From experience: Junie, and AI agent based on Sonnet 4, performs quite well. It can even write tests and fix them if they are failing.
Not saying the quality is great, but good enough eventually work and to pass as junior code.
Not sure how good OpenAI agent is, and if they used their coding agent Codex, and if they did then was it as-is or with some tuning? Not sure, they write it was “custom agent based on o3”.
They write all,the contestants have the same hardware, but did the agent run on the given machine, or in the cloud? Human brain is like 20-40W, so let’s say the upper limit given he has to move his hands - did the AI agent get the same wattage? I don’t think so.
From experience: Junie, and AI agent based on Sonnet 4, performs quite well. It can even write tests and fix them if they are failing.
Not saying the quality is great, but good enough eventually work and to pass as junior code.
Not sure how good OpenAI agent is, and if they used their coding agent Codex, and if they did then was it as-is or with some tuning? Not sure, they write it was “custom agent based on o3”.
They write all,the contestants have the same hardware, but did the agent run on the given machine, or in the cloud? Human brain is like 20-40W, so let’s say the upper limit given he has to move his hands - did the AI agent get the same wattage? I don’t think so.