Member of Technical Staff – Research Engineering
Hone builds AI that owns outcomes, not tasks: agentic systems that work towards goals over weeks and months, deploy themselves, and improve over time.
About Hone
Models keep getting smarter, but outside of coding most organizations can't absorb that intelligence, and the gap between model capability and the value derived from it keeps widening. The reasons are many, from human incentives to org design, but at the core we think it's a technology problem. Hone builds engines: AI systems that you hand a goal, like "keep the recruiting pipeline healthy." An engine builds itself, maintains its own agents and code, pulls in people when it needs judgment, and learns from every correction.
In October 2026, we announced a $60M seed round from Benchmark and Index Ventures.
Half our team are former founders: They’ve built companies, shipped products, and chosen Hone as the place to build next. The rest of us come from Cognition, Mercor, Ramp, Jane Street, Decagon, Brex and Harvey
Our customers are building with AI themselves: Cognition, Modal and Lucanet already use Hone, with customer perspectives featured on our website
Our founders have built from research to scale: They researched agents at Stanford, built the core of Devin at Cognition, led enterprise engineering at Mercor, built internal AI platforms, and founded companies that reached over one million users
We measure our work by economic impact: Engines are deployed across revenue, procurement, risk, product and operations, and we judge them by the results they create, not benchmark scores
Our values shape how we work: Own the Outcome, Set the Standard, Win with Class, and Commit to the Racing Line. Read more on our careers page
Read more: Bloomberg, Index Ventures’ investment thesis, and our launch announcement
The Role
Engines run for weeks and months inside real companies, where goals are rarely fully specified and feedback arrives late, partially, and through people. You'll own how engines improve: offline evals, simulations, memory, and the self-improvement loops that turn production feedback into verified changes to skills, memory, code and policies.
What You'll Work On
Simulations built from company context with simulated users and connectors to evaluate engines on tasks that take weeks or months.
Turning production traces into simulations that we can rewind, fork with one change, and run forward, keeping the simulated world consistent.
Memory systems that store, retrieve, and compact what engines learn.
Self-improvement loops that turn traces into verified changes to skills, memory, code and policies.
Grading and credit assignment over month-long trajectories.
Requirements
2+ years of engineering experience, or equivalent depth from research, open source or your own projects
Time at a fast-growing startup
You've shipped applied ML systems from scratch as an engineer, not only papers
You've run evals and experiments rigorously against production systems
Strong fundamentals: ML evaluation, experimental design, statistics, LLM agents
You use coding agents every day and still understand every line you ship
You build and tinker for fun because you genuinely care about how the technology works
Nice to Have
Harness, post-training or applied AI experience at a lab or applied AI team
Published research, especially on agent memory or continual learning
How We Hire
Intro with a co-founder.
Technical screen.
Research deep dive.
Two-day work trial in San Francisco.
References.
Offer.
Logistics
Location: San Francisco, in person, Sunday to Friday.
Compensation: Competitive salary + meaningful equity.
Visa: We sponsor visas and support you through the immigration process.
Benefits: Health and dental insurance, maternity and parental support, and relocation support.