Artificial Intelligence Engineer

AI Engineer - Agentic Systems | Remote (South Africa) | US Eastern hours

I'm hiring for a US-based startup that builds and runs custom AI agents for mid-sized companies. Not a software product; a managed service. They design agents around a client's real workflows, deploy them into the tools those teams already use, and own the tuning and expansion month to month.

They're past the "will anyone pay for this" stage and into "how do we deliver this repeatedly." Self-funded, revenue in the door, and this is the first dedicated engineering hire.

What you'd actually do

• Triage agent behaviour in production: the agents open their own tickets and other agents close most of them, so you handle the ones needing human judgement
• Diagnose where clients hit friction: output quality drops, context bloat, tooling that handles a data source badly
• Build better tooling for the agents themselves
• Onboard new clients, each bringing custom logic, writing instructions for how an agent should reason over a client's data warehouse, building tooling for the queries it will run repeatedly
• Wrap client data and APIs into MCP servers so agents can call them
• Sit in client conversations, scope the work, then build it

The stack

• AWS end to end: Lambda, SSM for secrets, isolated instances per agent
• Terraform and infrastructure-as-code throughout
• OpenClaw as the base harness, heavily customised and wrapped in a custom gateway
• Provider-agnostic across models
• Python

They are not building foundational models and want to avoid maintaining their own harness. The engineering goes into making an existing harness work properly for real users.

What you need

• Production experience building and running agentic systems, not prototypes
• Strong Python
• Hands-on AWS and infrastructure-as-code experience
• Orchestration frameworks: LangGraph, LangChain or equivalent
• Multi-provider API integration experience
• A real methodology for evaluating non-deterministic systems: how you detect drift, structure validation, and feed failures back into the loop
• Practical context window management: compaction, RAG-backed context, token budgeting
• Ability to work consistent US Eastern hours

Strongly preferred

• Hands-on time with open-ended agent harnesses rather than only tightly-scoped pipelines
• Client-facing or professional services experience
• MCP server development
• BI or data-heavy agent use cases