA, Ganapathy holds a ‘Leadership with AI’ certification from ISB. Before founding both companies, he spent many years honing his craft at Tata Consultancy Services. He is a highly sought-after speaker who shares his expertise globally. He has spoken at several events, including Ad Hoc Council, Nasscom, TiE, CII, HubSpot, and many others.