Member of Technical Staff at Fulcrum (San Francisco, CA)

Fulcrum· San Francisco, CA·6d ago

Role details

Work type

Hybrid

Employment

Full-Time

Job description

Compensation: $200,000–$500,000 + equity (depending on experience and scope)

Location: San Francisco

Start: ASAP • Applications: Rolling

There is a $5k referral bonus for this role. You can refer people through our form airtable.com/form.

We’re hiring a member of technical staff to work closely with the founding team. You'll shape both our products and research direction.

Key Responsibilities:

Backend engineering: Design and build scalable infrastructure and systems for AI agents, including containerization, and performance optimization
Machine learning: Contribute to agent science research, and create evaluations for agent performance and behavior
Frontend engineering: Develop oversight interfaces and frontend systems for coding agent monitoring and control
Shape product strategy and research direction in collaboration with the founding team

High-impact mission: Work on foundational infrastructure for how humans will control and trust AI-generated software.
Elite technical team: Colleagues with top-tier ML publications, open-source influence, and MIT alignment background.
Culture of research and engineering: The team publishes research, builds open-source software, and operates at the intersection of ML research and systems engineering.
Early stage: Opportunity to shape the company’s product, culture, and direction.
Remote/hybrid: Not specified on careers page; likely flexible for early-stage startup.
Hiring roles: Currently seeking a "Member of Technical Staff" and a "Generalist"—general skills valued; no explicit degree or experience requirements listed, just impact orientation.

One-liner: Fulcrum is an AI lab building infrastructure and interfaces for human-AI collaboration in software development, with a focus on "elicitation" to extract optimal performance from AI models.
Entity Type: Private (early-stage startup)
Headquarters: Not publicly available on their website or careers page; likely US-based (founders with MIT affiliation).
Founded: 2025 (GitHub org created February 27, 2025)
Founders: Founders are from MIT’s AI alignment group, with top-tier ML research publications (NeurIPS, COLM), open-source contributions, and international olympiad medals. Specific names are not listed on the website.

Primary industry/industries: AI/ML Infrastructure, Developer Tools, Agentic Software
Target customers: B2B; AI researchers, engineering teams, and enterprises building multi-agent software systems
Mission or purpose statement: "AI agents are going to write most of the world's software. We want to make sure humans can understand, control, and trust what they build."

[Druids]: A batteries-included, open-source Python library to coordinate and deploy coding agents across machines. It provides a state machine framework, pre-warmed cloud VMs, and a UI/CLI/MCP for managing agent teams. Solves the bottleneck of human oversight by scaling intent across many agents with verification and control flow.

Valuation/Market Cap: Not disclosed
Key Metric: Total Funding—not publicly disclosed (likely angel or pre-seed stage). Open-source community traction: Druids launch (March 2026) with beta users applying it to open math problems, ML "autoresearch," and software optimization.
Notable Investors/Partners: Not publicly disclosed
Growth Signals: Rapidly building a developer community (Druids is open-source); 21 public repositories on GitHub; early beta users in research and engineering; actively hiring for technical staff and generalist roles.

Deep technical moat in elicitation: Focus on getting the best outputs from frontier models by structuring tools and processes, combined with infrastructure to measure output quality.
Founding team caliber: Top-tier ML research (NeurIPS, COLM), open-source software contributions, international olympiad medals, and leadership at MIT’s AI alignment group.
Unique product philosophy: Treats AI agents as a distributed systems problem (state machines, VMs, parallelism) rather than a prompt engineering problem—offering reproducibility and control at scale.

Open-sourcing infrastructure: Druids is the core product, released open-source to build a community and ecosystem.
Human-AI collaboration: The company's core thesis is that smarter models alone aren't enough—human oversight scaling requires new infrastructure.
Research-driven: Continues to measure model capabilities, find failure modes, and develop elicitation techniques for "the hardest and fuzziest tasks."

Sources