Software Engineer, Staff: Applied AI, Science & Engineering

About the role

At Anthropic, we're building AI systems that are safe, beneficial, and transformative. Our mission is to develop AI that benefits humanity, and we believe the most powerful capabilities emerge when we thoughtfully bridge the gap between research breakthroughs and real-world applications.

Claude is getting good at research. Given a well-posed problem, it can read the literature, write and run simulations, analyze results, and iterate on a design across physics, materials, chemistry, and engineering. But most of the hardest problems in science and engineering aren't well-posed. They live inside companies and labs with their own data, tools, constraints, and experts, and progress depends on turning all of that into work a model can actually do and check.

Our Applied AI, Science & Engineering team takes Claude to the scientists and engineers working on hard problems in energy, materials, hardware design, and other physical-world fields. We build the tools  Claude needs to do real research and engineering work. Mostly that means software: the agent harnesses, integrations, evaluations, and working processes that turn a partner's problem into something Claude can make progress on and validate. We find real-world problems and test what we build through partner engagements, then feed our learnings back to our research and product teams. It's early, and the engineers who join now will shape how Claude gets applied to science and engineering.

This is an engineering role first. You'll spend real time with domain experts, become the person who knows how to get Claude working in their field, and prototype quickly alongside them. Most of your time, though, goes into building and hardening the systems that make that work repeatable, not into writing recommendations or configuring someone else's product. A science or engineering background is a big plus, but what matters most is that you can earn the trust of expert researchers, turn a fuzzy problem into something concrete and checkable, and ship.

Responsibilities

• Build the agent harnesses and research loops that let Claude carry a problem from literature review through hypothesis generation, simulation, analysis, and design iteration
• Connect scientific computing and simulation tools (finite element, CFD, electromagnetic, or molecular modeling codes, for example) so Claude can drive them reliably and at scale
• Design and build evaluations that tell us whether Claude's work is actually right, and be honest about where it isn't
• Sit with scientists and engineers at partner organizations to learn their problem, data, tools, and constraints, and turn open-ended questions into well-specified, verifiable tasks with clear success criteria
• Prototype with partners, ship pilots into their workflows, and cut anything that doesn't move the problem forward
• Turn what works in one engagement into shared tools, reusable components, and documented processes the next one can start from
• Be the technical voice on how Claude performs on science and engineering work, explaining results, limits, and tradeoffs clearly to researchers, engineering leads, and non-technical stakeholders alike
• Partner with our research and product teams to share where models fall short on science and engineering work, and help shape what comes next

You may be a good fit if you

• Have 8+ years of experience building software, with strong engineering fundamentals and comfort across the stack, from data pipelines to tooling to quick interfaces
• Have built real systems with large language models, including agents, tool use, or evaluations
• Have worked on technical problems in a science or engineering field, such as physics, materials, chemistry, or electrical or mechanical engineering
• Have a track record of zero-to-one work in startup or startup-like environments\
• Can work directly with expert users, understand their workflows deeply, and still keep the focus on building the right system rather than the one first requested
• Have good judgment about what can and can't be verified, and are comfortable saying a result isn't good enough yet
• Bring high agency, pick up new domains quickly, and hold strong opinions loosely
• Communicate clearly with researchers, engineers, and external partners, and care about the societal impacts of your work

Strong candidates may also have

• An advanced degree or research experience in a physical science or engineering field
• Hands-on experience with scientific computing and simulation, numerical methods, or optimization
• Experience in industrial R&D, a national lab, or a deep-tech company in areas like energy, materials, manufacturing, or hardware
• Experience building alongside customers or technical partners, for example in forward-deployed, applied AI, or solutions engineering roles, ideally where you also owned the software that came out of it
• Experience with ML research, RL environments, or evaluation design

Candidates need not have

• 100% of the skills listed above

• Formal certifications or education credentials

• Expertise in every scientific domain we work in

• Direct machine learning or AI research experience

Deadline to apply: None. Applications will be reviewed on a rolling basis.