The content-pipeline version of 'AI is making me dumb': you stop manually reviewing which pages get indexed vs. dropped. The model runs fine — your read on the data atrophies. You stop catching the systematic errors you're no longer looking for. #ai

Burned my GitHub Actions free minutes (2k/mo, private repos) by day 25. The culprit wasn't the slow jobs — it was CI re-running a 4-app matrix build on every bot content commit. Fix: paths-ignore content/**, let Cloudflare gate deploys. The cheap, frequent jobs bleed you. #buildinpublic

Qwen3.6-235B is open-weight, but Alibaba closed its flagship simultaneously. Pattern: smaller models open, frontier stays proprietary. For pipelines, open-weight lags frontier by one product cycle. Plan inference costs around that lag, not the benchmark. #ai

Three post-deploy checks I run after every Cloudflare Pages build After hitting a sitemap _redirects bug and a Bluesky image race condition, I added three targeted smoke checks that catch silent Cloudfl… https://dev.to/morinaga/three-post-deploy-checks-i-run-after-every-cloudflare-pages-build-411o

Why I'm betting on AI-curated directories when Google AI Overviews answer the same queries The core counterargument to my three directory sites is Google's own AI Overviews. Here… https://dev.to/morinaga/why-im-betting-on-ai-curated-directories-when-google-ai-overviews-answer-the-same-queries-33pf

My programmatic pages now get referral clicks from Perplexity, ChatGPT, and Claude combined that rival Google positions 4-6. The total is still small — but the conversion rate is different. Is anyone else tracking AI tool referral separately in analytics? #indiehackers

Cloudflare Workers runs at 300+ PoPs with no cold starts. Vercel Edge Functions run at fewer PoPs and can cold-start on free tier. For a 10K-page programmatic site, that TTFB gap shows up in CrUX data — which factors into Google ranking. Not theoretical. #webdev

Long context vs. RAG isn't a versus. Long context handles per-call reasoning over complex source docs. Vector retrieval handles 'which of 100K entities relates to this?' At scale, they compose: retrieval narrows the set, context does the reasoning. #ai

Five overlooked packages running my AI directory stack A curated look at tsx, Pagefind, Crawlee, eemeli/yaml, and @libsql/client — the unsexy dependencies doing most of the actual work. https://dev.to/morinaga/five-overlooked-packages-running-my-ai-directory-stack-230p

Five things that caught my attention this week in AI tools and open-source models Claude Code Agent View, ZAYA1-8B on AMD hardware, Harness productivity data, ServiceNow in every IDE, an… https://dev.to/morinaga/five-things-that-caught-my-attention-this-week-in-ai-tools-and-open-source-models-39gk

Three post-deploy checks I run after every Cloudflare Pages build After hitting a sitemap _redirects bug and a Bluesky image race condition, I added three targeted smoke checks that catch silent Cloudfl… https://dev.to/morinaga/three-post-deploy-checks-i-run-after-every-cloudflare-pages-build-1p21

Why I'm betting on AI-curated directories when Google AI Overviews answer the same queries The core counterargument to my three directory sites is Google's own AI Overviews. Here… https://dev.to/morinaga/why-im-betting-on-ai-curated-directories-when-google-ai-overviews-answer-the-same-queries-18fn

Cited inside a Google AI Overview: +35% organic clicks, +91% paid clicks vs. non-cited competitors on identical queries. Position 1 organic below the AI box is less commercially valuable than a citation inside it. Is anyone tracking AI citation rate as a primary KPI? #ai

Astro sites pass Core Web Vitals at 66% vs 30% for Next.js, per HTTP Archive and Chrome UX Report. For programmatic sites where CWV directly affects click-through in Search, framework choice is a baseline ranking variable — not a DX preference. #webdev

DeepSeek V4 Pro scores negative on hallucination benchmarks — it hallucinates more than open-weight peers despite leading on coding tasks. For factual content pipelines, leaderboard rank is the wrong filter. Task-level model selection beats defaulting to the top entry. #ai

Three post-deploy checks I run after every Cloudflare Pages build After hitting a sitemap _redirects bug and a Bluesky image race condition, I added three targeted smoke checks that catch silent Cloudfl… https://dev.to/morinaga/three-post-deploy-checks-i-run-after-every-cloudflare-pages-build-39ee

Why I'm betting on AI-curated directories when Google AI Overviews answer the same queries The core counterargument to my three directory sites is Google's own AI Overviews. Here… https://dev.to/morinaga/why-im-betting-on-ai-curated-directories-when-google-ai-overviews-answer-the-same-queries-2mp9

Conventional advice: cover the full entity graph for topical authority. My read: a 500-page site owning one tight niche outranks a 10K-page site spread thin across 50 topics. Breadth costs build time. Does it buy proportional ranking signal? My data says no. #ai

Embedding batch size is almost never tuned but directly affects throughput for large content pipelines. The OpenAI embedding API accepts up to 2048 inputs per call — most pipelines send one at a time. Optimal batch size depends on average token length. Worth measuring once. #ai

Most tool directories deploy Article JSON-LD. SoftwareApplication is the correct type — it has applicationCategory, offers, and operatingSystem fields that directly feed rich results. Schema type is free to switch. Most leave that equity on the table. #webdev

Static site search for Astro in 2026: why I picked Pagefind over Algolia and Lunr A practical comparison of Pagefind, Algolia DocSearch, and Lunr.js for an Astro SSG directory — build-time… https://dev.to/morinaga/static-site-search-for-astro-in-2026-why-i-picked-pagefind-over-algolia-and-lunr-pg1

How I built pairwise AI model compare pages with Claude Haiku and a budget cap The O(n²) combinatorics problem with AI directory compare pages, how I budget-capped Claude Haiku calls, and du… https://dev.to/morinaga/how-i-built-pairwise-ai-model-compare-pages-with-claude-haiku-and-a-budget-cap-ia0

Three post-deploy checks I run after every Cloudflare Pages build After hitting a sitemap _redirects bug and a Bluesky image race condition, I added three targeted smoke checks that catch silent Cloudfl… https://dev.to/morinaga/three-post-deploy-checks-i-run-after-every-cloudflare-pages-build-40nh

Why I'm betting on AI-curated directories when Google AI Overviews answer the same queries The core counterargument to my three directory sites is Google's own AI Overviews. Here… https://dev.to/morinaga/why-im-betting-on-ai-curated-directories-when-google-ai-overviews-answer-the-same-queries-4f7d

Counterintuitive: programmatic sites with public data changelogs earn more backlinks than those without. A changelog page is fresh, specific, quotable — and your pipeline already produces the data. No one exposes it because it feels internal. Would you publish yours? #webdev

GSC API returns indexation status per URL, for free. Most programmatic sites check it manually once a month. Automating a nightly diff of ‘submitted but not indexed’ catches drop-offs in 24 hours, not weeks. The data is already there — it’s just a fetch call away. #webdev

Running generation at temperature 0 for factual fields and 0.7 for prose isn’t just a quality call — it’s a consistency call. Exact fields (prices, dates, IDs) need deterministic output. Most pipelines treat temperature as one global setting when it should be per-field. #ai

Three post-deploy checks I run after every Cloudflare Pages build After hitting a sitemap _redirects bug and a Bluesky image race condition, I added three targeted smoke checks that catch silent Cloudfla… https://dev.to/morinaga/three-post-deploy-checks-i-run-after-every-cloudflare-pages-build-f8i

Why I'm betting on AI-curated directories when Google AI Overviews answer the same queries The core counterargument to my three directory sites is Google's own AI Overviews. Here… https://dev.to/morinaga/why-im-betting-on-ai-curated-directories-when-google-ai-overviews-answer-the-same-queries-5g3i

Bet: translating a programmatic site into 20 locales with an LLM has better ROI than the same budget on new English pages. Tail-language traffic adds up. The risk is duplicate content signals, not quality. Anyone running multilingual programmatic at scale? #ai

Sitemap `lastmod` and JSON-LD `dateModified` are separate freshness signals — syncing one without the other creates inconsistency. Most programmatic pipelines update sitemap on rebuild but never touch structured data timestamps. Two signals, one update routine. #webdev

Content pipelines accumulate invisible drift: same prompt, different model version, different output. Nothing in git tracks prompt × model × date as a single unit. Regenerating pages whose diff you can't audit is a silent problem most pipelines ignore. #webdev

Three post-deploy checks I run after every Cloudflare Pages build After hitting a sitemap _redirects bug and a Bluesky image race condition, I added three targeted smoke checks that catch silent Cloudfl… https://dev.to/morinaga/three-post-deploy-checks-i-run-after-every-cloudflare-pages-build-3l67

Why I'm betting on AI-curated directories when Google AI Overviews answer the same queries The core counterargument to my three directory sites is Google's own AI Overviews. Here… https://dev.to/morinaga/why-im-betting-on-ai-curated-directories-when-google-ai-overviews-answer-the-same-queries-54kg

ISR exists to avoid rebuilding 10K pages when 100 update. For programmatic sites with nightly full refreshes, a complete edge build takes 5 minutes anyway. Is ISR solving the right problem at indie scale, or is full rebuild simpler to reason about? #webdev

Anthropic’s Batch API cuts inference cost 50% for async workloads. Content pipelines running thousands of calls per night qualify by definition. Most devs use the synchronous endpoint out of habit. That’s not a pipeline constraint — it’s a configuration choice. #ai

HiDream-O1-Image shipped as open weights last week. Real question for content pipelines: can it handle OG image generation at build time without a paid API? Throughput and reproducibility matter more than aesthetic quality at 10K pages. The test is cheap — one GPU-hour. #ai

Three post-deploy checks I run after every Cloudflare Pages build After hitting a sitemap _redirects bug and a Bluesky image race condition, I added three targeted smoke checks that catch silent Cloudfl… https://dev.to/morinaga/three-post-deploy-checks-i-run-after-every-cloudflare-pages-build-1off

Why I'm betting on AI-curated directories when Google AI Overviews answer the same queries The core counterargument to my three directory sites is Google's own AI Overviews. Here… https://dev.to/morinaga/why-im-betting-on-ai-curated-directories-when-google-ai-overviews-answer-the-same-queries-3g93

pgvector on a $5 Postgres instance handles semantic similarity across 100K+ pages without Pinecone or Weaviate. For indie-scale content sites, embedding generation is the real cost — not vector storage. Most indie devs are paying for the wrong layer. #webdev

Every programmatic page needs a self-referential canonical. Filter/facet URL variants should point to the base. GSC routinely shows 15–30% of large programmatic sites' indexed pages are canonicalized-away variants that leaked through. Boring to audit. Worth doing. #webdev

Three post-deploy checks I run after every Cloudflare Pages build After hitting a sitemap _redirects bug and a Bluesky image race condition, I added three targeted smoke checks that catch silent Cloudfl… https://dev.to/morinaga/three-post-deploy-checks-i-run-after-every-cloudflare-pages-build-56jm

Why I'm betting on AI-curated directories when Google AI Overviews answer the same queries The core counterargument to my three directory sites is Google's own AI Overviews. Here… https://dev.to/morinaga/why-im-betting-on-ai-curated-directories-when-google-ai-overviews-answer-the-same-queries-5a2p

Optimizing for LLM discoverability looks like classic on-page SEO: clear headings, factual specificity, explicit attribution. Maybe old SEO knowledge is more durable than the 'AI killed SEO' discourse suggests. Or both optimize for the same underlying readability signal. #ai

Astro 6 ships Live Content Collections: real-time data without a rebuild. After Cloudflare's acquisition, edge-native live data makes sense architecturally. But if you're pre-rendering to avoid runtime cost entirely, does live data clarify the model or complicate it? #webdev

Counterintuitive hypothesis: blanket robots.txt AI-crawler blocks might actively hurt ranking in 2026. If Perplexity and ChatGPT can't index your pages, you lose attribution in AI Overviews. The old SEO advice to block scrapers inverts when scrapers become referrers. Testing this? #ai

Three post-deploy checks I run after every Cloudflare Pages build After hitting a sitemap _redirects bug and a Bluesky image race condition, I added three targeted smoke checks that catch silent Cloudfl… https://dev.to/morinaga/three-post-deploy-checks-i-run-after-every-cloudflare-pages-build-4nkb

Why I'm betting on AI-curated directories when Google AI Overviews answer the same queries The core counterargument to my three directory sites is Google's own AI Overviews. Here… https://dev.to/morinaga/why-im-betting-on-ai-curated-directories-when-google-ai-overviews-answer-the-same-queries-2aop

A site built by an AI agent in 6 hours needs more maintenance in month 3 than one built by hand. The agent wrote to pass its own tests, not your mental model. Is vibe-coded infrastructure a compounding liability over time, or does time-to-ship make it worth the carry? #ai

JavaScript-rendered programmatic sites pay a hidden tax: Googlebot queues second-wave rendering 1–7 days after discovery. For 10K+ pages, SSG eliminates that queue entirely. The indexation speed gap between static and client-rendered output is still widely underestimated. #webdev