Much of the public conversation about AI in the enterprise has collapsed into a single question. How many roles can it remove? Boards ask it, analysts model it, and vendors answer it with great confidence. It is the wrong question, and enterprises that optimize for it risk ending up with leaner teams and weaker systems.
History suggests a better one. When ride hailing arrived, it did not carve the existing taxi market into two halves and stop there. In a 2017 study, the San Francisco County Transportation Authority found ride hail vehicles making more than 170,000 trips within the city on a typical weekday, roughly twelve times the number of taxi trips. Our read is that the pool grew because getting across town became easy enough for trips that once depended on a favor from a neighbor or a spouse, or simply never happened.
Airbnb opened up towns, suburbs and smaller cities where travelers once had nowhere suitable to stay, making room for experiences that simply were not available before. In popular metros and tourist hotspots, the added supply tempered the prices hotels could command on peak days and during major events. With stays more affordable across the board, more people could travel, and the spending that followed reached local businesses and jobs well beyond the usual destinations.
Neither platform simply divided its market. Each lowered the barriers to, or the cost of, a service and released demand that had been waiting. AI can do the same inside the enterprise, provided the headroom it creates is pointed at revenue, margin and customer experience rather than written off as savings.
At Lirik, we start from a different premise. AI should replace effort, not expertise. We think of it as effort substitution rather than labor substitution. The distinction may sound semantic. In practice it determines whether an AI program compounds value over time or quietly erodes the knowledge that made the business work in the first place.
Every enterprise platform that has been in production for a few years carries two kinds of knowledge. The first is explicit: configuration, code, integration maps and whatever documentation survived the last three releases. AI tools are now remarkably good at reading all of it.
The second is tacit. It lives with the administrators, architects, analysts and business owners who have run the platform through reorganizations, acquisitions, regulatory change and shifting leadership priorities. Tacit knowledge explains how people decide, why a validation rule exists that nobody would design today, why one region’s quoting process diverges from the rest, and which integration needs watching at quarter close. It is the institutional memory of how the business actually operates, as opposed to how the system was originally specified. None of it is in the repository. All of it matters.
An AI agent can tell you what a system does. Only the people who built and ran it can tell you why. Remove those people and the AI does not become smarter. It becomes confidently wrong at scale.
Watch how a capable platform team spends its week. Many of its hours go to work that consumes attention without requiring judgment: triaging tickets, reproducing defects, writing test scripts, documenting changes, reconciling releases and chasing environment drift. This is effort. It is necessary, repetitive and increasingly automatable.
The judgment calls take up far less of the calendar, and they are where the value sits. Someone has to decide what to build and why, what to retire, when and how, and which exception the business genuinely needs versus which one it simply inherited. This is expertise. It is scarce, it is slow to rebuild, and it does not transfer through a handoff document. It is also part of your company culture, your decision framework.
The labor substitution model treats effort and expertise as one thing and cuts both. Our model separates them. We put AI to work on the effort so that the expertise can be redeployed to the decisions only it can make.
Our engagement model puts this principle into practice. We do not open an engagement by proposing to rewrite anything. We begin by taking on the work the client already has, and we sequence the program in three phases. Each phase earns the right to the next.
Run. We assume responsibility for managing the existing platform, using AI native tooling for monitoring, triage, testing, documentation and routine change. In this phase we lean heavily on the client team. Their people show us where the fragile parts are, what the business cares about in a given season, and which shortcuts carry history. We engineer that knowledge into the runbooks, test suites and agent instructions that power the service. The client gets time back almost immediately.
Resolve. Once we understand the systems well enough to change them safely, we turn automation towards technical debt and backlog that have accumulated for years. Work that was always important but never urgent finally gets done, because the cost of doing it has fallen. The client team is no longer tied to the mundane, repeatable work, and is free to direct what gets fixed first.
Refactor. With the platform stable, documented and understood, modernization becomes a business decision rather than a leap of faith. The knowledge that makes refactoring safe has already been captured, tested and proven in production.
The common way to measure AI in delivery is the share of existing tasks it can automate. That framing undersells the opportunity. Efficiency on yesterday’s work list is where the return begins, not where it ends.
Every enterprise has a backlog. Behind it sits a second one that never was written down. It holds the analysis a sales leader stopped requesting because the last request took two quarters, the integration a service team works around by hand, and the experiment nobody proposed because the queue was already full. Call it the invisible backlog. It is demand that the cost of delivery has quietly suppressed.
When AI lowers that cost, the invisible backlog surfaces, and that is where the real value lives. The goal is not to do the same work with fewer people. It is to do the work the business has wanted for years and could never fund, with the people who understand it best directing the effort.
This model asks something of our clients early on. We need time with their experts, candor about where their systems are fragile, and patience while we learn. We are explicit about that ask because everything that follows is built on it.
What comes back is worth far more. Experts stop spending their days on work that does not need them. Their knowledge is engineered into systems that keep working when people change roles. Their teams gain the room to take on the initiatives that move the business forward, including new capabilities, new products or services with new ways of serving customers.
None of this is a cautious position. We are building AI native delivery as aggressively as anyone in our market. The difference lies in what we believe the technology is for.
The enterprises that lead in the years ahead will be the ones that converted the most effort into capacity and pointed that capacity at the work that matters. Their people will remain at the center, doing more of what only they can do, while machines carry the rest.
That is the future we are building with our clients. The effort gets replaced. The expertise gets multiplied.
IndiQube Fort, 3rd Floor Malviya Nagar, Jaipur India, 302017
AWFIS – 26th Floor, GIFT One Tower, Gujarat International Finance Tec-City, Gandhinagar, Gujarat – 382355