Build1 publisher3 min readPublished
Keyword-to-URL mapping has to land before the agent writes the first React component
JetBrains puts weekly AI coding agent use at 90% of professional developers in mid-2026. A dev.to post argues the sites those agents scaffold collapse onto one section order and one URL, and treats page architecture as work that has to precede generation.
The Engineer · Build desk
What happened
- The same research put Claude Code at around 39% of professional developers worldwide and OpenAI Codex at 16% over the same period.
- A dev.to post argues that agent-built sites repeat the same hero section, cards, gradients, navigation, CTA and FAQ, and sometimes the same single-page structure.
- The post proposes running business strategy, design direction and SEO architecture before AI-assisted development, instead of letting the agent decide the design and then write the code.
Compiled by The EngineerSomething wrong?How this is made
Why it matters
- decision Route count and the keyword-to-URL map have to be settled before anyone types the prompt, because after generation the agent's default scaffold is the thing you are arguing with.
- cost Whoever inherits a shipped single-page build pays the retrofit: redirects, rewritten internal links, and per-page titles, metadata and schema that did not exist when one URL held everything.
- contradiction The survey evidence is about how often developers call an agent, while the argument needs evidence about what agents emit, and that half rests on one practitioner's pattern list.
A single-route build has one document head. One title element, one H1, one canonical, one meta description, one schema block. The dev.to post making this case says agent output often puts everything on one URL, which it calls acceptable for a simple landing page and restrictive once a business has multiple services, products, locations or search intents to address [7]. Ten service keywords then compete inside the same document, and an internal link has nowhere to point.
Splitting that apart afterwards is route work. Each new page needs its own topic, search intent, title, H1, content, internal links, metadata, schema and conversion goal [9]. The post's ordering puts the keyword-to-URL map first, so that a page exists because a query exists, and one homepage is not trying to rank for 30 unrelated keywords [8].
It stages the failure as a line from the site owner: "The website is ready. Now let's do SEO." [14] The post's position is that SEO belongs in the architecture from the start [15].
The adoption figures come from JetBrains' 2026 Developer Ecosystem Survey, and they measure how often developers call an agent at work. The gap between weekly and daily use is 22 points, so roughly one professional developer in five is invoking an agent in a given week without invoking it every day [16]. For the headline number to describe your team, your team has to resemble the pool that answered that survey between May and July 2026 [1].
Tool concentration is what makes the homogeneity argument plausible at all. In the same window JetBrains had the most-used agent at about 2.4 times the adoption of the second [17]. Whatever the leading tool emits when asked for a modern company website is therefore the most common starting point in the population, defaults included.
The evidence for convergence is thinner than the adoption evidence. The post lists recurring features, the same hero section, cards, gradients, navigation, CTA and FAQ, and sometimes the same single-page structure [5]. It does not say how many generated sites it looked at. On the section order it is careful: there is nothing inherently wrong with Hero, About, Services, Why Choose Us, Testimonials, FAQ, Contact, and the problem is applying it to every business without considering search intent, business type, customer journey, keywords, services, locations, content depth, internal linking or conversion goals [19].
The prescribed fix is cheap in the sense that it is all upstream: business research, competitor research, keyword research, search intent analysis, customer journey, website architecture and content requirements before the first React component [12]. Then hand the agent a design system, defining brand colors, typography, button and card style, border radius, spacing, navigation, image style, animation rules, icon style and mobile behaviour, and ask it to implement that system [11]. The worked example for an SEO agency is ten routes, including a city page at /seo-expert-jaipur/ and a separate /case-studies/ [13].
What to watch
- Whether JetBrains' next survey reports anything about what agents emit, such as route counts or template reuse, and not only how often developers invoke them.
- A measured sample of agent-generated sites, with section orders and URL counts compared, would turn the homogeneity claim into a number.
- Whether coding agents change their default scaffold to multi-route output with per-page metadata, removing most of the retrofit work.