about me.

if you're reading this, you're probably trying to ship ambitious ai agents without lighting production on fire. same. by day i architect enterprise agent systems at a large language model company. nights and weekends i run evalops, focused on making agents reliable enough for real work.

i don't just write prompts. i build systems where agents reason, plan, and execute complex workflows. using tools like dspy, mcp, and custom runtimes, i build agents that can improve their own performance over time and handle ambiguity without breaking.

my background in security (vanta, threatkey, snap) means i build these systems to be robust. i know that an agent isn't useful if it's hallucinating or leaking data, so i bake reliability and evaluation into the core architecture.

translation: i obsess over durable systems, crisp feedback loops, and agents that earn their keep in production.

quick hits

  • powered by analog notebooks, single-origin espresso, and aggressively curated playlists.
  • architects autonomous agents that can reason, plan, and execute complex workflows.
  • runs evalops at night to prove that reliability is the most important feature of any agent.
  • takes thinking breaks on san francisco's 49-mile scenic drive with a field recorder for stray ideas.
  • hosts friday pizza nights starring doughvid—the temperamental sourdough starter who's somehow family now.
jonathan haas in san franciscotap for the full frame

experience

senior staff security engineer

stealth llm company, sep 2025 - present

architecting secure agent infrastructure and evaluation systems for next-gen models.

  • designing secure runtime environments for autonomous agents
  • building model evaluation pipelines for reasoning and safety
  • scaling production guardrails for high-stakes enterprise deployments

senior product manager

vanta, oct 2024 - aug 2025

brought on to advance vanta's next-generation security automation vision.

  • led strategic security integrations across cloud, code, and infrastructure platforms
  • built partnerships with wiz, github, gitlab, and other key security vendors
  • developed comprehensive integration strategy for compliance automation ecosystem

co-founder & ceo

threatkey, oct 2020 - oct 2024

pioneered ai evaluation methods for compliance automation

  • built custom evaluation datasets for security compliance tasks
  • developed model-graded evaluation framework for policy analysis
  • created domain-specific benchmarks for legal text classification
  • achieved 95%+ accuracy through rigorous evaluation methodology

lead, security operations

carta, nov 2020 - jul 2021

led security operations for equity management platform

  • built security operations from ground up
  • implemented incident response protocols
  • left to go full-time on threatkey

beyond work

when i'm not building agents, you'll find me slow-walking san francisco's 49-mile scenic route, cooking dishes from my travels across 37 countries, or hosting pizza nights featuring doughvid—my temperamental but talented sourdough starter. i believe the best ideas come from diverse experiences, whether that's exploring a new neighborhood, perfecting a recipe, or debugging code at 2 AM.

connect

whether you're tackling agent orchestration, building products that matter, or just want to swap stories about san francisco's hidden gems, i'd love to hear from you.

email / linkedin / github