---
title: "Speaking"
description: "Talks, panels and podcasts by Sunil Kumar on agentic AI in production. Bio and headshot kit for organisers."
url: https://sunilkumar.ai/speaking/
author: Sunil Kumar (https://sunilkumar.ai/about/)
---

# Speaking

Talks, panels and podcasts by Sunil Kumar on agentic AI in production. Bio and headshot kit for organisers.

I speak to leadership teams, engineering audiences and industry events about one thing: what it takes to run AI agents in production in a real enterprise. The talks are built from Ailoitte's own deployments and from the 306 agents running in Leverge, with the client names removed and the mistakes left in.

Formats: keynotes of 20 to 40 minutes, panels, podcasts, and closed-door sessions with leadership teams. I am based in Bengaluru and travel for the right event. Every session below can be cut to the slot you have.

## Sessions I give

### Agentic in Production

_Keynote, 30 to 40 minutes. CIO forums, leadership offsites, industry conferences._

Why most enterprise AI agents never leave the demo, the three gaps they die in (accountability, cost and audit), and the four tests an agent has to pass before it touches real work. The audience leaves with a one-page definition of "production-grade" they can hold their own programme to.

### The refusal list

_Talk or workshop, 25 to 45 minutes. Engineering, risk and compliance audiences._

The one document every agent programme is missing: what the agent must never do, enforced in code rather than in prompts. How writing it first changes the architecture, the evaluation set and the price, with the support agent that could not say no as the worked example.

### Fixed price for agentic AI

_Talk, 20 to 30 minutes. CTOs, procurement and delivery leaders._

Why hourly billing fails for agents, the three artefacts that make an agent priceable, and how a small senior pod can carry the delivery risk. Includes the questions to ask any vendor, including us.

### Ten thousand runs a day

_Talk, 20 minutes. Finance and technology leaders in the same room._

What an agent costs at scale, why the pilot's cost per run is the wrong number, and the budget ceiling that turns a surprise invoice into a decision. Routing, caching and limits, and why the limit matters most.

### AI-DLC

_Talk or leadership session, 30 to 60 minutes. Engineering leadership._

How the software lifecycle changes when agents write, test and operate code: what moves to agents, what stays with people and gets more important, and what actually happened to team shape at Ailoitte.

### A half-day on your own programme

_Closed-door session with a leadership team._

Not a talk. We take your agent programme as it stands, work out where it sits between demo and production, name what will break, and leave with the three decisions to make this quarter. Under NDA, no slides.