Senior Software Engineer, AI Training - Argentina

G2i

Worldwide · remote (project dependent)Contract & freelanceSenior
Contribute human expertise to AI training, evaluation or data work. Check the official source for project requirements.

About the role

Contract | Remote, worldwide | $100-$200/hour depending on experience and location | 10-20 hrs/week

Please note that eligibility may vary depending on your location, as some regions may not be supported by the client.

Check out this Loom video for more details: https://www.loom.com/share/b0d1b0bf24c44ae8b95dca84b9db60e5

We're looking for highly experienced software engineer (SR+) to help evaluate the quality of interactions with modern coding agents such as OpenAI Codex and Claude Code.

This is not a traditional engineering role. You won't be writing production code. You'll be evaluating something harder: whether the model thinks like a great engineer.

WHAT THIS ROLE ACTUALLY IS

You will assess how AI coding agents behave in real-world scenarios, focusing on:

• Whether the response makes sense

• Whether the preamble and reasoning are useful

• Whether the output reflects strong engineering judgment

• Whether the interaction feels right to an experienced developer

This role is about engineering taste. Syntax correctness is the easy part.

WHAT YOU'LL BE DOING

• Evaluate AI-generated coding interactions end to end

• Judge whether outputs are useful, correct at a high level, and aligned with how a strong engineer would think

• Assess the quality of explanations and reasoning, not just the code

• Distinguish between levels of response quality (what makes something a 2 vs a 4)

• Give clear, opinionated feedback in writing on what worked, what didn't, and what felt off or misleading

• Help define what great looks like when working with tools like Cursor, Codex and Claude Code

WHAT WE MEAN BY "TASTE"

We're looking for engineers who can answer questions like:

• Does this feel like something a strong engineer would actually say?

• Is this explanation helpful, or just technically correct?

• Is the model guiding the user well, or just dumping output?

• Would this interaction build or erode trust? …

Official source: jobs.ashbyhq.com. Details can change; the platform's page is the final word. How we check listings.

Similar roles