Software Engineer - Agent Harness - Instabase

At Instabase, we're passionate about democratizing access to cutting-edge AI innovation to enable any organization to solve previously unsolvable unstructured data problems in their industry.  With customers representing some of the largest and most complex organizations in the world, and investors like Greylock, Andreessen Horowitz, and Index Ventures, our market opportunity is near infinite.

Instabase offers a consumption-based pricing model where customers can pay only for what they use, aligning directly with the value our products deliver. It empowers our clients to explore our AI Hub platform features extensively, enabling them to uncover crucial business insights. This customer-centric model allows Instabase to glean insights from diverse use cases and behaviors, ensuring we deliver top-tier solutions that provide unmatched advantages for everyday business operations.

With offices in San Francisco and Bengaluru, Instabase is a truly global company. We are people-first , and we've built a fearlessly experimental, endlessly curious, customer obsessed team who work together and help organizations around the world turn their unstructured data into insights instantly.

Role Overview

We are seeking a visionary systems-minded AI engineer to lead the design, architecture, and development of the Agent Harness . In this role, you will bridge the gap between advanced foundation models (LLMs) and reliable, production-grade execution.

You will not just be "vibe coding" prompts ; you will build the actual state machine, execution sandbox, tool-calling middleware, and security containment layers that allow AI agents to safely execute code, search the web, manipulate files, and query databases in real-time. This is a high-leverage role requiring deep expertise in distributed backend systems, runtime isolation, and stateful agent choreography.

What You’ll Do (Responsibilities)

• Harness Architecture & State Management: Architect the next-generation execution runtime (Agent Harness) that drives SuperApp’s planning, reasoning, memory retention, and tool-execution loops. Build reliable state machines capable of pausing, resuming, and versioning agent trajectories.
• Secure Sandbox Execution: Design and scale highly secure, isolated, and ephemeral environments (using Docker, gVisor, WebAssembly, or microVMs) to execute agent-generated code safely without putting host infrastructure at risk.
• Tool & Integration Frameworks: Build and maintain high-throughput APIs and integration layers (such as the Model Context Protocol - MCP) that connect SuperApp to external services, databases, web search utilities, and complex file processors.