Loop / Harness Engineer. The role, the market signal, and how to build it in your org.
The 2026 successor to the Prompt Engineer. Instead of prompting an agent step by step, designs the loop and harness the agent runs inside: the scaffolding of tools, memory, verification, and the act-observe-decide-repeat cycle that keeps an agent productive, on-goal, and verified while it runs unattended for hours. When a single agent run touches dozens of files, the leverage moves from writing a sharper prompt to designing a better loop. A skill in formation, named by practitioners at Anthropic, Google, and the frontier labs ("beyond prompt engineering", "thin harness, fat skills"), not yet a job-board title. The signal lives in commentary and practitioner writing, not postings.
In a workforce of people and AI agents, this role takes shape around the workflows it supports. Define its responsibilities and handoffs, then prepare people to perform that work.
4 postings · 4 distinct titles · from 264,613 real job postings · see the live data →
What the postings ask this role to do
38 tasks extracted from real Loop / Harness Engineer job descriptions, classified Automate / Augment / Human-only. Only 2.6% can be fully automated: companies are hiring this role for the judgment, not the keystrokes.
- Follow defined scope, guidelines, and deadlines.
- Design agent orchestration.
- Develop tools for agent interactions.
- Launch new models on cursor end-to-end.
- Tune model behavior for agent performance.
- Perform tests and issue root cause analysis.
- Mentor other team members.
- Shape the default cursor agent's personality.
- Ensure evals are the shared bridge between both teams
- Ensure that as the harness evolves the model evolves with it
- Own the feedback loop between north's harness and the modeling team
From the market's version of this role to your version of it
Start with the work, not the org chart.
Define what this role will own, then connect those responsibilities to practice and assessment.