Engineering Leadership · AI Platforms

Scale is not an
accident. It is a
design decision.

Nineteen years building the platforms behind India's largest consumer businesses. Today: end-to-end engineering for a digital estate serving 130M+ monthly users — and the AI systems rewriting how all of it gets built.

See the work Follow on LinkedIn
Himanshu Gulati — VP & Head of Engineering
Scroll
0
Monthly users
0
Events / day
0
Engineers led
0
Availability
0
Years
The approach

AI-native, not AI-adjacent.

Most organisations bolt AI onto the edges of how they already work. The harder and more valuable move is to let it change the shape of the engineering system itself — how software is written, tested, shipped and operated. Four convictions I build on.

01

Agents belong in production, not in pilots

A demo that impresses a steering committee proves almost nothing. An agent that resolves a real customer's request, on a real account, inside a live app — with evaluation and guardrails as first-class infrastructure rather than an afterthought — is a different engineering problem entirely. That gap is where most AI programmes quietly stall.

02

The org is the architecture

Systems inherit the shape of the teams that build them. If you want platforms rather than silos, you change reporting lines, career tracks and incentives — not just the codebase. Converting specialists into full-stack engineers, and testers into developers, changes what a team can build far more than any framework choice.

03

Open source is a strategy, not a preference

Choosing open foundations at platform scale is a decision about leverage: who controls the roadmap, where the cost curve bends, and how fast you can respond when the ground shifts. Migrating off proprietary infrastructure is rarely about licence fees — it is about never again being one vendor's pricing decision away from a re-architecture.

04

Reliability is what buys you speed

Teams that fear their own deploys move slowly, and no amount of AI tooling fixes that. Service objectives, real observability, honest incident review and an active-active posture are not the brakes on velocity — they are the reason a large organisation can ship continuously to a hundred million people without holding its breath.

Selected work

Platforms built at consumer scale.

Three systems that changed how a 130-million-user business builds and operates software. Described by problem and approach — the specifics stay where they belong.

01

An agentic assistant that answers for itself

The problem

Customer intent at this scale is enormously varied, and scripted flows only ever cover the head of the distribution. Everything else falls through to a human — expensive, slow, and a poor experience on both sides of the call.

The approach

An agentic platform on a fully open-source stack, with evaluation pipelines and guardrails built in from the first commit rather than retrofitted. Agents that don't just answer but complete the request — resolving the query and executing the service action behind it.

Where it stands

Live in beta inside the consumer app, handling real customer interactions end to end. Built to be extended by product teams rather than owned as one monolithic assistant.

Agentic AIOpen-source LLMsEvaluation & guardrailsService orchestration
02

Making a large engineering organisation measurably faster

The problem

AI coding tools arrived faster than the ability to tell whether they were helping. More code was being produced; whether it was better code, reviewed properly and shipped sooner, nobody could actually say. Existing products showed dashboards, not answers.

The approach

A productivity platform correlating delivery, quality and review signals into prescriptive insight and automated action — telling a team what to change, not merely what happened. Paired with touchless test automation and design-to-code generation as an early enterprise design partner.

Where it stands

Adopted across engineering teams well beyond the one that built it, with AI tooling in active daily use by the majority of engineers and measurable movement in cycle time and release frequency.

Developer productivityAI-assisted SDLCTest automationDesign-to-code
03

One platform, three surfaces, hours not days

The problem

Every product team was rebuilding the same scaffolding three times over — once for Android, once for iOS, once for Web — instrumenting, configuring and experimenting differently each time. Go-live took days, and no two surfaces behaved quite alike.

The approach

A channel platform treating multi-platform deployment as the primary abstraction, with observability, configurability and experimentation available out of the box rather than assembled per team. Built to be reused by other businesses, not just one app.

Where it stands

Go-live cycles compressed from days to hours for every team on it, and the platform now serves surfaces beyond the one it was originally built for.

Platform engineeringMulti-platformExperimentationObservability
The path

Nineteen years, one consistent bet.

That the durable advantage is never a single product — it is the platform underneath it, and the organisation capable of building one.

2022 — PRESENT
VP & Head of Engineering, Consumer Business
India's largest telecom operator

End-to-end engineering ownership for a consumer digital estate serving 130M+ monthly users — the super-app, contact centre platform, and the channel, data and AI platforms beneath them. 150+ engineers through eight direct leaders.

2018 — 2022
AVP & Head of Engineering, Consumer App
India's largest telecom operator

Built a compliant payments platform from the ground up, re-platformed a monolithic self-care stack to microservices with zero downtime, and replaced proprietary infrastructure with an open-source standard that still holds.

2010 — 2018
Senior Engineering Manager, Flights
India's largest online travel platform

Promoted through manager and deputy manager roles across hotels and flights. Unified duplicated booking funnels behind a common aggregation framework and moved a monolith to service architecture on a business measured in billions of GMV.

2006 — 2010
Software Engineering
Global technology consultancies

Enterprise Java and distributed systems across telecom and technology consulting — where the habit of caring about what happens at the ninety-ninth percentile started.

Writing

Notes from the inside.

What actually happens when a large engineering organisation meets AI — written from inside one, not from the outside looking in.

Essay

Your AI pilot was never the hard part

Every organisation can produce a compelling demo in a fortnight. Almost none can put an agent in front of a hundred million people. The distance is not model quality — it is evaluation, guardrails, fallback design, and what happens when the agent is confidently wrong at three in the morning.

Read
Essay

The most underrated hire in an AI-native org is your QA engineer

When AI writes more of the code, the scarce skill stops being production and starts being judgement — knowing what should be true, and proving it still is. The people who spent careers asking "but what if it isn't?" turn out to be exactly who you need.

Read
Essay

What five billion events a day teaches you about simplicity

At a certain volume, cleverness becomes a liability. Every additional moving part is another thing that fails at scale, at the worst moment, in a way staging never reproduced. The systems that survive are almost boringly simple — and getting to boring is the hardest work there is.

Read
Speaking

Talks on building at this altitude.

I speak at industry conferences on platform engineering, agentic AI in production, and what genuinely changes when a large organisation goes AI-native.

TM ForumPlatform engineering at telecom scale
Elastic ConferenceSolving large-scale problems with open platforms
Mentor · Young Tech LeadersMentoring emerging engineering leaders
  • Putting agentic AI into production for a hundred million users — and what breaks first
  • Measuring engineering productivity honestly in the age of AI-assisted development
  • Platform strategy: how open-source foundations change an organisation's cost and speed curve
  • Leading an engineering organisation through AI transformation without stalling delivery
  • Reliability at scale — service objectives, active-active design, and the culture that sustains them

Building something that has to work at serious scale?

I write on AI-native engineering, platform strategy and leading technical organisations. Follow along on LinkedIn, or get in touch about speaking, advisory and leadership conversations.

Follow on LinkedIn Get in touch