Senior Software Engineer: Agentic Coding (AI Training)

What you'll do

You work on agentic coding systems (the setups where an AI plans, writes, tests and iterates on code with little human intervention), and your job is to make them better by finding where they fail.

That means reviewing AI-generated code for correctness, efficiency, security and adherence to good practice; writing hard, realistic coding challenges designed to break the agent; and giving expert feedback on its reasoning chains and tool use. You also help build the evaluation frameworks, rubrics and testing harnesses that score the systems.

The core of the role is the failure hunt: identifying the edge cases where an autonomous agent goes wrong without any obvious crash. Work is task-based and asynchronous, at 10–40 hours a week you set yourself.

What you need

• Five or more years of professional software engineering experience, having shipped and maintained real systems
• Proficiency in one or more of Python, JavaScript/TypeScript, Java, C++, Go or Rust
• A solid grasp of software architecture, design patterns and system design
• Comfort reading, reviewing and debugging code you did not write, across unfamiliar styles and codebases
• Version control, CI/CD, testing frameworks and code review as everyday habits
• The ability to articulate precisely why a piece of code is good, bad or broken
• Genuine curiosity about LLMs and autonomous software development

Nice to have

• AI/ML systems, prompt engineering, or LLM-based tooling experience
• Familiarity with agentic frameworks such as LangChain, AutoGPT, CrewAI, or function calling
• A background in developer tools, compilers, static analysis or code generation
• Prior model evaluation or training work: RLHF, preference labelling, or similar
• Open-source contributions or a strong public engineering portfolio

What it pays

$80–120/hour, stated on the listing. Against the stated 10–40 hours a week, that is $800 to $1,200 for a light week and $3,200 to $4,800 for a full one.

The 10-hour weekly floor is realistic to run alongside a full-time job.

Worth knowing

Good:

• The rate is published up front, which most contract AI work is not
• A 10-hour floor works alongside a day job
• Task-based and self-paced, so there are no shifts to hold
• Reviewing agent output is a fast way to learn where current coding models actually break, which is useful well beyond this contract
• No prior AI-evaluation experience is required; senior engineering judgement is the bar

Less good:

• The band is wide. $80 and $120 an hour are set at contract, not by the ad
• Task-based work means income tracks available task volume, which is not guaranteed between projects
• Reviewing machine-written code for hours is repetitive, whatever the subject matter
• Hourly contract: no benefits, no notice period, your own tax to manage
• The scope is the least specific of the engineering contracts here. Expect assignments to be defined project by project

About this listing

Posted by Alignerr on its own jobs board under the Coding category, confirmed open on 6 September 2026. The $80–120/hour rate is the ad's own published figure. No closing date and no country restriction are stated, so both are left open. Onboarding and payment run at the platform level. See Alignerr.