AI Agent Engineer

Job Description

What You Will Do

• Ship with agents. Use Claude Code, Codex, Cursor, Copilot, or other tools across planning, implementation, testing, debugging, refactoring, documentation, and review. Deliver real features in production with the quality bar you would hold for hand-written code.
• Own the output. Treat what the agent writes as untrusted until you have reviewed it: watch for hallucinations, wrong assumptions, hardcoded secrets, missing validation, injection-prone patterns. Run it, confirm it works, iterate. For authentication, authorization, cryptography, and session handling, use our standard implementation rather than the model's invention.
• Fix the context, not the prompt. Keep CLAUDE.md, coding standards, architecture notes, and module dependencies current so the agent gets it right the first time. Feed SAST and dependency findings back with call path and data flow, so it writes a real patch instead of a suppression.
• Build team-level tooling. Develop prompts, skills, sub-agents, and MCP (Model Context Protocol) integrations for recurring work, plus guardrails that keep agents in bounds: architectural limits, forbidden patterns, off-limits folders. Keep the context lean and watch token cost.
• Use agents across the whole lifecycle. Integration tests and coverage gap analysis, regression tests generated from real production incidents, root-cause analysis with logs and traces pulled in, ADRs (Architecture Decision Records), and runbooks drafted with AI and checked by you.
• Review AI-assisted work well. Run context-aware AI review on your own diffs, separate true positives from false positives, and turn recurring false positives or escaped bugs into rules the team keeps.
• Share what works. Show your setup to the team, review other people's AI-assisted pull requests, and get at least one thing you built adopted beyond your own repo.

What You Will Need

Must Have

• Daily use of coding agents on real work. Not a course, not a demo. Code with users, where the agent did a large share of the writing and you owned the result.
• At least one project you can walk us through end-to-end with these tools: what you delegated, what you kept, where it failed you, and how you caught it.
• Agentic experience: tool and function calling, multi-step workflows, state, error recovery, human-in-the-loop.
• Context discipline. Show how you set an agent up to succeed — project files, standards, examples, scoped tasks — and how that cut your rework.
• Solid engineering fundamentals. Clean code, testing, error handling, security, performance. Agents make weak fundamentals more expensive, not less.
• The judgment to know when the agent is wrong. Speed is easy to fake. Judgment is not, and it is what we will test.
• Willingness to share it. You do not need to have run a training program. You do need to be the kind of person who shows a teammate the setup instead of keeping it.

Language is not a filter. Go and PHP/Laravel are preferred because that is what we run; Python, TypeScript, or anything else is fine if the rest is there.

Nice to Have

• Team-level tooling you built and other people used: shared skills, sub-agents, MCP servers, slash commands, or a review or test agent running in CI.
• Security habits around agents: forbidden patterns, least privilege on the tools and credentials an agent can reach, secrets kept out of AI context, AI-suggested packages verified before they are merged.
• Test and debugging depth with AI: coverage gap analysis, regression tests written from production incidents, root-cause work using observability data.
• LLM work: prompt engineering, agent SDKs, or orchestration frameworks such as Strands Agents, LangChain, or LlamaIndex.
• Agentic patterns beyond the basics: RAG, MCP, skills, hooks, plugins.
• Experience designing scalable, highly available systems, ideally in Go.
• Curiosity about how these systems actually work under the hood: benchmarking them, breaking them, improving them.

On Seniority

We are open on seniority. Mid-level and Senior-level candidates are both evaluated, and your level comes from what you have shipped, not from the number on your CV. The one thing we do not flex on: agents are already how you work, not something you are planning to try.