About
Paysend is transforming how people and businesses move money around the world. Its products and services help millions of customers stay connected globally.
Responsibilities
•Design end-to-end AI-assisted delivery workflows, including agent responsibilities, execution boundaries, decision points, human intervention, failure handling, recovery, and escalation
•Design and refine AI-driven pipelines that automate coding, code review, testing, and deployment
•Build and optimise agentic workflows using Claude Code and the Anthropic API, including custom tooling, MCP server integrations, sub-agents, and stateful orchestration logic
•Engineer robust prompts, benchmark suites, and automated evaluation datasets; measure and improve pipeline accuracy, latency, and API costs
•Define workflow-specific success criteria and operational evidence, including correctness, reliability, intervention and escalation rates, failure modes, and recovery effectiveness
•Integrate Anthropic capabilities into internal developer platforms, including tool use, structured outputs, prompt caching, and batch processing
•Handle production work including sandboxed execution, strict error handling, retries, observability, guardrails, and edge cases
•Own systems after implementation by observing their behaviour in engineering workflows, investigating failures and incorrect behaviour, and improving them based on operational evidence
•Partner with engineers across the organisation to identify automation opportunities and safely roll out agentic tools with human-in-the-loop review
Requirements
•Strong software engineering background and experience designing, building, and operating production systems
•Strong hands-on experience with Python or TypeScript
•Ability to decompose end-to-end engineering workflows and identify decision boundaries, failure modes, dependencies, observability needs, and recovery paths
•Hands-on experience with Claude Code beyond casual experimentation, including its strengths, limitations, and configuration (CLAUDE.md, sub-agents, hooks)
•Experience building with LLM APIs, including prompt engineering, tool/function calling, structured outputs, context window optimisation, agentic loops, and evaluation
•Familiarity with MCP server integration and custom workflows
•Experience turning prototypes into production systems, including CI/CD integration, containerized execution, reliability engineering, and monitoring
•Pragmatic judgment about where AI automation helps and where deterministic logic or human oversight is needed
Nice to have
•Experience with agent orchestration frameworks or custom agent harnesses/tooling
•A track record shipping internal developer platforms or CLI tooling
•Contributions to open-source AI frameworks, developer tools, or MCP servers
Conditions
•Applicants from outside Serbia are welcome; quarterly meetings at the Engineering Hub in Belgrade are expected
