After shipping large scale consumer AI products at Microsoft and now a senior engineer at Google Deepmind, Ankit Jain is rethinking how knowledge work gets done and how to make organisations AI native.
Ankit Jain has spent much of his career working on the systems that turn powerful AI models into products people can actually use. A computer science graduate of Punjab Engineering College, Jain joined Microsoft as his first job in 2019.
Right after his undergrad, he got the opportunity to work on software he had played around as a kid. “From using Microsoft Word in my school’s computer lab as a kid to building it was a dream come true”, he says. Before the launch of generative AI, his work landed multi lingual dictation in products like Outlook and Word. He advocated for expanded dictation features in regional languages - a problem he had seen his mother personally struggle with. His insights on customers led and identifying core product bugs led to a strategic rollout that spurred a 17% usage increase within a month. His work then evolved to large scale backend systems powering generative AI experiences across products like Word, Powerpoint, Outlook and Teams.
His focus increasingly moved towards some of the harder problems surrounding AI models: retrieval, evaluation, context, latency and reliability.
He also built an agent that automated parts of a Responsible AI evaluation workflow, reducing a process that could take around three weeks to approximately five minutes. He took up this effort which led to productivity increase across the Office Org at Microsoft. In separate work, he helped reduce latency for AI responses using web search from around 46 seconds to 15 seconds.
His work also led to product innovation. He engineered a system for creating podcast-style experiences from documents and on technology related to automatically presenting information in PowerPoint. Patent applications were filed for both.
After Microsoft, Jain moved to Google DeepMind in late 2025, continuing his work around applied AI and knowledge work.
Asked what advice he would give to students entering computer science today, Jain says technical foundations still matter, but they are no longer enough on their own. He believes as intelligence gets commoditized with AI, human relationships, empathy and human agency become even more valuable. When building becomes easier with AI, the skill of knowing what to build becomes more important. He believes one of the ways to get ahead in an AI first world is to better understand humans.
On his advice for transforming knowledge work within organisations, he believes that the next real value of AI will be harnessed not by adding AI to existing workflows, but by rethinking and reimagining how work itself is done. “The model is only one part of the system,” Jain says. “The harder question is whether the AI has the right context, whether its output can be trusted and whether it actually reduces work for the user.” His experience has shaped a distinction he believes will become increasingly important: the difference between organisations that are merely AI-enabled and those that are genuinely AI-native.
“Most companies are adding AI to workflows that were designed before AI existed,” Jain says. “Being AI-native means asking whether those workflows should exist in the same form at all.”
Most business processes were created around human and software limitations. Information is read, analysed, documented, reviewed and passed between teams. AI can make each of those steps faster.
Jain argues that the larger opportunity is to reconsider the workflow itself.
“If intelligent systems had existed when many of today’s processes were designed, we probably would not have designed them the same way,” he says. One of the biggest challenges, he believes, is organisational context. Important information is often fragmented across documents, messages, meetings and people, while old decisions can remain accessible long after they have been superseded.
Future enterprise AI systems, Jain argues, will therefore need to understand not just documents, but relationships between projects, decisions and information over time.
“A confident answer based on outdated context can create more work than it saves,” he says.
Across his career, Jain’s work has moved from building software at scale to generative AI, agents and the broader redesign of knowledge work.
For him, the next phase of enterprise AI will not be defined simply by putting an assistant into every product.
It will begin when organisations ask a more fundamental question: if intelligence had always been available, would we have designed work this way at all?
