Software Engineer for Agentic AI (Junior–Mid Level)

Agentic Software Engineer (Junior to Mid Level)Location: Kuala Lumpur, beside Semantan MRT station. View office locationHybrid: 3 days in the office, 2 days from homeType: Full-timeWe build AI agents that change how work gets done. Come build them with us.Everyone here builds. Including the CEO. You will never report to someone who only knows how to talk about the work. We judge people by what they ship, not how many years they have on a CV. Bring something you’ve made.Who we arePantas is an AI-native software company in Kuala Lumpur. We started in climate data and became Asia’s first PCAF-accredited financed emissions platform, live with banks and corporates across Southeast Asia. That gives us something most AI startups don’t have: real customers, real data and real problems.Now we are building the next layer. AI products that do the work for people, not dashboards people have to stare at.What you’ll build

• Pantas Brain. A company brain inspired by Andrej Karpathy’s LLM Wiki. It turns everything an organisation knows into a living knowledge base that people and agents can use.
• Axsona. AI Twins. Synthetic populations of demographically accurate AI agents that simulate how real audiences react to strategy, messaging and products. Test the message on a simulated population before it hits the real one.
• Agentic sustainability platform. Agents that collect data, flag anomalies and carry a case from one person to the next without anyone chasing emails. We design from the end state and work backwards. We don’t patch yesterday’s workflow.We’re done building CRUD. If you want to wire up forms and database tables, this isn’t the job. We build software that reasons over information, uses tools and moves work forward. You’ll build the systems that make it dependable enough for real customers.Everyone builds. Including the CEO.No managers who only know how to talk. Our CEO writes code, builds products alongside the team and runs monthly workshops on AI agents, MCP, context engineering, harness engineering and loop engineering. Weekly AI and engineering sharing sessions mean what one person learns, everyone gets. Our engineering leads ship every week.Everyone is expected to code and build, business development team included. BD colleagues who had never written code have built agents that now do real work.Your teammates will challenge your thinking, review your work and help you get unstuck. You bring the initiative.A team that makes you sharperOur engineering team is 18 strong, with a very high talent density. Three engineers hold a Master’s and one a PhD in Computer Science, from top local and international universities. Expect your peers to raise your game. We’ve built our own proprietary AI algorithms from the ground up. Every month we review the latest research papers and turn what works into how we build.Flying cars, not faster horsesWe don’t want 10% better. Anything bigger than a day of work goes through four questions first:
• Outcome. What result do we actually want?
• Assumptions. What are we taking for granted that might be wrong?
• The unreasonable version. What would this look like if it were 10x better, not 10% better?
• Proof. What evidence would convince someone else that it works?Then we aim for the unreasonable version and work back to what we can ship this week.Hunger beats tenureEveryone ships. Impact beats talk. Your title doesn’t get you a pass.We don’t count years. We count what you shipped.
• Your contribution decides your growth. Not your title, not how long you’ve been here.
• A junior who ships beats a senior who talks.
• Ownership comes fast if you earn it. Early-career engineers here own real features used by real customers.
• Seniority isn’t the problem. Seniority without hunger is.How we build
• No spec, no build. No evidence, no done. Every piece of work starts with a spec and ends with proof.
• AI-native pipeline. Spec → implementation tickets → AI coding → evals from acceptance criteria → end-to-end tests → production.
• Loop engineering. Agents test, eval and fix until green. You investigate failures and verify it solves the real problem. You own the judgment and the taste.
• Best tools, paid for. Claude Code is our default and every engineer gets it. Plus Codex, Grok, and more.
• Frontier and open models. Claude, GPT, Gemini, DeepSeek, Qwen, Kimi and GLM, on Vertex AI, Databricks, OpenRouter and Hugging Face.What you’ll do
• Build backend and full-stack features for Brain, Axsona and our agentic sustainability platform.
• Work directly with LLMs: retrieval, agents, tool use, MCP servers and evals.
• Turn specs into production code with AI coding agents, and prove it works with evals and tests.
• Integrate with real customer systems (banks, enterprises, data sources) and keep data clean, cited and auditable.
• Prototype in hours. Harden in days.
• Help write our AI engineering playbook: the skills, harnesses and workflows the whole company runs on.Where you start
• Junior engineers. Start with a well-scoped feature. Use feedback and code review to sharpen your work, then grow into bigger ownership. Ask questions early and explain how you know your code works.
• Mid-level engineers. Own a feature or integration from problem definition to release. Make the design trade-offs, define how success gets measured, and own quality and reliability.What we look forYears don’t matter. Evidence does.
• You build things. Show us. A repo, a product, a demo, an agent, a side project people actually use. Internships, open source and weekend projects all count.
• You build with AI, every day. Claude Code, Codex, Cursor or similar. This is a hard requirement. If you haven’t built anything with AI yet, this isn’t the role.
• You can code. At least one modern backend stack (Python/Django, Node.js or similar) and solid SQL.
• You care about quality. Clean code, tests, and the judgment to spot when AI output is wrong.
• You’re hungry. You ship, take feedback and ship again.
• You’re fine with ambiguity. Fast-moving startup, small team, lots of work with non-engineers.Bonus points
• You’ve built something with LLMs: agents, RAG, MCP, evals or prompt orchestration.
• Cloud (GCP, AWS, Azure) or data/ML infrastructure (Vertex AI, Databricks, SageMaker).
• Strong opinions on context engineering, harness engineering or agentic coding. Bring them.This is not for you if
• You want a manager to tell you exactly what to do every day.
• You think AI is a fad, or a shortcut that saves you from thinking.
• You care more about your title than what you ship.How we hireNo LeetCode. No whiteboard puzzles. We stopped doing coding interviews because they don’t tell us anything useful. We look at what you’ve built and how you use AI to build it. Expect to walk us through real work.What you get
• Build products that change how work gets done, on top of a climate platform banks rely on.
• Learn from a CEO and engineers who build alongside you, not managers who just review code, slides and Jira tickets.
• Real ownership early. Judged on contribution, not tenure.
• Weekly AI and engineering sharing sessions, monthly AI workshops, and the best AI tools on the market, paid for by us.
• Competitive salary and ESOP after probation. Everyone shares in what we build.
• Hybrid work: 3 days in the office, 2 from home. The office is beside Semantan MRT station.We celebrate diversity and are proud to be an equal opportunity employer. We do not discriminate on the basis of race, religion, colour, national origin, gender, age, marital status, sexual orientation, or disability status.If you’d rather build the future of work than talk about it, show us what you’ve built.Apply: jobs@pantas.com. Send your CV or a short intro, plus one thing you built. A repo, demo, screenshots or short write-up all work. Tell us what you contributed and how you used AI. Please don’t share confidential employer or client material.