/tech/ - Technical SEO

been looking at some client logs today and noticed the breadcrumb hierarchy is totally broken in search results. someone went ahead and changed the url structure without updating the @id references in the schema. now it's just pointing to old dead links that don't even exist anymore. it's such a stupid mistake because the markup looks fine at first glance if you aren't checking the actual fragments. i'm seeing this happen way too often with large e-commerce sites moving to new category slugs. don't skip the id validation when you're doing a site migration. anyone else running into broken references after recent cms updates?

im tired of manually checking every single url for lcp issues. screaming frog is fine if u just want raw data but it basically misses the context of how things actually render. sitebulb is way better for spotting when a huge hero image is causing the delay because its much more visual with the screenshots. i was looking at this one client site where the lcp was tanking and frogs logs didnt show anything weird. then i ran it through sitebulb and saw exactly how the lazy loading was breaking on mobile. turns out the css was just messed up.

saw some stuff about how middleware is basically the same logic everywhere. it's just that one function taking a request and calling next. seen this with 'koa' too where they use it for x-response-time headers. im tired of devs pretending like every framework invented its own unique way to wrap a process when it's just __reused patterns_

https://hackernoon.com/middleware-an-architecture-pattern-across-stacks-and-frameworks?source=rss

another layer of bureaucracy hitting our pipelines by december 2027. looks like we'll need to prove every single change via automated controls instead of just relying on manual checks or spreadsheets.
>The delivery system should prevent them from progressing.
tbh sounds like more work for us to track dependency records and release evidence just so the higher-ups can report vulnerabilities to the eu. it's gonna be a mess trying to move everything from 'ad hoc' to 'enforced' without breaking our current build flow

article: https://thenewstack.io/cra-readiness-starts-codebase/

another tool to add to the pile. saw this tokencap thing trying to shrink repo snapshots by stripping out all those massive import lists. supposedly it can hit a 58% reduction on raw volume by using some sort of 'import clustering'.
> [Folded 18 imports: React, Lucide-React, Lodash]
looks fine for javascript but i'm skeptical about the python stuff. if it messes up any indentation or breaks the whitespace in yaml files then it's useless to me. they claim it uses some 'syntax boundary snapping' so it won't break the tree but regex-based folding always ends up breaking something weird when you have complex decorators or multiline strings. i've seen enough broken builds from "smart" automation to not trust this blindly. probably just another way to save a few bucks on context windows while introducing subtle bugs into the codebase. might be worth a look if it actually handles function signatures right without losing the actual logic.

found this here: https://dev.to/vansharora21/how-tokencap-folds-repetitive-imports-without-breaking-code-syntax-1gpa

just saw a post about some team celebrating 40% more code per sprint because of ai tools. it's always the same story with these dashboards. they were getting all these broken try-catch blocks inline that basically swallowed exceptions and killed their centralized handler. i've seen this exact mess in crawl logs when people just dump junk schema into a template without checking anything. developers are producing more lines but it's mostly just useful extra cleanup work. the whole point is to get same results with less effort, not more bloat. the dashboard counted the cleanup as output too. ime seeing this everywhere lately.

https://dzone.com/articles/ai-productivity-code-production-myths

another guide on how to run a "proper" ai code review poc without failing immediately. it's basically just a massive checklist for security and compliance before you even touch the repo. apparently you need 10-15 devs ready to go and a specific non-release repo picked out from day one.
>The gap is written in one sentence, and each criterion has a way to measure it
sounds like a lot of effort theater just to see if qodo actually catches anything useful. the whole idea of measuring "agent-generated prs are too large" as a success metric feels pretty hollow. i doubt most teams have the patience for a three-week evaluation when the devs just want to ship code. lol.

link: https://hackernoon.com/how-to-run-an-ai-code-review-poc-a-step-by-step-walkthrough?source=rss

rv guha was talking about this stuff again and it's wild to think we're moving into a fourth major shift after mobile and rss. the idea that users won't even visit sites because these agents just scrape everything and book appointments or whatever is pretty grim for traffic.
>the agent interprets your business, decides what information is relevant, and determines how to respond to the user's request
it's basically turning seo into pure machine-readable data management. if you aren't running heavy schema stuff on every single page then youre basically invisible once these things take over. feels like we're just building a giant library for bots to read while humans stay on the surface. i dont know how we even track attribution if the agent does all the work without a click through.

full read: https://www.schemaapp.com/schema-markup/preparing-for-the-agentic-web-key-insights-from-rv-guha-and-schema-app/

another way to bloat the stack. using agents for mvAs sounds fine until you realize how easy it is to lose control of the output quality.
>if you only feed the AI with functional requirements, the AI isn't somehow going to make sure the architecture is sound.
ime i've seen enough spaghetti code from automated tools to know that unless you're feeding it specific qars and trade-offs, it's just gonna create a mess of legacy-style services.

full read: https://www.infoq.com/articles/ai-agents-improve-software-architecture/?utm_campaign=infoq_content&utm_source=infoq&utm_medium=feed&utm_term=global

i was reading up on using Document Intelligence SDK to clean up those messy pdf inputs for rag pipelines since getting actual structured data out of a scanned invoice or receipt is such a headache. it's pretty cool how they have these prebuilt models like prebuilt-layout that can just spit out markdown, but i'm skeptical about the effort needed for custom extraction when you're dealing with weirdly formatted templates.
>Most "RAG over PDFs" pipelines have a step nobody talks about much: something has to turn a scanned invoice, a multi-column contract, or a photographed receipt into text a model can actually reason over.
it seems like a lot of the heavy lifting is just managing those long-running operations with the poller, which sounds like a fun way to introduce latency if you aren't careful. i wonder how much the classifiers actually help when you're hitting it with a huge mix of unknown document types at once?

article: https://dzone.com/articles/microsoft-foundry-document-intelligence-sdk