• I'm Venky Chivukula — eighteen-plus years in technology, and the throughline is making regulated infrastructure production-ready at scale. Most recently, that has meant rebuilding operations around AI agents — and deciding exactly how much those agents can be trusted when privacy and compliance are on the line. (pause)

Setting the Quality Bar for Agents

• Today I'm Director of Technical Program Management at a global financial-services enterprise. On our Internal Developer Platform — 300+ production clusters, 10,000+ engineers — we put agents to work on developer and on-call toil, and I owned the bar they had to clear before they were trusted. (pause)
• We started with the toil we already knew how to fix, because that's the only kind where you can prove an agent got it right. Every agent was scored on consistency, accuracy, and auditability against real historical cases, and earned the right to act one stage and one class of action at a time — anything irreversible or touching a compliance control stayed human-approved. (pause)
• Eight agents went through that process. Alongside the platform work, deployment lead time went from two weeks to under 45 minutes, Sev-1/Sev-2 incidents fell 40%, and automated compliance held at 99%+.

Production Readiness for Privacy & Compliance Infrastructure

• I also owned production readiness and engineering quality for the systems that keep our privacy and compliance infrastructure reliable — deletion and data-subject request pipelines, PII classification, consent, and audit logging. (pause)
• Those systems fail silently: a deletion that only partly runs still looks green. So I made readiness measure correctness, not just uptime — correctness targets, synthetic test customers, and daily reconciliation that caught silent failures within 24 hours instead of weeks later in an audit. (pause)
• An incident pattern agent analyzed every incident and near-miss, and each confirmed pattern became a readiness check for every system — about 15 of them — with people approving every change to the bar.

Catching Defects Before They Ship

• I led the rollout of agentic code review and automated test generation across our engineering teams, after aligning Security, Legal, and Core Engineering on the guardrails first. (pause)
• It cut legacy technical debt 70% and improved code quality 40% — and for privacy systems, it caught defects like unclassified personal data before they ever reached production.

Regulation Readiness — GDPR at Amazon Ads

• Before that, I spent four years as a Principal TPM at AWS and Amazon Ads. (pause)
• At Amazon Ads, I led the ad-supported streaming infrastructure for Prime Video and Freevee — 120M+ monthly active users, 99.99% availability — including the GDPR program for the advertising rollout. (pause)
• We built consent, data minimization, and deletion into the shared APIs and SDKs every partner used, so compliance was a property of the infrastructure, not a sign-off at the end.

Scaling Playbooks Across Teams

• Both the agent evaluation bar and the privacy readiness bar started with one team — and I scaled each across the organization. (pause)
• The approach: standardize only the non-negotiables, ship everything else as reusable tools instead of documents, and let the agents carry each new lesson from one system to every system.

Capacity: GPU, AI & Tokenomics

• At AWS, I also ran GPU capacity planning for ML training and inference across three Availability Zone launches, with zero Sev-1 readiness-related incidents in the first 30 days after GA. (pause)
• I apply the same discipline to agents: cost per agent task measured against the toil it replaces — tokens are just the newest kind of capacity.

Fit

• What draws me to this role is that it's the exact problem I've been working on: automating PREQ work with agents, where the output informs privacy and compliance decisions and "pretty good" isn't good enough. (pause)
• From what I've read on Meta's engineering blog about Privacy Aware Infrastructure, the Risk org has already made privacy a property of the infrastructure itself. (pause)
• The next step is making the operations of that infrastructure agentic, without giving up the correctness bar that makes it trustworthy.

Closing

• I'm genuinely excited about the chance to define what trustworthy agentic operations look like for Meta's Risk org — setting the bar agents have to clear, deciding where humans stay in the loop, and scaling that playbook across risk areas.