Skip to content
Search jobs inside your AI agent. Claude, ChatGPT, Claude Code, Codex, Gemini, Cursor and Grok
Connect

Tenarai

Senior AI Engineer

Posted Updated

Description

Project info:


We are a small engineering team that brings agentic AI into end-to-end quality engineering for a large travel-technology platform. We build and own a shared agent harness, distributed as a plugin for Copilot CLI, Claude Code and VS Code. It includes:

  • skills and agent instructions
  • hooks and guardrails
  • MCP and CLI tools
  • a deterministic eval suite
  • a knowledge base for agents
  • browser exploration and recording tools

Several distributed QA and product teams use it every day to design, write, heal and maintain automated tests. We measure our work by engineering hours saved and by the quality of what gets merged, not by how much AI we use.



Responsibilities:


  • Design, build and maintain skills, instructions, hooks, MCP and CLI tools, and subagent definitions. These ship as one plugin to several agent clients.
  • Design tools for a model: names, descriptions, parameter shapes, output size and error messages, so that agents pick the right tool and use it correctly.
  • Keep the always-loaded context small and load procedures on demand. Find out why an agent stalled, looped or ignored an instruction, then fix the harness, not only the prompt.
  • Gate every harness change behind deterministic evals and CI. A quality regression is a failed build.
  • Run work in parallel across several agents and worktrees. Pick the model for each task, and make the cost of each workflow visible.
  • Sit with QA and product engineers, get their tacit knowledge out of them, and turn it into skills and decision records.
  • Show sceptics that agents work on their tasks, including manual and exploratory testing, with before/after numbers.


Job requirements:


Using agents every day is the baseline, not a qualification. Each item below is something we will ask you to explain or show from your own work during the interview.


1. You build for agents, not only with them
  • You have built agent extensions that other people use: skills or instructions, hooks, MCP servers or custom tools, subagents, or plugins. You can walk us through one of them, including its history: what broke, how users or models misused it, and what you changed.
  • You know at least one agentic coding harness (Claude Code, Copilot CLI, Codex, Cursor or similar) at the configuration and extension level. You can explain how it loads instructions, when hooks fire, how permissions work, and how it connects tools and MCP servers.
  • You can describe a tool or skill you redesigned because the model kept using it wrong, and explain why the new version worked.


2. You understand how an agent works inside

You can explain each of these with an example from your own work:

  • The agent loop. Model call, tool call, tool result, next turn. Where the loop stalls, repeats itself or stops too early, and what you did about it.
  • Context as a budget. What fills the context window (system prompt, instructions, tool schemas, tool outputs, history), what happens at compaction, and why always-loaded instructions should stay small.

Senior AI Engineer - Agentic EnablementPage 2 of 5

  • Prompt caching. What breaks the cached prefix, why stable ordering matters, and how caching changes cost.
  • Delegation. When a subagent saves money and when it multiplies cost, what an isolated context loses, and what a good brief contains.
  • Non-determinism. Why one green run proves little, and how you evaluate agent behaviour instead.


3. You prove quality and cost with numbers
  • You have built an eval or regression gate for prompts, skills or agent behaviour. You can show what it checks, a regression it caught, and where it stops being useful.
  • You can show what a real workflow costs (tokens, model mix, cache hits or a similar measure). You can also describe a change that lowered the cost and how you checked that quality did not drop.
  • You can give an example where a cheaper model was enough, and one where it wasn't.


4. You orchestrate parallel agent work
  • You can show work you split across several agents or worktrees: how you broke it down, the briefs you wrote, how you integrated the results, and the bad output you caught before it was merged. Having three chat tabs open does not count.


5. You understand both product and quality engineering
  • You can talk through how a product is built, released and maintained and how its quality is engineered. That covers a test strategy and CI pipeline you shaped, and failure triage or flakiness you fixed.
  • We don't need a "developer who does AI" or a "QA who does AI". We need someone who understands both sides deeply, whatever their mix of background.


6. You keep sensitive data safe
  • You can describe a guardrail you built or enforced that keeps personal, payment or customer data out of prompts, logs, reports and repositories, including data from test environments.


7. Enablement

You can describe specific situations from your own experience where you:

  • got working knowledge out of experts who didn't volunteer it, and turned it into artifacts that others reused;
  • changed a sceptic's mind with a working demo on their own task and measured results, rather than with slides;
  • challenged an inflated metric without alienating the person who reported it;
  • kept pushing a change through months of slow adoption without burning out, including how you paced yourself and when you escalated;
  • coached a remote team from another culture where people tend to defer to seniority, and made it safe for them to admit gaps;
  • made a practice stick by building it into tooling (hooks, CI checks, templates) rather than running a one-off training.
  • TypeScript / Node.js, Playwright, REST/XML/JSON API testing
  • GitHub (PRs, Actions, reusable workflows), Jenkins, Allure
  • Shipping one plugin to several agent clients; plugin marketplaces
  • Browser tooling: Chrome extensions (MV3), DevTools Protocol, traces/recording
  • Knowledge graphs or RAG-style memory for agents
  • Jira / Rally / Confluence APIs
  • Travel or airline domain knowledge (reservations, offers/orders, NDC)
  • Measuring developer productivity or AI ROI

Benefits:


General benefits - depends on the form of employment


  • Hybrid work model combining office & remote work
  • Attractively located office with collaboration spaces
  • Onsite parking space for employees
  • Referral program with financial bonus
  • Life Insurance
  • Budget for development (including language courses and others), clear career path with the possibility to gain experience in international environment
  • Access to internal Learning Platform with multiple trainings oriented for professional growth


Lifestyle benefits:


  • Access to MyBenefit platform (Multisport included)
  • Team Building activities
  • Charity initiatives
  • Working environment promoting diversity and inclusion


Health benefits:


  • Private medical care - Platinum Package

Skills

What Senior AI Engineering jobs ask for — and how much of it you have →
Apply

See also

AI Engineering jobs by country — openings, pay and top skills →