A 4B coding agent, post-trained only with RL on synthetic software tasks, reaches 61.5% on SWE-bench Verified and beats models many times larger. No distillation from a frontier teacher. The missing ingredient is not a bigger model. It is tasks calibrated to what the current checkpoint can almost, but not quite, solve.

The pipeline estimates how often the agent already resolves a candidate task, then rewrites the ones that are too easy or too hard until they sit in a learnable band. As the policy improves, the curriculum moves: from heavily scaffolded bug fixes toward shorter statements and more diverse change types. A simple tool harness (read, write, edit, glob, bash) lifts the same 4B base far more than it lifts large models. Failures remain mostly reasoning, not tool use. Effective reward hacking is held at zero by the harness, not by hoping the agent is honest.

AI coding agents have fundamentally inverted the old trade-off: languages once chosen for human speed and ecosystem breadth (Python, TypeScript) are now being evaluated by what is easiest for agents to write, debug, and maintain at scale.

Strongly typed systems languages with fast compiler feedback loops (Rust, Go) have become dramatically easier for AI than traditionally “easy” languages. The barrier to high-performance code has collapsed.

AI tools make it dangerously easy to fix bugs while skipping the learning.

The default loop is simple: paste a spec or error message → model returns a fix → symptom vanishes → ship. Over thousands of these interactions, the messy struggle that once built deep mental models quietly disappears.