Steven M. Ledbetter

(845) 379-0641

NYC Area, NY

Summary

Daily Claude Code power user building agentic developer tools. Author of "Structure Beats Scale," a white paper showing structured review pipelines outperform brute-force generation in LLM code synthesis. Co-inventor of a patented ML model (US12149553B1). 20 years in product management, three successful exits, founder background — the creative hacker spirit this role demands.

Employment

Meta

2023 – Present

Delivering the most performant data pipelines in the world.

Product Manager, AI & Data Infrastructure

  • Own the freshest, fastest SQL engine at Meta — used by ML resarchers, ads and content serving in Real Time.
  • Developed a declarative MCP, catalog, and orchestration agent swarm that continuously optimizes warehouse queries with a novel safety/alignment mechanism — converting research prototypes into production infrastructure.
  • Co-led the transition of the PM function in Data Infrastructure to an AI-Native, agentic development role with 10x productivity — identifying how emerging LLM capabilities could transform internal workflows.
  • Led capacity conversations for 20+ internal customers with their finance and capacity teams, synthesizing conflicting feedback into actionable specifications.

Supernatural (acquired by Meta)

2022 – 2023

Led product through acquisition by Meta.

VP of Product

  • 3x ARR and 2x user retention across millions of members; described as "the closest Meta ever came to realizing the metaverse" (Pranav Dixit, Business Insider).
  • Guided product through $400M+ acquisition by Meta, surviving an FTC antitrust challenge — navigating extreme ambiguity from first-principles.
  • Built 0-1 real-time multiplayer infrastructure supporting 10k+ concurrent users — shipping an ambitious, commercially successful product in a novel technology space.
  • Led a team of 17 PMs, UXR, and Program Managers through hypergrowth and acquisition.

Elevate Security (acquired by Mimecast)

2020 – 2022

First product hire; built and steered product team to acquisition.

VP of Product

  • Co-inventor, US12149553B1: novel ML risk-scoring model and automated intervention system — involved in research translation, model selection, evaluation design, and tuning.
  • First product hire; built the Product Team from scratch including PMs, POs, Designers, and Technical Writers.
  • Repositioned product against legacy security awareness training vendors by reframing the category around measurable risk reduction — differentiating on outcome metrics vs. completion rates.
  • Pivoted company to new products that 2x ARR with a 60 NPS score and less than 5% churn.

Tonal

2018 – 2020

Owned engagement experience, pre-launch through scale.

Director of Product Management, Engagement

  • Joined pre-launch as the first PM working horizontally across hardware, firmware, software, and content teams — shipping a unified product experience in a novel technology space.
  • Increased engagement 2x and customer satisfaction 50% through data-driven experimentation, applying research from behavioral psychology to product metrics.
  • Developed automated content production pipeline and workflows that increased output 27x.

Habitry (acquired by Lift The Bar)

2013 – 2018

Co-founded, built, and sold.

CEO, Head of Product, Co-Founder

  • Co-founded and sold a Motivation Science consultancy; built an iOS behavior change platform with 72% daily adherence and 68% retention at 365 days, serving 16,000 users across 6 continents.
  • Translated academic research (Self-Determination Theory, Motivational Interviewing) into product specifications that drove measurable behavior change outcomes.
Education

University of Chicago

2004

Bachelors in Philosophy

Harvard University

2001

Undergraduate Coursework in Economics

John F. Kennedy University

2014

Masters in Sport Psychology

Writing

https://github.com/smledbetter/structure-beats-scale

White paper: structured review pipelines outperform brute-force generation in LLM Python code synthesis across 3,000+ runs on HumanEval, MBPP, and competitive programming benchmarks using Sonnet and Mercury 2.

https://smledbetter.com/blog

Blog series on AI-native product management — sprint methodology with AI agents, coaching agent teams, and building encrypted apps with LLMs.

Skills

Product Leadership

Agentic AI & LLMs

Machine Learning

Data Infrastructure

Python & SQL

Privacy & Security

Open Source Projects

Flowstate

https://github.com/smledbetter/flowstate

Sprint-based development workflow framework for Claude Code.

  • Quality gate framework with automated checks (tests, types, lint, coverage) for AI-assisted development.
  • Three-phase sprint structure (Think, Execute, Ship) that prevents scope drift on complex work.
  • Metrics pipeline tracking token usage, session time, test counts, and code coverage across sprints.

Uluka

https://github.com/smledbetter/Uluka

Security claim verification CLI for code.

  • Extracts security claims from comments, README, and JSDoc, then verifies they match actual implementation.
  • Multi-language code parsing via Tree-sitter with taint-tracking data flow analysis across files.
  • Optional Claude-powered verification for nuanced security properties, with deterministic rules running first.

Nexa Scheduler

https://github.com/smledbetter/Nexa-scheduler

Compliance-aware workload scheduler for GPU and CPU clusters on Kubernetes.

  • Built a privacy-first Kubernetes scheduler enforcing data sovereignty, org isolation, and confidential compute constraints at scheduling time — including topology-aware GPU placement and gang-scheduling for multi-node training jobs.
  • Fail-closed design with preemption policies and quota enforcement ensures pods stay Pending rather than compromise isolation guarantees.
  • Full audit trail of every placement decision with structured JSON logging and Prometheus metrics for queue depth, scheduling efficiency, and resource utilization.

Weaveto.do

https://github.com/smledbetter/Weaveto.do

Privacy-first, agent-augmented task coordination for decentralized teams.

  • End-to-end encrypted task coordination platform using Olm/Megolm (vodozemac WASM) with zero-knowledge relay.
  • WASM-sandboxed custom agents with encrypted state and Ed25519 signature verification.
  • No accounts — device-bound identity via WebAuthn PRF with ephemeral-by-design encryption keys.