Eleven years in enterprise SaaS, the last three building agentic AI that enterprises run in production: composable agent workflows at Aisera, and an AI platform with 50+ agents and 20+ MCP connectors serving 1M+ users at RingCentral.
Follow one request end to end. The console on the right replays a real workflow pattern; each step of the story lights up the part of the trace it describes.
No form, no ticket category, no "please select from the following." The agent takes the request as it is spoken and works out what is really being asked.
Which systems hold the answer, in what order to ask them, and what a good outcome looks like. The plan is visible, so a human can see the reasoning before anything happens.
Identity, endpoint, security, CRM, support, HRIS. Each call is scoped to what the requesting person is allowed to see, and each result is kept with its source.
This gate is the product. Routine, low-risk actions run. Privileged ones route to the right approver. Nothing ambiguous happens silently.
An account is unlocked, a brief is delivered, a laptop is ordered. The person is told what happened and why, and the trail is there for anyone who needs it later.
Most "AI assistants" stop at a chat window. The hard product work starts when the model has to read from ten systems, decide what to do, and be trusted to do it. That is the layer I own.
One governed agent runtime, many business domains. At Aisera, enterprise customers deployed composable agent workflows without custom engineering; at RingCentral, a 144-tool marketplace is becoming a headless, API-first AI platform, and a multi-domain assistant spans IT, HR, Sales, Finance, and Support.
Agents are only as useful as what they can reach. I have shipped 20+ enterprise MCP connectors, the first six in a single quarter, set the reference architecture every connector since has followed, and launched usage-based billing for what agents consume.
The work that gets an agent past the CISO and the CFO. An AI security agent that scans every codebase before launch; telemetry that diagnoses agent behavior in production; and a three-phase ROI system that gives executives productivity in numbers.
Five products across two companies, one thread: give agents real access to enterprise systems, make them useful to the people who rely on them, and engineer the trust needed to let them act. Open any project for the full case study.
Working positions, earned from shipping. I am writing each of these up in more depth.
Thumbs-up rates and resolution counts drift away from real user value within weeks of launch, and the dashboards keep smiling. Agent products need a standing practice for detecting that decay, and a culture willing to reset the numbers when it happens.
An agent that does everything autonomously is a liability; one that asks permission for everything is a chatbot with extra steps. The design work is deciding, per action and per role, what runs, what asks, and what escalates.
Models are already capable enough for most enterprise workflows. What blocks deployment is who the agent is when it touches a system of record. Non-human identity, scoped service principals, and audit trails are where agent programs are won or stalled.
Eleven years in enterprise SaaS across two continents: engineer, then product manager, then three years building agentic AI that enterprises run in production.
Own strategy and roadmap for AI Garage, RingCentral's enterprise AI platform: a 144-tool marketplace with 50+ live agents becoming a headless, API-first product for 1M+ users. Launched 20+ enterprise MCP connectors, the first six in one quarter. Drove Ringo, a multi-domain assistant on Claude spanning IT, HR, Sales, Finance, and Support, and lead agent-first IT support across seven enterprise systems. Built an AI security agent that cut deployment friction 35%, agent telemetry and ROI measurement for executives, and usage-based billing for MCP consumption.
Owned agentic AI platform products for Aisera's enterprise SaaS offering: composable, self-serve agent workflows that customers and implementation partners deployed without custom engineering, driving $150M+ in joint business value. Built agent analytics that cut abandonment 27% and raised task success, and orchestration blueprints that cut implementation time 42% and accelerated deal closure 26%.
Drove development and go-to-market of a digital learning platform MVP with analytics-driven cohort personalization.
Led a 29-person cross-functional team on enterprise SaaS products and ran 100+ customer interviews that shaped the roadmap, lifting adoption and engagement 10%. Earlier, as an engineer, built AI-driven geo-fencing and automated content management modules for ManageEngine MDM, used by 500K+ enterprise users.
Students' Leader, Peer Career Consultant, and Teaching Assistant for Product Definition and Validation.
I am most useful on problems where the model is the easy part and the platform, permissions, and proof of value around it are the hard part.