AI Agent Engineer (Contract): Build the Agent Layer

Top 3 reasons to join us

• Build a greenfield AI agent suite end-to-end
• Remote, real autonomy over your work
• Direct access to the founder & senior engineer

Job description

About PantoMartech
PantoMartech builds marketing technology apps. One of our apps is an AI-powered product that helps online retailers improve their digital advertising performance. It connects to their e-commerce store and gives them a far more accurate picture of what's actually driving results. We're a small, fast-moving team, and we're now building the layer that turns accurate data into automated decisions: a suite of AI agents that read a store's real performance, ad performance, and customer behavior — and act to improve outcomes.

Why this role exists

We've built the hard part — the data, the enrichment, and customer models. The product and UI for the agent layer are already designed. We need an engineer to build the agents themselves.

This isn't one agent — it's a growing suite to support everyday e-commerce operators.

Every agent ships in stages: recommend → one-click approve → autonomous within guardrails. Trustworthy, explainable output is the product. "Black box" is a failure mode for us — each action must show its reasoning and sit in an audit log.
 

What you'll build

• The reasoning and orchestration layer for multiple agents (tool use, planning, structured outputs, evals) over our e-commerce performance, ad performance, and customer behavior data
• Read and write integrations with ad-platform APIs — Meta, Google Ads, TikTok — for pulling performance and pushing audiences and budget actions
• Recommendation logic plus the "why" explanations that make every output auditable
• The agent-facing screens (mockups are ready: recommendation feeds, approval flows, autonomous activity logs, permission/guardrail controls)
• Evaluation harnesses to measure recommendation quality and catch regressions
• Safety scaffolding: dry-run modes, scoped permissions, and rollback paths for anything autonomous

Your skills and experience

Who we're looking for

We want a strong generalist who can take a designed feature and ship it end to end — and who uses AI tools (including Claude) to move fast without cutting corners on safety. You won't get detailed tickets; you'll get a clear goal, ready designs, and direct access to the founder and our senior engineer.

You're a great fit if you:

• Have shipped real, production software end to end (backend, data, and a clean web UI)
• Have built LLM-powered features beyond prompt-tinkering — tool/function calling, multi-step orchestration, structured outputs, and basic evals
• Are comfortable integrating third-party / OAuth APIs (ad platforms a strong plus)
• Default to explainable, safe automation — you build the audit log and the dry-run before the "go live" switch
• Work well solo and async, and use AI coding tools effectively to punch above your weight

Nice to have

• E-commerce / performance-marketing background (ROAS, CPA, COGS, contribution margin, LTV)
• Hands-on with the Anthropic API / Claude (tool use, evals)
• Experience with agentic products that take real-world actions, not just chat

Engagement

• Freelance / contract, project-based — roughly 3–6 months to build the first few agents, with strong potential to continue as the suite grows
• Remote within Asia
• Competitive compensation for the market

Why you'll love working here

Work that matters, end to end Own the agent layer for a real product, take ready designs and ship them to production. No detailed tickets, no micromanagement: a clear goal, finished mockups, and direct access to the founder and our senior engineer.

Autonomy and flexibility

• Fully remote within Asia
• Async-friendly, you manage your own schedule
• Project-based contract (~3–6 months), with strong potential to extend as the agent suite grows

Modern tooling, real impact

• Build with a current AI stack, LLM orchestration, tool use, evals, and the Anthropic API / Claude
• Ship agentic features that take real-world actions, not just chat
• Work on explainable, safe automation that customers actually trust

Compensation

• Competitive compensation for the market