Job description
Our client is looking for Agentic AI Engineer to build and deploy AI agents that automate real business workflows — reducing operational cost, headcount dependency, and turnaround time.
This is a hands-on implementation role, not R&D.
If your experience lives in notebooks, courses, and demos, this role is not for you.
What You'll Own Design and build multi-agent systems that replace or augment manual business processes Integrate agents with internal tools (CRM, ERP, databases, APIs, communication platforms) Optimize agent performance for cost per task — token efficiency, latency, and accuracy Set up observability and monitoring so agents don't fail silently in production Work directly with business stakeholders to translate process pain points into agent workflows Maintain and iterate on deployed agents — not a build-and-forget role Remote working At Substance, we’re all about action, not just talk.
If your profile aligns with what we need, you’ll hear from us within 1-2 weeks.
If not, no fluff—just know we value your interest and will keep you in mind for future roles where your skills can make an impact.
We focus on real connections and meaningful matches, so when the right opportunity comes, we’ll be ready to make it happen.
Getsubstance.
co Pte. Ltd. | EA License No: 24C2398 Minimum 2 production-deployed agentic systems — not POCs, not demos, not coursework.
Real systems, real users, real business impact.
You must be able to speak to what broke, how you fixed it, and what it saved Proficiency in Python, LangChain/LangGraph or equivalent orchestration frameworks Experience with tool-calling, RAG pipelines, memory management, and multi-agent coordination Familiarity with MCP servers and API integrations Cost-aware engineering mindset — can justify model choice (when to use GPT-4o vs Haiku vs Sonnet) based on task requirements, not preference Understanding of guardrails, human-in-the-loop design, and failure handling How We Evaluate You — Interview Process Theory will not get you through.
Expect: Production walkthrough — show us a live or previously deployed agent.
Walk us through the architecture, what failed, and how you resolved it Cost breakdown — explain the token/compute cost of a system you built and how you optimized it Live build task — given a business process, design an agent workflow on the spot What Good Looks Like Agents running in production, not in slides Clear ROI per agent — time saved, cost reduced, error rate dropped Builds lean — chooses the cheapest model that does the job reliably, not the most impressive one What Will Get You Rejected Portfolio of tutorials and Jupyter notebooks with no deployment history Can explain LangChain architecture but never shipped one to production Optimizes for technical elegance over business outcome What You Won't Be Doing Academic research or model training Waiting for perfect requirements — expected to drive clarity from ambiguous briefs Building without a business case
This job post has been translated by AI and may contain minor differences or errors.
