Kovva

Own meaningful, end-to-end areas of the product — spanning the agent harness, evaluation systems, platform integrations, and user interface.

About Kovva

At Kovva, we're building AI Agents for Media Buying teams. Our clients are Independent Agencies, Holding Companies and Brands. We integrate with all the tools media buyers use — DSPs, social platforms, search, third-party creative ad servers, measurement and ad verification partners — as well as productivity tools like Slack, OneDrive, Zoom and Email.

Kovva automates the mundane work that consumes the media buyers' day, such as cross-platform pacing checks and discrepancy monitoring, campaign QA and troubleshooting. We know this world from the inside, and we're moving fast. This is the ground floor, with all the opportunity and responsibility that implies.

The Role

You will own meaningful, end-to-end areas of the product, spanning the agent harness, evaluation systems, platform integrations, and user interface. We are a lean team with minimal hierarchy, significant engineering autonomy, and a fast path from idea to production.

Our development workflow is agent-native: coding agents are embedded throughout exploration, implementation, testing, review, and debugging, while engineers remain accountable for the quality of what reaches production. You will also work directly with customers and observe media buyers using the systems you build.

What You'll Do

• Integrate a new advertising platform. Design and implement authentication, a robust API client, read and write tools with appropriate approval controls, a synchronization pipeline into Postgres, and a dedicated platform sub-agent. You will also develop model instructions that capture the platform's real-world API behavior, including nuances and constraints that may not be fully reflected in its documentation.
• Own an agent-powered product surface. Media-plan imports, campaign building and quality assurance, scheduled briefings, and Slack and email entry points are each distinct products with their own agent architecture, tools, evaluations, and failure modes. You may take ownership of one from design through production.
• Eliminate a recurring class of failure. When something is silently going wrong in production for a particular customer, you will reconstruct the agent's behavior using BigQuery event logs and traces, identify the underlying mechanism, and address the root cause.
• Extend the agent harness itself. Improve foundational capabilities such as tool gating, context compaction, large-payload handling, and the review system that determines whether the operator's work is ready to present to a user.

What We're Looking For

• 5+ years backend engineering experience. You have several years of production Python experience, or equivalent depth in another language that you can transfer quickly. You are comfortable with asynchronous systems, Postgres, schema design, background jobs, and the practical failure modes of distributed systems.
• Hands-on experience with LLM agent systems. You have built and shipped an agentic product involving tool calling, multi-step orchestration, context management, and structured outputs. You also treat prompts and instructions as versioned, testable engineering artifacts.
• Comfort with nondeterministic systems. You can diagnose systems in which the same input may produce different outputs. You use evaluations, replay, tracing, and telemetry to understand behavior.
• Enough frontend experience to complete the feature. You are comfortable with TypeScript and React and can ship the interface required for a feature without handing it off. Prior experience with Next.js and Tailwind is helpful but can be learned on the job.
• A strong sense of ownership. We are a small team responsible for a large product surface. You take responsibility for carrying work from an initial idea through implementation, validation, and production.

Preferred

• Experience in advertising technology or marketing technology, including DSPs, ad servers, attribution, or the structure of media plans. This domain knowledge is difficult to find, and we value it highly.
• Deep experience with Google Cloud, particularly Vertex AI, Cloud Run, BigQuery, and Pub/Sub.
• Experience building multi-tenant SaaS products with rigorous data-isolation requirements.

Our Stack Why Kovva

• You'll matter immediately — as an early hire, the systems you build become the foundation for everything that follows. This role has leverage that doesn't exist at larger companies.
• The product is genuinely novel — Kovva replaces ~70% of a media buyer's manual workload. No competitor does what we do cross-platform. You'll be building something you can actually believe in.
• Founding team with deep domain credibility — our co-founders built a DSP, ran trading teams, and led engineering and data science. We have deep agency relationships and domain expertise. We're solving a real problem we've experienced.
• Direct line to customers — you'll work directly with media buyers using the systems you build, and what you learn becomes product decisions. Fast loops, no bureaucracy, real impact.