Explore what's new in AI

Zed 1.23 adds JSONL and NDJSON table previews

The update makes line-delimited JSON easier to inspect inside the editor, reducing the friction of debugging logs and data exports without extra tooling.

Trq212 recommends asking agents to teach unfamiliar domains first

When users cannot specify requirements precisely, repeated prompt iteration becomes a common failure mode; teaching the domain before planning can improve task framing.

Linear Loops generates QA plans when issues and pull requests reach review

Teams can automatically collect relevant issues and PRs, identify testing needs, and share a QA plan when a release enters review.

Factory lets Jira teams assign coding tasks to its Droid agent

Droid can take ticket context, write code, open a pull request, and report progress back to the issue, bringing agent work into an existing team workflow.

Blume 2.2 adds agent-assisted migrations for documentation sites

The CLI can move existing documentation projects into Blume with Codex, while new manual guides help teams handle migration details themselves.

OpenCode uses agents to triage issues and open pull requests

OpenCode is applying agents to issue responses and closure workflows, a concrete example of maintainers automating repository operations rather than simply disabling intake.

A study finds 124 unique tool jobs across 11 coding-agent harnesses

The analysis maps 380 tool entries to 124 distinct jobs and finds at least 65% overlap among most harnesses, suggesting substantial room to share tool implementations.

Codex adds a /visualize command to its app

The command gives developers a way to create visual representations from a coding session, extending the agent workflow beyond text and code review.

Sam Newman argues LLMs lack causal models for reliable operations

The distributed-systems author warns that fluent models do not understand action consequences, so deterministic controls remain important for protecting production systems.