About the role
The agents team makes models that use dense toolsets, recover from mistakes and finish the job. An agent with fifty tools, partial information and a long task has to choose, call, read results, notice when something went wrong and try again without looping or quietly giving up. The team works on the model behaviour and the scaffolding around it.
Clexerl Go is built on the premise that agents do the coordination and people keep the judgement. This team's research is its engine: tool selection, planning, permissioning and handoff to a human at the right moment. Findings go straight back to Post-Training and Reinforcement Learning as training tasks, and to Evaluations as new agentic suites.
Work here is grounded in real traces. The team reads failed runs from internal use, categorises why they failed, and fixes the class, whether that means a better tool schema, a training environment or a change in how context is managed across a long task.
What you’ll do
- Design agent architectures covering planning, tool selection, state management and error recovery for tasks running hundreds of steps.
- Define tool interfaces and schemas that models use reliably, and study how description wording, argument types and error messages affect success.
- Build context-management strategies, such as compaction, scratchpads and sub-agent delegation, that preserve task state over long runs.
- Analyse failed trajectories at scale, cluster the causes and feed them into training environments and evaluations.
- Implement permissioning and confirmation flows so agents ask a person before taking irreversible or high-impact actions.
- Prototype end-to-end agent behaviours inside Clexerl Go with product engineers and measure them against real workflows.
What you’ll bring
- Strong software engineering background in Python or TypeScript, with experience building systems that call external services and handle failure.
- Hands-on experience building LLM agents, including function calling, structured outputs and multi-step planning.
- Understanding of how models fail on long tasks: compounding errors, forgetting earlier instructions, tool-call hallucination and loops.
- Skill in building evaluations from real traces and turning findings into concrete improvements.
- Awareness of security issues specific to agents, such as prompt injection and over-broad permissions.
- Comfort moving between research questions and product constraints without losing rigour.
Nice to have
- Experience with the Model Context Protocol or other tool-integration standards.
- Research on multi-agent coordination and delegation.
- Background in browser automation, GUI agents or robotic process automation.
- Experience with workflow engines, durable execution or transactional rollback for side-effecting actions.
Work you might take on
- 01
Build a tool-routing layer that picks from hundreds of tools and measure the accuracy gain over putting every schema in the prompt.
- 02
Create a prompt-injection test suite for agents reading email and documents, and train against the failures it finds.
- 03
Implement a checkpoint-and-resume mechanism so an agent can recover from a mid-task error without repeating completed side effects.
- 04
Design a sub-agent delegation scheme and compare it with single-agent baselines on long scheduling and CRM tasks.
Compensation
$300K – $500K
The US base salary range for this role. Every offer also includes equity in Clexerl and, when it spins out, in the venture you help build. We tell you the reasoning behind your number.
Benefits
Health, fully covered
Medical, dental and vision for you and your dependants, premiums paid in full.
Two kinds of equity
Equity in Clexerl, plus a stake in the venture you help build when it spins out.
22 weeks parental leave
Fully paid, for every parent, however your family arrives.
Rest that sticks
Flexible time off with a four-week minimum we actually check on, plus a company-wide week in December.
Learning and compute
A yearly budget for courses, conferences and books, and compute credits for side projects.
401(k) with a 4% match
Matched from your first paycheque and vested immediately.
How we hire
Apply 5 minutes
Send links to what you have built and a few lines on why this role. No cover letter needed.
Intro call 30 minutes
A conversation with the hiring manager about the work, the team and what you want next.
Work sample 2–4 hours
A paid take-home or a pairing session on a real-shaped problem. You pick the format.
Studio day One day
Meet the team, go deep on your work and ours. Remote or in person.
Offer Within 5 days
A written offer with the full range, the equity and the reasoning behind both.
Opens an email with a short template. Links beat résumés.