Codex Ultrafast cuts triage to 15 min, Webframe sheds Chromium for WebView
Every URL the pipeline pulled into ranking for this issue — primary sources plus the supporting and contradicting findings each Researcher returned. Inline citations in the issue point back here.
Sources
How Ultrafast cut incident investigation from hours to minutes at OpenAI openai.com
See how OpenAI engineers use Codex Ultrafast to investigate incidents in minutes. Explore faster security analysis, code refactoring, and daily workflows.
webframe: evaluated as a scraper and a sandbox simonwillison.net
Research: webframe: evaluated as a scraper and a sandbox webframe is a neat little project from Jeff Lindsay which lets you “run a web page like a program” - you point it at a URL or file and it executes it in a hidden headless web view and outputs console.log() to standard output and any exceptions to standard error. It’s written in Go on top of the glaze library, which provides access to a system browser (“WKWebView on macOS, WebKitGTK on Linux, WebView2 on Windows”) so it doesn’t need to bun…
References
MindStudio analysis of Codex Ultrafast pricing mindstudio.ai
one complex task burned 17% of a weekly quota on Ultrafast compared to only 4% on standard mode
AgentHanks incident-response case study (Datadog + PagerDuty) agenthanks.com
MTTR… dropping from 47 minutes to just 17 minutes… time to the first diagnostic action from 18 minutes down to three
Vals.ai SWE-bench / SRE leaderboard vals.ai
While OpenAI reports a SWE-bench Verified score of 85% for GPT-5.3-Codex, independent SRE leaderboards for root cause analysis currently top out at 56.2%
The Hacker News — AI hallucinations in security ops thehackernews.com
hallucinated guidance can lead an AI to delete sensitive databases or disable firewall rules under the guise of ‘containment’
Skadden — Responding to AI-driven cyber incidents skadden.com
‘Codex Red’… detailed an incident where a developer used Codex to respond to a Linux compromise; the agent inadvertently masked symptoms (such as silencing cryptominer-related fan noise) and generated ‘noisy’ commands that complicated the subsequent investigation
dev.to — Using Codex for code review dev.to
Codex successfully addressed local syntax issues but failed to recognize deeper anti-patterns, such as tight coupling between the domain and the database
crg.eti.br — Glaze project page (Cesar Gimenes) crg.eti.br
Glaze binds to the WebView runtime already present on the user’s machine — WebView2 on Windows, WKWebView on macOS, and WebKitGTK on Linux — and uses purego to call OS APIs directly, so projects can cross-compile from any host without a C toolchain.
github.com/crgimenes/glaze github.com
AppWindow wraps a standard http.Handler in a native window; on macOS and Linux it communicates over Unix sockets, on Windows over TCP loopback, and supports custom app:// schemes so no public web server is exposed.
playwright.dev — WebView2 automation docs playwright.dev
Playwright connects to WebView2 via the Chrome DevTools Protocol by launching the host app with —remote-debugging-port=9222 and then using chromium.connectOverCDP().
context.dev — Playwright vs Selenium context.dev
Playwright’s auto-waiting feature pauses actions until elements are actionable, providing a level of reliability that lightweight tools currently struggle to match; however, launching a full browser instance per URL makes large-scale scraping (e.g. 90,000 sites) resource-intensive.
simonwillison.net/tags/slop — Willison on his own research repo simonwillison.net
Willison describes the contents of simonw/research as ‘total slop’ and has suggested GitHub add an ‘exclude from search’ flag so quarantined AI-generated reports don’t pollute indexes.
explainx.ai — Slopocalypse: AI slop in 2026 explainx.ai
curl maintainer Daniel Stenberg now instantly bans submitters of AI-generated vulnerability reports, and projects are deploying tools like ‘SlopGuard’ to score and quarantine PRs showing high markers of unreviewed AI generation.