Software Engineer: AI Harnesses & Tooling

Big Wombat LLC; United States; Remote; USD 215,000 - 325,000; Full time

Benefits offered by this job

Equity

Remote work

Founding-team role

Job summary

Yaklabs is building a desktop AI application that turns knowledge workers into AI power users, without docs, API keys, or knowing what to ask for. One surface for agentic AI work: proactive, personalized, progressively powerful.

The team is seeking a Software Engineer to design the agent harness, runtime, and extension system, owning the trust boundary and the inference stack. This is a founding-team role with impact on product direction and culture.

Qualifications

• You understand agent runtimes underneath: tool-calling loops, context windows, partial failures in multi-step tasks.
• You have opinions about what’s wrong with most agent frameworks.
• You think in compounding loops. You’ve designed systems that get better through use.
• Strong systems instincts. You think about failure modes, resource limits, isolation, and concurrency.
• You’ve worked with LLM APIs at the protocol level: streaming, tool use, structured outputs.

Responsibilities

• Design and build the agent runtime: dispatch, execution lifecycle, tool calling, error handling, retries, cancellation.
• Build the extension system, where FDEs, users, and agents themselves customize the harness.
• Own the trust boundary: sandboxing, permissions, credentials, and which actions an agent takes alone vs. approved.
• Build the inference layer: provider abstraction, streaming, and cost tracking.
• Design the context system: how context is assembled, compressed, and managed across long sessions.
• Own observability. Every agent action should be a structured event you can query, export, and audit.

Job description

Software Engineer: AI Harnesses & Tooling

Engineering, Remote (US or EU timezones), Full-time

You build the harness: the system that turns an LLM into a reliable agent that people, FDEs, and other agents can extend. A good harness is worth as much as a good model, and most of the ones out there aren’t good yet.

Problems you’ll work on

These are examples, not lanes. You will move between them, while taking real ownership of the ones that match your interests.

• Accountable memory: How can an agent build a durable model of a user’s world without accumulating stale facts or confident fiction? Give what it learns provenance, confidence, and a lifecycle as the underlying world changes.
• Multi-agent collaboration: How can several agents work toward one outcome without losing authority, context, identity, or completed work? Delegation must survive concurrent execution, partial failure, retries, and process restarts.
• Context and retrieval on the user’s machine: A person’s documents, messages, and calendar are spread across a dozen systems, and most of it should never leave their device. How do we index that so an agent pulls the right context out of a large and messy corpus, keep the index fresh as their world changes, and keep it fast on a laptop?
• Routing work across models: Every provider exposes a different mix of capabilities, prices, limits, and failure modes. How should the harness match each task to a configuration, preserve continuity through a long-running session, and fail over when a provider stalls or runs out of quota?
• Trust and permissions: What should an agent be allowed to do on its own, what needs a person to say yes, and how does a company prove afterwards what it did?

What you’ll do

• Design and build the agent runtime: dispatch, execution lifecycle, tool calling, error handling, retries, cancellation.
• Build the extension system, where FDEs, users, and agents themselves customize the harness to hill-climb against real problems. It’s a microkernel question: what’s in, what’s out, what’s exposed.
• Own the trust boundary: sandboxing, permissions, credentials, and which actions an agent takes alone, which need approval, and which are forbidden. Every action is safe by default, auditable afterwards, and overridable by the user.
• Build the inference layer: provider abstraction, streaming, and cost tracking, with many providers behind one interface.
• Design the context system: how context is assembled, compressed, and managed across long sessions.
• Own observability. Every agent action should be a structured event you can query, export, and audit.

Who you are

• You understand agent runtimes underneath: tool-calling loops, context windows, partial failures in multi-step tasks. You have opinions about what’s wrong with most agent frameworks.
• You think in compounding loops. You’ve designed systems that get better through use, and you know the harness should adapt around the person using it.
• Strong systems instincts. You think about failure modes, resource limits, isolation, and concurrency before you think about features.
• You’ve worked with LLM APIs at the protocol level: streaming, tool use, structured outputs. You know where the sharp edges are.
• You build abstractions that don’t leak. You’ve swapped the implementation under a system and its consumers didn’t notice.

Bonus

• You’ve built or contributed to agent frameworks or harnesses.
• You’ve worked with sandboxed execution environments.
• Systems-language experience (Rust, Go) alongside TypeScript.

About Yaklabs

Yaklabs is building a desktop AI application that turns knowledge workers into AI power users, without docs, API keys, or knowing what to ask for. One surface for agentic AI work: proactive, personalized, progressively powerful. The gap between AI power users and everyone else isn’t talent, it’s setup. We close it.

The team: Alex Goddijn spent four years at Palantir, then founded an agent harness company that was acquired by Poolside AI. Seb Goddijn led internal AI at Ramp. We’re building Bonsai so that kind of leverage isn’t limited to companies with their own internal AI team.

We have funding. Product and design work in person in New York City, and some engineering roles are remote. Small team, high intensity, no passengers.

The deal

• Early team. $215k to $325k base salary for US-based hires, plus meaningful equity.
• In person in New York City for product and design roles. Some engineering roles are remote, and each posting lists its location.
• You’ll shape the product, the culture, and the company. This is a founding-team role.