Deepak Nair

I write about how enterprise leaders turn AI capability into systems that improve how work gets done, earn trust, and get better over time.

My work connects models and systems, people, and enterprise readiness: the workflows, ownership, evidence, and operating choices that determine whether AI creates durable value.

I currently work at Google, shipping advertiser-facing AI agents. If you use Google Ads support, I’d love to hear your feedback on how we’re doing.

Working through an AI operating-model, workforce, or evaluation decision? Email me.

Operating thesis

Frontier models are improving quickly. Enterprise value depends on more than model capability: it depends on workflows, tools, policy, human oversight, evaluation, and trust. My work explores how those pieces come together in customer experience systems.

Three lenses

• Technology: What can models and agents actually do?
• People: How do people understand, supervise, trust, and work with them?
• Enterprise readiness: What policies, tools, workflows, and measures make deployment useful and safe?

What I’m working on

• SupportBench: an early-stage, open research benchmark for evaluating AI-native customer support systems.
• Small, reproducible experiments on memory, retrieval, tool use, escalation, and policy adherence.
• Open-source tools that help engineers build support systems customers can trust.

Practical tools

• Enterprise AI operating-model checklist: seven decisions to pressure-test before expanding an AI workflow.

What I believe

• Real-world AI applications should be evaluated as systems, not just as LLMs.
• Making AI agents reliable in enterprise settings is hard but possible.
• Model capabilities are often ahead of how reliably we know how to use them.
• There is immense value in the enterprise waiting to be unlocked.

Research Notes

• The Work of Making AI Matter in the Enterprise
• The Questions Behind Enterprise AI Strategy
• A Practical Enterprise AI Readiness Diagnostic
• SupportBench, Stage 3: When Escalation Replaces Action
• SupportBench, Stage 2: A First Local Pilot
• SupportBench, Stage 1: Defining the System We Want to Evaluate
• Evaluating AI-Native Customer Support Systems
• Coming next: a second local baseline and customer support as a proving ground for enterprise AI.

About this work

This is an independent, early-stage research project, not a finished standard or a model leaderboard. The goal is to share evidence, methods, wins, and failures in public. Examples use synthetic data or publicly available information.

You can email me at hey@deepnair.ai, find me on X, or connect on LinkedIn.