Build an assessment that measures how well someone can orchestrate AI agents to finish a real task, and prove it works with a live agentic workflow.
Kapsule is figuring out how teams should actually use AI agents to get operational and technical work done. Right now, there is no clean way to tell whether someone is good at running agents. We want to fix that.
We need you to design a practical assessment that measures agent-management skill. Not just "do you know what Claude is," but: can you break a problem into agent-sized pieces, write prompts that hold up, spot when an agent is stuck or producing junk, and coordinate several agents running in parallel toward one useful outcome.
Start by getting hands-on with the current landscape — OpenClaw, Manus, Perplexity, Claude, ChatGPT, and whatever else is worth including. Compare what each is good and bad at. Then design the assessment itself: sample tasks with a clear objective, a scoring rubric, workflow documentation, and side-by-side examples of strong vs. weak agent coordination.
Deliverable: an assessment framework we can actually run on candidates, plus a live demo of an agentic workflow you built using it. Show us the prompts, the hand-offs between agents, where you intervened, and the final output.
You'll get access to our internal notes on how we currently evaluate hires, a stipend for paid agent tools during the internship, and weekly review calls with David to pressure-test your rubric before you finalize it.