Anthropic
Software Engineer, Tokens and Prompt Structures
$320K–$405K
San Francisco, CA +2 · Full-time · Posted Sep 16, 2026
- Location
- San Francisco, CA
New York City, NY
San Francisco, California, United States - Experience
- 5+ years of software engineering experience; the posting also says required years will correlate with internal job-level requirements.
- Education
- Bachelor’s degree or an equivalent combination of education, training, and/or experience; relevant field of study demonstrated through coursework, training, or professional experience.
- Visa sponsorship
- Visa sponsorship is available, but not for every role or candidate; if an offer is made, Anthropic says it will make every reasonable effort to obtain a visa and provides immigration-lawyer support.
About the job
Owns and improves the libraries and abstractions that encode text and multimodal data and define Claude’s prompt structure, helping research ideas move from experiments into production.
What you'll do
- Maintain and improve encoding libraries; design abstractions for encoded data and prompt structures.
- Run experiments on feeding structured data to Claude and adapt libraries to new research directions.
- Optimize encoding performance across dependent systems and support production launches of research ideas.
What they ask for
- 5+ years of software engineering experience, including meaningful time maintaining libraries, SDKs, or developer-facing APIs.
- Familiarity with ML terminology and LLM architecture sufficient to work with researchers and understand experiment results.
- Experience carrying out complex refactors in large codebases.
- Strong communication skills and ability to collaborate with researchers and engineers.
Nice to have
- Experience with tokenizers or other text/data encoding systems, long-term library maintenance, or performance optimization.
- Python and/or Rust; reinforcement learning or model training infrastructure.
Pay
$320,000 — $405,000 USD (base salary)
Tools
Python, Rust
AI focus
Runs experiments on how to feed structured data to Claude; adapts encoding libraries for research directions and ships multimodal data types and other research ideas to production.
Model-written summary · the original posting is authoritative