🍿 3 fixes that cut my AI bill in half (+ latest news)

PRESENTED BY SONARQUBE

We all hate security scanners that flood our PRs with false positives. It's the quickest way to get a dev to ignore alerts altogether.

SonarQube Hunter Agent is different.

It’s an AI-powered agent that hunts for business logic flaws and actually validates them before bugging you.

It runs targeted playbooks to prove a vulnerability is real.

When it finds one, it drops a plain-English explanation into your SonarQube issue list showing exactly where the intent and implementation mismatched.

It's like having a dedicated AppSec engineer reviewing your logic ✨.

Ever hit your AI usage limit by Wednesday and wondered where all those tokens actually went?

It's usually not the model pricing - it's your workflow. Maximum effort on trivial tasks, rework from the wrong first attempt, and flagship models doing jobs a mini model could handle quietly double your bill.

How to Use It

→ Dial down your effort level - stay on medium by default, use low for boilerplate, and save high/max for genuinely hard bugs

→ Always start in plan mode - ~2,000 tokens of planning can save you ~200,000 tokens of rework

→ Start high, downgrade fast - get it working with Opus first, then run the same prompts through a mini model like Haiku and switch if it passes

🚀 The Takeaway

Match the effort and the model to the task. Want all 7 fixes? Watch the full video here​

AI

CopilotKit just open-sourced OpenDots - always-on AI coworkers you can text, call, or @mention in Slack. Each one has its own role, permissions and its own computer. Clone it and see how far "AI teammate" has actually come.

Claude Opus 5.5 can (finally) generate motion graphics - except it doesn’t output video at all. It writes code that draws every frame. Showreels and launch videos from a single prompt, and the examples are wild.

Can’t keep up with AI papers? 1kpapers summarizes the most important ones from 2025–2026, sorted by topic, lab, citations and code. Bookmark this one.

One of my favourite features shipped on Clerk - Audit Dashboard activity with admin logs.

​Ava Studio is a weird one: set a goal and it keeps producing campaigns, UGC ads and videos from one workspace, on its own. Is this what a "content machine" looks like?

TanStack Intent 0.5 solves a problem you didn’t know you had: your agent skills silently go stale every time your library changes. Now you can type-check the code examples inside your skills and review what each source change broke.

Patch Next.js today. The September security release fixes a high-severity SSRF in Image Optimization plus several cache-poisoning bugs. Upgrade to 16.3.8 or 15.5.27 - check if your remotePatterns config makes you a target.

​
​ GREAT READ

Next.js is no longer slow, thanks to Instant Navigations that dropped inside of Next.js 16.3

Clerk now supports multiple enterprise SSO connections per Organization - useful for teams managing multiple identity providers.

​🔥 Love it!​

​😐 It’s okay​

​👎 Could be better​

If you’ve published a blog post or shipped something, feel free to reply to this email and it comes straight to my inbox. I’m always looking for great community work to consider featuring.