--- title: "About" description: "Who Sunil Kumar is, what he believes about agentic AI in the enterprise, and how Ailoitte Technologies came to be." url: https://sunilkumar.ai/about/ author: Sunil Kumar (https://sunilkumar.ai/about/) dateModified: 2026-09-02 --- # About Who Sunil Kumar is, what he believes about agentic AI in the enterprise, and how Ailoitte Technologies came to be. I am Sunil Kumar, co-founder and CEO of Ailoitte Technologies, an AI-native engineering and product studio headquartered in Bengaluru with a US entity in Dover, Delaware. I started building the company in 2017. Today it has more than 70 engineers, has shipped over 300 products, and those products serve more than 50 million people in healthcare, fintech, enterprise and government. Forbes listed Ailoitte among its Top Innovative AI Companies in 2025, and the company holds ISO 27001 and ISO 9001 certification. ## How I got here I trained as a computer scientist, a BTech and then an MTech with a specialisation in AI, and I started the way most people in this industry start: as a coder, one function at a time. The work I remember from those years is not the code. It is the number of times I watched a good system fail for reasons that had nothing to do with the code, because nobody had decided who owned it, what it was allowed to do, or what it was worth. That pattern followed me into the years I spent as a solution architect, working with the leadership teams of more than fifteen enterprises on their technology. I saw the same two failure modes everywhere. Large companies had the experience and the data but moved slowly, because every decision had to survive a committee. Startups moved fast but lacked the depth to build things that would still be running in five years. I started Ailoitte in 2017 to put the two together: the technical depth of an enterprise team with the speed of a startup, on fixed prices, so that the client paid for outcomes and not for hours. The name is a small story. "Aloitte" is Finnish for "to start" or "to grow". We added the "I" for AI, because the company was built around AI from the first day, not retrofitted to it later. Eight years and three hundred products later, the failure mode I care about has a new shape. It is the AI agent that looks finished in a demo and is nowhere near finished for production. I have watched it fail the same three ways in company after company, and I have watched what it takes to get one across. That is what this site is about. ## What I work on The demo proves the model can do the task. Production proves the organisation can live with the agent doing it every day, at scale, with an audit trail, inside a budget. Most enterprise agent projects die in the gap between those two proofs. My work is getting them across it. At Ailoitte we do that with what we call AI Velocity Pods: small senior teams on fixed-price, outcome-based engagements. Fixed price forces the discipline that agentic work otherwise avoids. You have to decide up front what "done" means for an agent, and that decision is most of the engineering. The productised version of that discipline is [Leverge AI](https://leverge.ai), Ailoitte's product, which I lead. It has three parts. A library of 306 ready-to-deploy agents across 13 business functions, from sales, finance and procurement to customer service, HR and legal, each with a published runtime workflow and named human approval for any irreversible action, deployed inside the CRM, ERP or ticketing system a company already runs. [Leverge Design](https://leverge.ai/solution-design), which turns a brief and a folder of documents into an engineering-ready package: architecture, data model, backlog, threat model and statement of work, generated from one versioned context so they cannot drift apart. And, in progress, an orchestrator that lets an enterprise assemble and deploy its own agentic workflow from that library in a few clicks, under the same approval and audit model. ## What I believe Eight things, each learned the expensive way. 1. **Demos are cheap.** The demo proves the model can do the task. Production proves the organisation can live with the agent doing it every day. Only the second one is worth money. 2. **An agent is not production-ready until its failures are bounded, observable and owned** by someone with a pager. Not one of the three; all three. 3. **The refusal list comes before the prompt.** What the agent must never do is the first document on any agent project, and it is enforced in code, not in English. 4. **Nothing irreversible without a named person.** Every one of the 306 agents in Leverge has a named approver for irreversible actions. That single rule is why enterprises say yes. 5. **Governance is an engineering problem before it is a policy problem.** ISO 27001 and ISO 9001 are not paperwork to us; they are the shape of the delivery process. 6. **Price the run before the rollout.** An agent without a budget ceiling is not a product; it is an invoice waiting to happen. 7. **Small senior teams beat large ones** on agentic work, because the whole team has to have read every prompt and every failing trace. That is why we work in pods and why we can fix the price. 8. **The lifecycle changes when agents do the work.** I call the version we use AI-DLC. Roles change, reviews change, and the definition of a release changes. Speed and quality were never a contradiction; they were a process problem. ## Timeline - 2017: Started Ailoitte Technologies in Bengaluru. Three projects in the first year, no billing disputes, a habit we kept. - 2018: A team of twenty and the first client in the United States. - 2019: Partnered with Apna as it grew past a million downloads. - 2020: Went remote-first and put AI tooling into every delivery workflow. - 2021: Opened the healthcare practice: ABHA and ABDM-compliant platforms for Indian providers, and the compliance discipline that came with them. - 2022: Clients in six markets: the US, the UK, the UAE, Australia, Singapore and Canada. - 2023: The hundredth product shipped, and the first industry awards. - 2024: Launched AI Velocity Pods, fixed-price delivery for agentic AI, and stopped selling hours. - 2026: Released Leverge AI with the team: 300+ ready-to-deploy agents for every major department, and Leverge Design for enterprise solution design. - 2026: Started writing here under the banner Agentic in Production. ## Elsewhere I run [rightagent.ai](https://rightagent.ai), a content brand and newsletter on the same subject. Ailoitte's own site is at [ailoitte.com](https://ailoitte.com) and the agent platform is at [leverge.ai](https://leverge.ai). For background on the field, the Wikipedia entries on [intelligent agents](https://en.wikipedia.org/wiki/Intelligent_agent) and [large language models](https://en.wikipedia.org/wiki/Large_language_model) are decent starting points.