We can't find the internet
Attempting to reconnect
Something went wrong!
Hang in there while we get back on track
Coverage is incomplete: some monitored sources or documents could not be processed. This brief covers the available verified material.
Coverage is incomplete: some monitored sources or documents could not be processed. This brief covers the available verified material.
🔥 TOP SIGNAL
The agent bottleneck is moving downstream—from generating code to controlling the artifact and the environment around it. Simon Willison released commit-rewriter 0.1 after coding-agent cruft and private-repository issue IDs made commit messages unfit for publication; the Python web app provides an editor and creates a timestamped branch before rewriting. Peter Steinberger is attacking the other side of the problem with folder-clone worktrees he says should be ~80% faster and use less disk, while Theo’s sponsored Codesmith demo used actual CI logs to right-size runners.
The practical pattern is to make agent work cheap to isolate, inspect, and undo—not just cheap to generate.
⚡ TRY THIS
Clean agent-authored history before publication. Run
uvx commit-rewriter path/to/repo—or omit the path from inside the repository. Edit the messages in its web UI, then submit: it creates a timestamped branch for rollback and rewrites every commit from the first edited commit through the latest.Route subagents to the model they actually need. Lydia Hallie’s Fable 5.1 tip is
CLAUDE_CODE_SUBAGENT_MODEL=opus: Fable spawns subagents, but hers rarely need Fable-level reasoning, and she says this saved “so much usage.” Addy Osmani amplified the tip. Treat it as a field anecdote, not a benchmark.Make CI tuning log-driven. In a Blacksmith/Codesmith setup, ask the agent to inspect real CI logs, identify CPU-heavy and lightly used jobs, recommend runner sizes, and apply the one-line GitHub Actions change. Theo reports three real T3 Code PRs merged in under five minutes and almost halving CI time; because the segment is explicitly sponsored, reproduce the before/after in your own pipeline.
Put a size and verification gate around generated code. A Vienna discussion summarized by @pidotdev warned that agents generating too much code poison their own context (“slippery slop”) and called out unreadable Astra code. Theo separately reports Fable putting a screenshot of a bug it introduced into a PR without noticing the bug. Keep diffs bounded, then require a fresh test or rendered-output review; a screenshot is evidence to inspect, not acceptance.
📡 WHAT SHIPPED
commit-rewriter 0.1— Simon Willison. A Python web app born from preparing Datasette security-release commits whose agent cruft and private issue references were not fit for publication; it preserves a rollback point before rewriting history.Faster worktrees are headed for a release or dev channel—not stable yet. Peter Steinberger says folder clones across APFS,
brtfs, XFS, and ReFS should make worktrees ~80% faster and save disk space. The implementation is written in Rust; he will test it for several weeks before trying to get it into Codex.T3 Code now streams responses by default. Theo says the behavior is not token-by-token and will be toggleable before stable. He also says he is concerned T3 Code burns through Codex and Claude Code limits quickly enough to risk bans, then jokes that “the real rate limit was the quality of the client.”
🎬 GO DEEPER
- Theo — Codesmith/Blacksmith CI-agent demo. Watch the opening sponsor segment for the concrete loop: inspect actual CI logs → right-size runners → make a one-line GitHub Actions change → validate the resulting PRs. Theo reports three real T3 Code PRs in under five minutes and almost-halved CI; keep the sponsor context in mind.
Study the worktree substrate. Peter points to the Rust
fs-safeimplementation behind the folder-clone experiment; it is a future performance test, not yet a settled Codex feature.Romain Huet — Astra’s artifact-feedback loop. Huet asked Astra in Codex to build a Blender model and fly-by, then noted a ~30-minute first model/orbit, checks in daylight and dusk, and a follow-up request that built the interior, including 88 individually modeled window panes. The reusable pattern is inspectable artifact → environmental checks → concrete follow-up, not the nostalgic subject matter.
Editorial take: The alpha is not more autonomous generation; it is a reversible loop—usage-aware subagents, clone-backed worktrees, log-based feedback, and independent acceptance.
- Codesmith (Blacksmith): In a sponsored demo, Theo used Codesmith on T3 Code’s GitHub Actions. The agent inspected real CI logs, identified jobs with high or low CPU usage, and recommended right-sizing runners; moving the workflow to Blacksmith required changing one line in the existing GitHub Action configuration.
- Theo reports that he merged three real PRs filed by Codesmith during the demo; the process took under five minutes and nearly halved T3 Code’s already-fast CI time. He also claims Blacksmith delivered more than 50% performance improvement on its own and 60%+ total savings versus GitHub Actions when speed and price were combined.
-
Simon Willison released
commit-rewriter0.1, a Python web app he built while preparing Datasette security-release commits that contained coding-agent cruft and private-repository issue references unsuitable for publication. -
Replicable workflow: run
uvx commit-rewriter path/to/repo(or omit the path when already inside the repository), edit commit messages in the web interface, then submit the changes; the tool creates a timestamped branch of the current repository state for rollback and rewrites commits from the first edited commit through the latest one.
- Peter Steinberger says the next release or dev channel will make worktrees approximately 80% faster through folder clones on APFS, BRTFS, XFS, and ReFS, while also saving substantial disk space.
- The filesystem-safe implementation is written in Rust and available at https://fs-safe.io/copy.html. Steinberger plans to test it for several weeks and, if it helps most users, try to integrate it into Codex.
- Subagent model routing: Lydia Hallie reports that setting subagents to Opus on Fable 5.1 saved substantial usage because Fable frequently spawns subagents that rarely need Fable-level reasoning. The configuration is
CLAUDE_CODE_SUBAGENT_MODEL=opus. Addy Osmani amplified this as a practical tip, so this is a firsthand anecdote from Lydia rather than Addy’s own benchmark.
John Lindquist shares three personal apps he uses daily: Image Studio, which manages image-generation work across Codex, Grok, and Gemini; GM, a Google Messages rewrite built in GPUI with agentic and convenience features; and Eggo, a 24/7 real-time voice desktop agent/avatar. Kent C. Dodds says his upcoming video with @kodykoala will explain how he improves the process of building personal software for daily problems and how developers can adapt the approach to their own tools.
Theo warned that T3 Code may drive users to burn through their Codex and Claude Code usage limits quickly enough that he is concerned about potential bans. In a follow-up, he wrote that “the real rate limit was the quality of the client.”
@steipete reports a practical Linux troubleshooting workflow: on a Dell XPS with a non-working webcam, he told Codex about the problem; Codex rebooted the machine, and the webcam then worked. This is a firsthand example of using a coding agent for system-level debugging beyond source code.
Peter Steinberger reports that after fixing numerous small performance issues that mattered little for single-user use, the system was made suitable for teams; it is currently handling roughly 80 sessions without apparent strain.
- Firsthand Astra-in-Codex workflow: Romain Huet asked Astra in Codex to reimagine Peach’s Castle in 3D and create a fly-by video. The initial Blender model and orbit took about 30 minutes, and Astra checked its work in daylight and at dusk. Huet then requested an interior view; Astra built the ground floor and guided him through the doors, including 88 individually modeled glass panes.
- Replicable pattern: start with an external 3D scene and camera movement, have the agent inspect the result under multiple lighting conditions, then iterate through a concrete follow-up request for interior geometry and a walkthrough. This was presented as a small experiment rather than a production workflow.
@pidotdev’s summary of a Vienna discussion featuring Armin Ronacher (@mitsuhiko) and @badlogicgames warned that agents generating too much code can poison their own context—“slippery slop.” The practical takeaway is to keep agent-generated changes bounded and context focused, especially when code must preserve tight control; the discussion also emphasized that “code matters wherever control matters.”
DHH accepted Grok’s offer to work on Omarchy and asked whether Grok could pledge a discretionary token budget. Grok’s proposed assistance includes auditing Omarchy code and configurations for vulnerabilities, hardening agents and packages, crafting plugins and skills, debugging crashes, refining documentation, and tightening an existing Grok CLI/Bot integration. This is a proposed collaboration rather than a reported hands-on workflow; the posts provide no execution steps, benchmark, or outcome.
Riley Brown describes a firsthand remote-agent workflow: using an iPad and the “GPT remote” to use Codex while fully controlling a Mac that runs at home 24/7. He plans to publish a 90-minute video covering the complete setup, but this post does not yet provide configuration steps.
Theo (@theo) reports a Fable failure mode: although he likes the tool, it occasionally introduces a bug, includes a screenshot of the resulting issue in a pull request, and fails to notice the bug itself. Treat Fable-generated changes and PR evidence as requiring independent human review.
Theo reports that T3 Code now streams responses by default, using non-token-by-token streaming. He says the behavior will be toggleable before the feature reaches stable.
- DHH reports firsthand that an XPS14 running the bespoke Omarchy Quattro RS kernel handled Linux kernel compilation, two 4K streams, one 1080p stream, Spotify, and an agent terminal concurrently without stuttering. In a follow-up, he added Tetris Effects and said the system remained smooth.
