Description
Scale’s General Agents team is trying to solve a difficult productization problem: build agents general enough to work across many enterprise customers without turning every deployment into a completely custom system.
That requires architecture beyond basic prompt chains.
The engineer will design agents that combine LLM reasoning with external tools, APIs, databases, memory, and explicit control logic. A useful system may need to plan several steps, execute actions, recover from failures, preserve context, and behave consistently even when each customer’s data and integrations are different.
Generalization is therefore one of the main technical goals. Instead of building one workflow that works for one client, the team wants reusable agent architectures that can be deployed across recurring enterprise problem domains.
Evaluation is part of the engineering loop. Engineers will build datasets, environments, metrics, and testing frameworks that measure reliability and business impact under production conditions—not just benchmark performance in a laboratory setting.
The role also covers deployment and continuous improvement. Once an agent reaches customers, production traces and failure modes become new sources of information for redesigning prompts, tools, control logic, memory, or model behavior.
Modern agent techniques such as multi-step reasoning, tool calling, planners, multi-agent patterns, and context optimization are directly relevant. Fine-tuning methods including SFT, RLVR, and LoRA can also be used when system-level prompting alone is not enough.
Strong Python engineering is expected because these systems need to remain testable, observable, and maintainable. Experience integrating LLMs with APIs, services, cloud infrastructure, and databases is particularly useful.
At Staff level, the position also contributes to technical direction and agent-development standards across the organization.
This is a good fit for someone who enjoys the boundary between applied ML and software engineering: frontier techniques matter, but they only count when the resulting agents behave reliably for real users.
Curated opportunity. Please verify details and apply via the original link below. No Signals are required for this project/job.
https://scale.com/careers/4658162005