# Coding agent

Coding agents change code through tool calls, but a call carries no intent, and permission rules, logs and LLM monitors see only command text or after-the-fact effects. Episteme makes a type-plus-reason intent declaration mandatory for every tool call, with the harness recording actual effects and checking them against the declaration. On ImpossibleBench, whose tasks can be passed only by cheating, the journal exposed 44 of 49 misbehaving runs across two open-weight models, a 90% exposure rate, all through the agents' own stated reasons.

Coding agents change code through tool calls, but a call carries no intent, and permission rules, logs and LLM monitors see only command text or after-the-fact effects. Episteme makes a type-plus-reason intent declaration mandatory for every tool call, with the harness recording actual effects and checking them against the declaration. On ImpossibleBench, whose tasks can be passed only by cheating, the journal exposed 44 of 49 misbehaving runs across two open-weight models, a 90% exposure rate, all through the agents' own stated reasons.

The PROBE benchmark uses 2,457 bugs that maintainers fixed over 18 months in snapshots of 20 GitHub repositories as a reference, and asks whether coding agents can find those bugs autonomously across a whole repository. Across ten frontier models under a partitioned audit harness, the best configuration recovers only 20.8% of the reference bugs with precision no higher than 8.1%, yet an independent panel confirms 84% of findings that do not match the reference set as real bugs, showing that agents find a different distribution of bugs than maintainers fix.

The PROBE benchmark uses 2,457 bugs that maintainers fixed over 18 months in snapshots of 20 GitHub repositories as a reference, and asks whether coding agents can find those bugs autonomously across a whole repository. Across ten frontier models under a partitioned audit harness, the best configuration recovers only 20.8% of the reference bugs with precision no higher than 8.1%, yet an independent panel confirms 84% of findings that do not match the reference set as real bugs, showing that agents find a different distribution of bugs than maintainers fix.

JetBrains shipped Air in its 2026.3 IDEs, bringing coding agents into the developer's existing environment with one place to track session progress, review changes and send feedback. JetBrains frames the problem as a shift in bottleneck: as agents write more code, review becomes the constraint, and it cites a report finding that 49% of developers now spend four or more hours a week reviewing AI output and correcting prompts. Through the IDE's MCP server and built-in skills, agents can run inspections, build and run configurations, query databases and use the IDE's own refactorings, and in IntelliJ IDEA and CLion they can also drive the debugger.

JetBrains shipped Air in its 2026.3 IDEs, bringing coding agents into the developer's existing environment with one place to track session progress, review changes and send feedback. JetBrains frames the problem as a shift in bottleneck: as agents write more code, review becomes the constraint, and it cites a report finding that 49% of developers now spend four or more hours a week reviewing AI output and correcting prompts. Through the IDE's MCP server and built-in skills, agents can run inspections, build and run configurations, query databases and use the IDE's own refactorings, and in IntelliJ IDEA and CLion they can also drive the debugger.

Impeccable turns design direction into commands a coding agent can call, so vague requests such as bolder, quieter or polish land on a specific target. It also supplies a review command that scores design against usability heuristics and a slop detector built from 62 deterministic rules, with no LLM or API key, that flags interface tells recurring across models. It is strongest on existing pages and components; building new pages is the harder mode.

Impeccable turns design direction into commands a coding agent can call, so vague requests such as bolder, quieter or polish land on a specific target. It also supplies a review command that scores design against usability heuristics and a slop detector built from 62 deterministic rules, with no LLM or API key, that flags interface tells recurring across models. It is strongest on existing pages and components; building new pages is the harder mode.

Scientific foundation models usually ship with aggregate benchmark scores, while the performance limits that matter in use surface only after deployment through scattered reports. AutoSPECTRA constrains a general-purpose coding agent into a reproducible protocol: define a measurable generalization axis, run controlled comparisons against fixed reference methods, and confirm on data not used to select the axis. Across four biological foundation models, ESMFold2, STATE, CONCH and AlphaMissense, the audits tied performance shifts to protein sequence composition and relatedness to pretraining data, representation of perturbations in training, effective spatial resolution, and residue-level genomic, evolutionary and structural context, and rejected several initially plausible hypotheses.

Scientific foundation models usually ship with aggregate benchmark scores, while the performance limits that matter in use surface only after deployment through scattered reports. AutoSPECTRA constrains a general-purpose coding agent into a reproducible protocol: define a measurable generalization axis, run controlled comparisons against fixed reference methods, and confirm on data not used to select the axis. Across four biological foundation models, ESMFold2, STATE, CONCH and AlphaMissense, the audits tied performance shifts to protein sequence composition and relatedness to pretraining data, representation of perturbations in training, effective spatial resolution, and residue-level genomic, evolutionary and structural context, and rejected several initially plausible hypotheses.

Remote sensing models have long been built by hand, with experts designing a training recipe for each new task and tuning it against a benchmark. RSI4EO casts a task as two parts: a harness that fixes the data, pretrained weights and compute an agent may use, and a sealed verifier that returns only a score. A general-purpose coding agent then writes the training code, trains models and improves them recursively. On four benchmarks with published results, spanning optical, SAR and multispectral time-series imagery, the agent-built models surpass the state of the art, and with about 1% of the training labels they keep most of that accuracy.

Remote sensing models have long been built by hand, with experts designing a training recipe for each new task and tuning it against a benchmark. RSI4EO casts a task as two parts: a harness that fixes the data, pretrained weights and compute an agent may use, and a sealed verifier that returns only a score. A general-purpose coding agent then writes the training code, trains models and improves them recursively. On four benchmarks with published results, spanning optical, SAR and multispectral time-series imagery, the agent-built models surpass the state of the art, and with about 1% of the training labels they keep most of that accuracy.

Impeccable is a command set for coding agents that steers design with words such as polish and bolder, reviews interfaces with critique and audit, and flags templated design tells before shipping using 62 deterministic rules. With the same model, Fable 5.1, and the same brief for a fourteen-room family guesthouse, the page without Impeccable took the tasteful default of a cream page, thin serif and pill buttons, while with it the page was built around one picture of the house at night seen from the water. It is strongest on interfaces that already exist; building new pages is the harder mode.

Impeccable is a command set for coding agents that steers design with words such as polish and bolder, reviews interfaces with critique and audit, and flags templated design tells before shipping using 62 deterministic rules. With the same model, Fable 5.1, and the same brief for a fourteen-room family guesthouse, the page without Impeccable took the tasteful default of a cream page, thin serif and pill buttons, while with it the page was built around one picture of the house at night seen from the water. It is strongest on interfaces that already exist; building new pages is the harder mode.