Someone quote-tweeted my blog last week with: “Bro you have a link to an early 2000’s looking blog lol stop”
It stung. Mostly because they weren’t wrong.
I’ve been running Hexo with the Icarus theme since 2024. It works. It’s fast. It costs me basically nothing on S3 + CloudFront. But let’s be honest, it was dated. Good enough for the pre-AI-agents era when you just needed a free framework to spin something up quick. No excuse in 2026 though, when coding agents can help you ship a maximalist, visually impressive personal site in an afternoon. The sidebar widgets, the slightly-off spacing, the default blue links. I’ve been so focused on writing content and building AI workflows that I never stopped to ask if my blog looked like something people would actually want to read.
That tweet was the wake-up call. So I did what I do nowadays: I asked Claude for a deep research of the options, considering the landscape in 2026. Specifically: which static blogging framework plays best with AI coding agents?
Because I’m not rebuilding this thing by hand. Claude Code and Cursor are doing the heavy lifting. The framework needs to work with them, not against them.
Here’s what I found.
Astro is the answer for most of us
I’ll save you the suspense. After looking at everything, Astro is the clear winner for developer blogs in 2026.
Astro (v5.17 stable, v6.0 in beta) hit 56,600+ GitHub stars and 900,000+ weekly npm downloads in 2025 [1]. Cloudflare acquired them in February 2026 [2], which tells you everything about where the momentum is heading. This isn’t a framework that’s going away.
Why does it work so well for blogs specifically?
Zero JavaScript by default. Astro ships pure HTML and CSS. Your blog loads instantly. Near-perfect Lighthouse scores without trying. When you do need interactivity (a search widget, an animation, a comment system) you use Astro’s “islands architecture” to hydrate just those components. The rest of your site stays static.
Content Collections are AI-friendly. The Content Layer API gives you type-safe Markdown and MDX with Zod schema validation and frontmatter type-checking [3]. This matters more than you’d think. When I ask Claude Code to generate a new blog post, it can reliably produce the right structure because the schema is explicit. No guessing. No “I think the frontmatter looks like this.”
View Transitions with two lines of code. Smooth page-to-page animations that feel like a native app. No JavaScript library. Browser support is at 85%+ now that Firefox shipped native support. Combined with Tailwind CSS, built-in image optimization (AVIF/WebP), and the ability to drop in Framer Motion via React islands, Astro can look genuinely impressive without the bundle bloat.
The theme ecosystem is massive. Over 1,000 themes on astrothemes.dev [4], plus 400+ on Statichunt. AstroPaper (4,200 stars, clean and minimal), Fuwari (Obsidian-like aesthetic), AstroWind (SaaS-clean). Nearly all use Tailwind with dark mode out of the box. Compare that to my Hexo theme situation: a handful of options, most last updated in 2022.
AI agents actually build with it successfully. This is the part that sealed it for me. One developer rebuilt their entire WordPress site in Astro using Claude Code in 3 hours. Another migrated 63 Jekyll posts in “a few hours.” A Ghost-to-Astro migration took 5 hours with 75% memory reduction [5]. These aren’t hypotheticals. People are shipping production blogs in a day.
The one pain point: AI agents sometimes forget the --- frontmatter separators in .astro files. You need to be explicit about it in your prompts. Cursor rules for Astro exist on cursor.directory [6], and there’s a dedicated Claude Code skill that covers Content Collections, deployment, and i18n.
Deployment? Astro outputs pure static files, so it works anywhere. S3 + CloudFront (my current setup, and I’m keeping it), Cloudflare Pages, Vercel, Netlify, GitHub Pages. No vendor lock-in, no server-side runtime to worry about.
Downsides? The .astro file format is proprietary (JSX-like but not JSX), so AI training data is smaller than for React. The ecosystem is younger than Hugo’s. For sites with thousands of pages, Hugo still builds faster. Some people think Astro is overkill for a simple blog. Maybe. But the blog starter template is dead simple, and you’re not paying for the extra capability unless you use it.
Next.js: maximum AI reliability, maximum overkill
Next.js (v15.5 stable, v16 in beta) has 130,000+ GitHub stars [7] and the largest training corpus of any framework. AI agents just know Next.js better than anything else. Someone once said “the best examples of AI building something good in one shot were using Next.js and TailwindCSS,” and that matches my experience too.
The cursor.directory ecosystem proves it: 12+ dedicated Next.js rule sets [6] covering App Router, TypeScript, Tailwind, SEO. Astro has 1-2. If you lean heavily on AI, this breadth of tooling is real.
The React ecosystem also gives you the highest visual ceiling. Framer Motion, shadcn/ui, Radix primitives. If “wow factor” means complex scroll animations and interactive data viz, Next.js has the deepest toolkit.
But here’s the thing. Next.js is massively over-engineered for a blog. App Router, React Server Components, streaming SSR, ISR, CSR. You don’t need any of this for a content site. One benchmark found Astro loading pages in 746ms with 27 requests versus Next.js at 21.85 seconds with 72 requests for similar content [8]. The React runtime ships whether you need it or not.
I’ve built a Next.js blog before (this one: mygutjournal). It works fine. But the overhead is real. If your blog is part of a bigger React app, Next.js makes sense. If it’s a standalone blog? You’re carrying unnecessary weight.
Themes: Fragmented across Vercel’s gallery, GitHub, and third-party sites. The tailwind-nextjs-starter-blog [9] is the community standard and it’s legitimately good. Nextra works well for docs-plus-blog. But discovery is harder than Astro’s centralized gallery.
Deployment note: Next.js really wants to live on Vercel. You can do a static export to S3 + CloudFront, but you lose the SSR, ISR, and image optimization features that make Next.js worth picking in the first place. If you’re on AWS, you’re better off with a framework that’s static-first.
Hugo: unbeatable speed, terrible AI story
Hugo (v0.146+) builds at roughly 1ms per page. A thousand pages in 2 seconds. Nothing else comes close. Single Go binary, zero dependencies. It powers the Kubernetes docs, Smashing Magazine, and Cloudflare Developer Docs [10]. 86,000+ GitHub stars. The thing is battle-tested.
Recent releases added LaTeX typesetting, KaTeX math, Obsidian-style callouts, and Tailwind support. If you have 5,000+ pages, Hugo is still king.
But here’s where it falls apart for me. Hugo’s Go templates are the least AI-friendly templating language of any major SSG. The syntax ({{ if eq .PageNumber 1 }}) is counterintuitive. The dot (.) context shifts within templates confuse every AI agent I’ve tried. Go templates are a tiny fraction of AI training data compared to JavaScript/TypeScript.
I looked for Hugo-specific cursor rules. Nothing. Claude Code skills for Hugo? Nothing. “AI built my Hugo site” success stories? Nothing.
That’s a dealbreaker in 2026. If I’m going to have an AI agent help me rebuild my blog, the framework needs to be something the agent can actually work with reliably.
Theme ecosystem: Hundreds of themes at themes.gohugo.io. PaperMod is popular and minimal. Blowfish and Congo bring modern Tailwind designs. But honestly, a lot of Hugo themes feel dated. The aesthetic skews functional, not beautiful.
Other downsides: Steep Go template learning curve. No MDX (shortcodes instead). Hard to extend since it’s compiled. And the lock-in is real. Several developers report wanting to leave Hugo but finding migration too painful.
Eleventy and SvelteKit: niche picks
Eleventy (v3.1.0) is the principled minimalist’s choice. Zero client-side JS, zero framework lock-in, 11 template languages, builds that handle 4,000 Markdown files in under 2 seconds. NASA, CERN, W3C, and Google use it. It joined Font Awesome in 2024 for long-term sustainability.
The trade-off: ~100 starter projects versus Astro’s 1,000+. Themes lean functional, not visually striking. AI compatibility is middling — JavaScript-based (good), but underrepresented in training data (bad), and supporting 11 template languages can confuse AI agents about which syntax to use. If you’re the kind of person who writes their own CSS from scratch and enjoys it, Eleventy is great. That’s not me right now.
SvelteKit (v2.20+, Svelte 5) has the smallest blog theme ecosystem (~20 templates) but uniquely elegant built-in transitions. Svelte compiles to minimal JS with no runtime. The problem? Svelte 5’s runes paradigm shift broke AI compatibility. Agents consistently generate outdated Svelte 4 patterns. Unless you’re already deep in the Svelte ecosystem, this isn’t where you start.
Worth watching
Zola (Rust-based, 14,900+ stars) is interesting. Some benchmarks show it 4x faster than Hugo. Single binary, built-in taxonomy, pagination, RSS, search. Its Tera templating is easier than Go templates. But the ecosystem is tiny and AI training data is minimal.
Lume (Deno-based) is gaining traction. Deno’s own docs migrated from Docusaurus to Lume citing better Core Web Vitals. Worth watching, too early to recommend.
AI-native builders like Bolt.new, v0, and DeepSite generate entire sites from prompts. They’re impressive demos. But for a developer who actually wants to own and evolve their blog, the combination of a real framework + AI agent (Astro + Claude Code) is more powerful and flexible.
The stack I’m going with
For anyone in my position (static blog, needs to look modern, building with AI agents), here’s the stack:
- Framework: Astro 5.x
- Styling: Tailwind CSS v4 + Typography plugin
- Content: Markdown/MDX via Content Collections
- Visual polish: Built-in View Transitions + Framer Motion via React islands where needed
- Theme base: AstroPaper or Fuwari, then customize with AI
- Deployment: AWS S3 + CloudFront. I already have the infra, and Astro’s static output drops right in.
astro buildproduces adist/folder, sync it to S3, invalidate CloudFront. Done. Costs pennies. - AI workflow: Claude Code with Astro-specific prompts; always remind it about
.astrofrontmatter separators
This maximizes the intersection of visual impact, performance, AI compatibility, and simplicity. The content-first architecture means AI agents spend time on design instead of fighting framework complexity. The island architecture means adding interactive elements never kills your page speed.
What this means for my Hexo blog
Yeah, I’m migrating. Hexo served me well. The S3 + CloudFront infrastructure stays, it’s cheap and it’s fast. I’m just swapping the framework and theme. The Icarus theme was fine for a while. But “fine for a while” doesn’t cut it when someone screenshots your blog to roast you on Twitter.
The framework landscape has shifted. AI compatibility is now a real factor in choosing your stack. Frameworks with larger JavaScript-ecosystem footprints and structured content models have a measurable advantage, because AI agents can generate and modify their code more reliably. That single factor narrows the field to Astro.
I’ll document the migration when it happens. Knowing how these things go with Claude Code, it’ll probably take an afternoon.
References
[1] Astro GitHub repository (56,600+ stars, 900K+ weekly npm downloads). https://github.com/withastro/astro [2] Cloudflare acquires Astro (February 2026). Hacker News, 596 points. https://news.ycombinator.com/ [3] Astro Content Collections documentation. https://docs.astro.build/en/guides/content-collections/ [4] Astro Themes directory (1,000+ themes). https://astrothemes.dev/ [5] AI migration success stories: WordPress-to-Astro in 3 hours (Claude Code), Jekyll 63-post migration (Claude Code), Ghost-to-Astro in 5 hours with 75% memory reduction (Emergent Minds). [6] Cursor Directory — AI rules for frameworks. https://cursor.directory/ [7] Next.js GitHub repository (130,000+ stars). https://github.com/vercel/next.js [8] Roger Wong, performance comparison: Astro 746ms/27 requests vs Next.js 21.85s/72 requests. [9] tailwind-nextjs-starter-blog. https://github.com/timlrx/tailwind-nextjs-starter-blog [10] Hugo-powered sites: Kubernetes docs, Smashing Magazine, Cloudflare Developer Docs. https://gohugo.io/showcase/