TanStack Strikes Back

Today’s issue: Too much git, Tesla’s cure for developer loneliness, and love letters to your bros written in code.

Welcome to #530.

---

TanStack Strikes Back

I imagine trying to compete against Next.js feels a little bit like going up against The Galactic Empire. And while Tanner hasn’t blown up the Death Star and discovered that Pete Hunt is actually his dad, TanStack Start hitting 1.0 this week is evidence that the rebellion is alive and well.

TanStack announced the release candidate for Start back in September 2025 and since then the framework has exploded in growth, with companies like Lovable making it the default framework for all of their apps. And while this has been more of a slow build than an overnight success, it is interesting to consider what has enabled Start to become Next’s biggest competition.

What TanStack got right:

• Simpler primitives: Start benefited from seeing where Next’s App Router frustrated developers. Nested directory conventions, caching directives, and a strong push toward RSC added complexity that didn’t appeal to everyone. TanStack offered a more familiar React model: routes, loaders, and server functions, with explicit control over where code runs and how pages render.

• Type-safe by default: TanStack connects types across route parameters, search parameters, loaders, and server functions. Next offers typed routes and server actions too, but TanStack makes that end-to-end connection a core part of the framework, catching more mistakes as you move between navigation, data loading, and backend calls.

• Platform agnostic: Next supports other hosts, but its close ties to Vercel can make that choice feel more consequential. Start offers deployment options across multiple providers and runtimes, letting teams choose a framework without committing to a hosting provider. Keeping those decisions separate makes Start an easier bet.

• Performance: Teams switching from Next.js to Start have reported substantial gains. Inngest cut local page loads from 10–12 seconds to 2–3 seconds, Railway cut builds from over 10 minutes to under two, and Lovable reported 49% lower median time to first byte after its broader migration.

Bottom Line: TanStack Start is winning over the hearts of the agent developer alliance … which means Emperor Rauch trying to lure Tanner to the dark side will probably be one of our top Bytes predictions for 2027.

---

Bridge the gap between AI and quality

Getting agents to produce quality code requires the right tools. Fortunately, one of those tools, SonarQube, just got a lot easier to use.

If you’re not familiar with Sonar, it’s been analyzing code for over 17 years and processes 750 billion lines of code daily. And its new SonarQube MCP Server puts that analysis inside your AI coding agent, giving your agent access to Sonar’s entire code intelligence platform.

With the SonarQube MCP Server, you can:

• Ask “Are there any new vulnerabilities in this file?” in plain English
• Check your project’s quality gate status without leaving your IDE
• Update an issue’s status or mark it as a false positive directly from your AI coding agent

Set up the SonarQube MCP Server for free and help your agents start shipping better code.

---

1. Anthropic built their own source control platform that can keep up with the 30,000 merges they do a day. Rumor is that 29,000 of these are from Jared Sumner’s clawd army.

2. Speaking of git platforms, Jovi De Croock wrote about how he built his own git platform for agents that doesn’t have PRs. PRs really were just love letters to your bros written in code… RIP 💔.

3. Apify’s new plugin connects your coding agent over MCP to live data from Google Maps, Instagram, X, and LinkedIn, with skills, hooks, and rules built in. Use code BYTES for $25 in credits. [sponsored]

4. Jenna Smith and James Nelson wrote about the intricacies of building a filterable menu that works for everyone.

5. SpaceXSI (did I get this right?) just released Grok Voice 3.0, the same model that lonely Tesla commuters use as a substitute for friendship.

6. The TypeSafe AI team released Jev 1.14.0 alongside experimental image support, making your dream of building “Hot Dog / Not Hot Dog” a reality.

7. Only idiots write tests – modern engineering teams like Notion, Wiz and ElevenLabs use Meticulous to maintain e2e UI tests that cover every edge case of your web app. [sponsored]

8. Shopify just released an overhaul of Hydrogen, their framework for building custom Shopify sites. This new version is framework agnostic but comes with bindings for React and Vue out of the box.

9. Cheng Lou broke the internet again with another “toy project” called Handwired that converts JavaScript into a neural net that runs in the browser and works without training? I give up trying to explain it. Go see it for yourself.

10. Hyperagent lets you hire AI staff to take on roles across your business. They run in the cloud with their own browser, shell, and file system, connect to your tools, and hand work off to each other. Your whole team can work with the same agents—not just the person who set them up. [sponsored]

11. Astro released version 7.4, which adds a new flag astro build --timings that gives you a breakdown of how long everything took, separate build and runtime image services, and experimental parallel prerendering support.

12. Nick Cammarata wrote an article that will take you an hour to read, uses words that you don’t know (like polysemanticity), but will teach you everything you want to know about the interoperability of neural networks.