Software Engineer, AI Application

Engg; United States; On-site; USD 140,000 - 180,000; Full time

Job summary

Plane is hiring for a product engineering role focused on enabling AI agents to participate in work lifecycle. You will read context from workspaces, perform actions inside them, and hand control back to humans at the right moment.

This is a hands-on engineering role centered on Python development and building tooling that agents depend on. You will design and ship tool surfaces, manage retrieval and embedding models, and own the end-to-end experience from data access to user-facing interfaces.

Qualifications

• Two or more years building production software.
• At least one year shipping features backed by large language models.
• Strong Python skills and production debugging experience.
• Experience building agent systems that call tools and manage loops.
• Working knowledge of embedding models, vector databases, and retrieval quality issues.
• Experience building an evaluation harness for non-deterministic systems.

Responsibilities

• Build agent workflows that operate on Plane data, including planning loops and multi-step execution.
• Design and ship MCP servers exposing Plane capabilities as tools with safe schemas and error handling.
• Own the end-to-end retrieval layer, including chunking, embeddings, vector storage, and hybrid search.
• Create agent interfaces with streaming output, visibility of steps, and gating for approvals.
• Establish evaluation harnesses with datasets and regression suites to assess changes.
• Instrument tracing, token cost, and latency across agent runs to reduce failures.
• Design a robust permission/audit model so agents only access permitted data.
• Evaluate models and tooling and advise Plane on adoption.
• Collaborate with backend, frontend, and product engineers to ship prototypes.

Skills

Python

LLM integration

Agent systems

Vector databases

Debugging under traffic

Tools

MCP server

Retrieval systems

Vector store

Job description

ABOUT PLANE

Plane's mission is to build the infrastructure the world's work runs on. Every organization runs on three things: the projects it's driving, the knowledge it keeps, and the requests it fields. Plane brings all three into one open, adaptable platform: simple enough for any team to adopt, dependable enough for organizations to build on. And we are building it for a future where humans and AI agents do that work together. Plane began in public on GitHub at the end of 2022. Since then it has grown into a work management platform used by teams around the world: 55,000+ stars, 5,000+ forks, and a contributor community that reads our code and files our issues. Organizations run Plane as a managed Cloud service, on their own infrastructure, or inside fully isolated environments. Building in the open keeps us close to users and raises the standard for everything we ship. Plane is the #1 work infrastructure in aerospace, defense, financial services, and other regulated industries: organizations whose requirements for control, auditability, and data residency rule most software out. When the strictest buyers pick a system of record, that choice means something. Adoption is growing fastest on Plane Cloud and in sovereign clouds, deployments that keep everything inside a country's own borders and rules. Plane is backed by top investors and built across San Francisco, London, and Hyderabad. We work in tightly knit teams, stay close to users, and care about the visible product as much as the unglamorous details that make software dependable. People own problems end to end, and we add process only when it helps the work.

HUMANS AND AGENTS

We believe the next decade of work will be done by humans and AI agents together. Not agents replacing people, and not a chat window bolted onto software built for humans clicking around, but both working in one system of action, where an agent's work is as visible and as accountable as a person's. Most software treats AI as a feature. We treat agents as a kind of worker, and that changes what the system underneath has to be. Agents are only as good as the context they can see and the state they are allowed to change, so shared context, explicit state, durable history, and accountable action are not items on our roadmap. They are the product. Plane is built so that when an agent acts, the humans responsible can see what happened, why, and on whose authority, and the record survives. This is what the infrastructure is for: making the future where humans and agents work together useful, legible, and fully within the organization's control. Every role at Plane is some part of building that.

ABOUT THE ROLE

Plane is where work gets planned, tracked, and shipped. You will build the layer that lets agents participate in that work directly, reading context from a workspace, taking action inside it, and handing control back to a human at the right moment. This is product engineering, not research. You will write Python, design the retrieval and tool surfaces that agents depend on, and own what you ship once real customers are using it.

WHAT YOU'LL DO

• Build agent workflows that operate on Plane data, including planning loops, tool calling, multi-step execution, and recovery when a step fails.
• Design and ship MCP servers that expose Plane capabilities as tools, defining the schemas, permissions, and error semantics that make them safe for a model to call.
• Own the retrieval layer end to end, covering chunking strategy, embedding models, vector storage, hybrid search, and reranking.
• Build the agent interfaces that customers actually touch, including streaming output, intermediate step visibility, approval gates, and clean interruption.
• Establish the evaluation harness for everything you ship, defining datasets, regression suites, and the offline and online metrics that decide whether a change goes out.
• Instrument tracing, token cost, and latency across agent runs, and drive down the failure modes that show up in production.
• Design the permission and audit model for agent actions so an agent can only reach what the acting user can reach.
• Evaluate models, frameworks, and techniques as they land, and make a clear recommendation on what Plane should adopt and what it should skip.
• Partner with backend, frontend, and product engineers to take a working prototype to a shipped surface.

WHAT YOU'LL BRING

• Two or more years building production software, including at least one year shipping features backed by large language models.
• Strong Python and experience operating services in production, including debugging them under real traffic.
• Experience building agent systems that call tools, covering the orchestration, the failure handling, and the loop control that keeps them from running away.
• Working knowledge of retrieval systems, including embedding models, chunking tradeoffs, vector databases, and the retrieval quality problems that follow.
• Experience building an evaluation harness for a non-deterministic system and using it to make a ship or no-ship call.
• Examples of AI features you shipped to real users rather than to demos or notebooks.

NICE TO HAVES

• You have built or published an MCP server.
• You have run agents against production data with a real permission model behind them.
• You have worked on developer tools, project management, or another product with a complex domain model.
• You have contributed to open source. Plane is open source, and you will see your work in public.

TECH

Python is the primary language for AI services at Plane. You will work across model APIs from major vendors, MCP for tool exposure, an agent orchestration layer, a vector store alongside Plane's primary database, and the tracing and evaluation tooling that sits around all of it. You will have real input into what that stack looks like as it grows.

WHY PLANE?

Every company says it is different. We will try evidence instead. The scope is real. We are a passion team s