AI for Builders

AI for Builders: Agents Patched a Thousand Kernel Holes. Five AI stories for engineering leaders: the headline, what it means for your team, and a verdict (worth a pilot / wait a quarter / ignore), plus one thing to try this week. STORIES 1. OpenAI's autonomous pipeline found 948 Linux kernel privilege escalations and wrote the patches (OpenAI) — verdict: worth a pilot — https://go.kasidwala.com/openai-linux-kernel-patches 2. Five engineers rebuilt X.com with AI agents in under six months, first post in 2.7 seconds instead of 22 (X) — verdict: worth a pilot — https://go.kasidwala.com/x-rebuilt-by-agents 3. Parcel 3 rewrites the bundler in Rust and reports cold builds up to 100x faster than v2 (Parcel) — verdict: worth a pilot — https://go.kasidwala.com/parcel-3-rust-rewrite 4. Tencent's Hy4 sees screens and uses computers, but the open weights are still a promise (Tencent) — verdict: wait a quarter — https://go.kasidwala.com/tencent-hy4-model 5. OpenAI's Images 2.6 adds GIFs and two API models, 50 percent faster (OpenAI) — verdict: worth a pilot — https://go.kasidwala.com/openai-images-2-6 TRY THIS WEEK Run one agent security scan on a small service with human review on every finding, write an AGENTS.md at your repo root with the lint, typecheck, and test rules every change must pass, and benchmark the Parcel 3 alpha against your current bundler on one project using your own cold build times. Tools: Codex with its security plugin, a repo-root AGENTS.md file, and the Parcel 3 alpha. Tied to stories 1, 2, and 3. ABOUT Mustufa Kasidwala is a software leader with 20+ years of experience, working at the intersection of engineering leadership and GenAI-accelerated software development. Connect: https://go.kasidwala.com/linkedin AI DISCLOSURE This episode was produced with the help of AI, including AI-generated voices. AI Generated

AI for Builders: Siri Opens to Every Developer. Five AI stories for engineering leaders: the headline, what it means for your team, and a verdict (worth a pilot / wait a quarter / ignore), plus one thing to try this week. STORIES 1. Apple opens Siri AI to every developer through App Intents (Apple) — verdict: worth a pilot — https://go.kasidwala.com/apple-siri-ai-developer-preview 2. Postman auto-installed plugins into Codex and Claude Code, and tried to approve its own hook (dev.to) — verdict: worth a pilot — https://go.kasidwala.com/postman-agent-plugins 3. Maple ported 287,000 lines of React in 3 days, and proving nothing changed was the real work (Maple) — verdict: worth a pilot — https://go.kasidwala.com/maple-react-to-foldkit-port 4. Netflix lets agents find performance bugs, with canaries and a human deciding (ByteByteGo) — verdict: worth a pilot — https://go.kasidwala.com/netflix-ai-performance-agents 5. Cloudflare stopped rebuilding whole DNS zones for small changes, cutting peak memory 85% (Cloudflare) — verdict: worth a pilot — https://go.kasidwala.com/cloudflare-dns-zone-builds TRY THIS WEEK Audit your agent configs, profile one slow service into a shared pattern catalog entry, and write a done rule before your next refactor. Tools: Codex and Claude Code config files, your profiler, a Git Markdown catalog. Tied to stories 2, 3, and 4. ABOUT Mustufa Kasidwala is a software leader with 20+ years of experience, working at the intersection of engineering leadership and GenAI-accelerated software development. Connect: https://go.kasidwala.com/linkedin AI DISCLOSURE This episode was produced with the help of AI, including AI-generated voices. AI Generated

AI for Builders: The $1,000 Firefox Exploit. Five AI stories for engineering leaders: the headline, what it means for your team, and a verdict (worth a pilot / wait a quarter / ignore), plus one thing to try this week. STORIES 1. An open-weight model built working Firefox exploits from public patches, for about $1,000 each (Epoch AI) — verdict: worth a pilot — https://go.kasidwala.com/firefox-patch-gap-exploits 2. OpenAI rebuilt Git for its monorepo with Pando, cutting median push latency nearly 5x (OpenAI) — verdict: wait a quarter — https://go.kasidwala.com/openai-pando-git 3. Coding agents are collapsing their tool catalogs into a single exec tool (Phil Schmid) — verdict: worth a pilot — https://go.kasidwala.com/agents-one-exec-tool 4. LangChain's Managed Deep Agents add a browser with human sign-in handoff and async subagents (LangChain) — verdict: worth a pilot — https://go.kasidwala.com/langchain-deep-agents-browser 5. Vercel merges Data Cache into Runtime Cache, with no code changes but a billing change (Vercel) — verdict: worth a pilot — https://go.kasidwala.com/vercel-runtime-cache TRY THIS WEEK Pick one coding agent your team already uses. Give it a single exec tool inside a sandbox, with command-level logging turned on, and run it on a real but low-risk task. At the same time, pull your fleet report and write down how long browsers on your team actually take to update after a release. Record three things: how many round trips the exec tool saved, how long an audit of one run took, and your browser update lag. Tools: your coding agent and its sandbox, your fleet update report. Tied to stories 1 and 3. ABOUT Mustufa Kasidwala is a software leader with 20+ years of experience, working at the intersection of engineering leadership and GenAI-accelerated software development. Connect: https://go.kasidwala.com/linkedin AI DISCLOSURE This episode was produced with the help of AI, including AI-generated voices. AI Generated

AI for Builders: The Weekend's Always-On Coding Agent. Five AI stories for engineering leaders: the headline, what it means for your team, and a verdict (worth a pilot / wait a quarter / ignore), plus one thing to try this week. STORIES 1. DeepSeek open sources a proactive coding agent that watches a GitHub repo and returns pull requests (GitHub) — verdict: worth a pilot — https://go.kasidwala.com/deepseek-superagent 2. A tampered Claude Code extension reached developers through Cursor's marketplace (The Hacker News) — verdict: worth a pilot — https://go.kasidwala.com/cursor-claude-code-backdoor 3. Cisco gives a 27-year-old engineer unlimited AI tokens to rewrite 98% of its legacy code in 24 months (Fortune) — verdict: worth a pilot — https://go.kasidwala.com/cisco-legacy-rewrite 4. A benchmark measures what the same app costs a coding agent to build across JavaScript frameworks (Nolan Lawson) — verdict: worth a pilot — https://go.kasidwala.com/framework-cost-benchmark 5. One company installs no package version until it is seven days old (TLDR Dev) — verdict: worth a pilot — https://go.kasidwala.com/seven-day-install-delay TRY THIS WEEK First, ask whether anyone on your team installed or updated Claude Code from Cursor's marketplace on October 29. If anyone did, check the machine for the svc-telemetry key in authorized_keys and rotate what it could reach, from a clean machine, then turn on an extension allow list for one team. Second, try the pattern from DeepSeek's super agent design on a throwaway repo: a separate GitHub account for the agent, work as tickets, isolated sessions, and pull requests as the only output, and record how often it needed a human. Tools: DeepSeek Harness and the dsh-superagent prototype, your extension allow list. Tied to stories 1 and 2. ABOUT Mustufa Kasidwala is a software leader with 20+ years of experience, working at the intersection of engineering leadership and GenAI-accelerated software development. Connect: https://go.kasidwala.com/linkedin AI DISCLOSURE This episode was produced with the help of AI, including AI-generated voices. AI Generated

AI for Builders: Open Weights, Agent Intrusions, and the Adoption Gap. Five AI stories for engineering leaders: the headline, what it means for your team, and a verdict (worth a pilot / wait a quarter / ignore), plus one thing to try this week. STORIES 1. Canada invests up to C$1.6 billion in Cohere for sovereign AI (Prime Minister of Canada) — verdict: wait a quarter — https://go.kasidwala.com/canada-cohere-stake 2. Mistral opens the weights of its trillion-parameter Large 4 model (Mistral) — verdict: worth a pilot — https://go.kasidwala.com/mistral-large-4 3. GitHub makes Copilot Fusion generally available as agents start over 40% of pull requests (GitHub) — verdict: worth a pilot — https://go.kasidwala.com/github-copilot-fusion 4. CrowdStrike reports an agent framework running intrusions against more than 200 organizations (CrowdStrike) — verdict: worth a pilot — https://go.kasidwala.com/crowdstrike-agent-intrusions 5. White & Case holds back its newest legal agent as lawyers hesitate (The Information) — verdict: wait a quarter — https://go.kasidwala.com/white-case-legora-agents TRY THIS WEEK Pick one real coding task, such as a bug fix or a dependency update, and run it through GitHub Copilot Fusion. Before the agent starts, write down what a good result looks like. Then review it the way you would review a junior developer's work, and write down how long review took, what it cost, and how many times a human had to step in. Tools: GitHub Copilot Fusion and your existing review process. Tied to stories 3 and 5. TRY THIS WEEKEND Build a read-only triage agent: a scheduled job that lists pull requests open for more than seven days, writes a short summary of each one, and posts it as a comment. It must never merge anything, and it must log every action it takes. Give it the smallest permissions that let it read and comment. Tools: Copilot agent features or any agent tool you already have, plus your repository's scheduled jobs. Tied to stories 3 and 4. ABOUT Mustufa Kasidwala is a software leader with 20+ years of experience, working at the intersection of engineering leadership and GenAI-accelerated software development. Connect: https://go.kasidwala.com/linkedin AI DISCLOSURE This episode was produced with the help of AI, including AI-generated voices. AI Generated