How to apply for AI Engineer

About Elicit

Elicit is a public benefit company building an AI research assistant that helps researchers review literature and make better decisions. The product is a scalable ML system based on human-understandable task decompositions, with supervision of process rather than outcomes. This is a chance to work on AI systems where transparency and systematic reasoning are central design goals, not afterthoughts.

About the role

As an AI Engineer at Elicit, you will build and improve a language-model-based research assistant used for tasks like literature review. The role sits at the intersection of software engineering and applied ML: you will work on the harder backend parts of a scalable ML system. The work directly supports Elicit's goal of increasing good reasoning in the world and expanding understanding of safe AGI architectures.

A typical day

A typical day likely involves writing and reviewing backend code that connects language models to Elicit's research workflows. You might design or refine a task decomposition, inspect model outputs for transparency and reliability, and discuss architecture decisions with the team. Ask Elicit during interviews how they structure engineering time between product work and research.

Who Elicit is looking for

• A software engineer who enjoys difficult backend work and is comfortable building with large language models and generative ML systems.
• Someone who thinks in terms of human-understandable task decompositions and process supervision, not just outcome optimization.
• A builder who cares about transparency and systematicity in ML systems, and can reason about how those properties are implemented in practice.
• Someone interested in AI safety architecture questions and in making good reasoning more affordable for non-experts as well as pushing the frontier for experts.

Tips for this application

• Read Elicit's blog posts about AI engineering and their approach to process supervision, then reference specific ideas in your application.
• Look at Elicit's product and describe concretely how you would improve or extend its literature review capabilities.
• Show backend engineering depth: describe systems you have built that handle difficult data, model, or infrastructure problems.
• Explain how you have worked with large language models in production, including how you handled reliability, evaluation, or transparency.
• Connect your work to Elicit's stated goal of supervision of process, not outcomes, and to their public benefit company structure.

What to cover in your cover letter

['Why you want to build a scalable ML system based on human-understandable task decompositions.', "Specific backend or ML engineering work you have done that maps to Elicit's product and architecture.", 'Your view on process supervision versus outcome supervision, and how you have applied it.', 'How you think about making good reasoning more affordable for non-experts while pushing the frontier for experts.']

Research before applying

• Read Elicit's blog, especially posts about AI engineering and their approach to process supervision.
• Use the Elicit product yourself on a real literature review task and note where it works and where it fails.
• Listen to or read swyx's Latent Space podcast discussion where the term AI engineer was suggested.
• Look into what a public benefit company is and how Elicit's structure relates to its stated mission.

Likely interview topics

Based on the job description, expect questions about:

• How you would decompose a research task like literature review into human-understandable steps for an ML system.
• Your experience building backend systems that support large language models at scale.
• How you would implement supervision of process rather than outcomes in an AI research assistant.
• Trade-offs between transparency, capability, and scalability in ML system design.
• How you would evaluate whether Elicit is actually increasing good reasoning for its users.

Common mistakes to avoid

• Treating this as a generic ML engineer role and ignoring Elicit's emphasis on process supervision and human-understandable task decompositions.
• Focusing only on model training or benchmarks without showing backend software engineering strength.
• Writing about AI safety in abstract terms without connecting it to Elicit's specific architecture and product.