January 2025 Web Dev Highlights: Trademark Clashes, JavaScript Updates, and Doom in a PDF | The Digest #4

Why Fora Soft still publishes this digest a year later

January 2025 is now more than a year behind us, but almost every storyline that started in that single month still pays and still bites. The Deno vs Oracle JavaScript trademark case is live discovery. The Google-requires-JavaScript policy reshaped every SEO vendor’s pricing. DeepSeek-R1 forced OpenAI and Anthropic into a public pricing retreat. Vite’s Rust bundler shipped as default. Every one of those January stories is still on the whiteboard of our custom software development engagements.

We’re republishing the digest as a retrospective because most January-2025 news articles read, in April 2026, as obvious in hindsight. This one calls out the calls that aged well and the ones that didn’t. If you’re a product founder or CTO scoping a web build today, read this as a decision checklist — each section ends with the practical implication for the stack, the budget, or the vendor you pick.

For context on how we apply this kind of lens inside client engagements, see our spec-driven agentic engineering approach and the AppyBee booking platform case where we run the React / React Native / PHP / Node microservices stack at real scale.

JavaScript tools and frameworks — what shipped

The JS story in January 2025 is three releases — Deno 2 (late 2024, rolling into production through Q1 2025), the Vite + Rolldown beta, and the Ladle React component tool. Each fixes a different bottleneck.

Deno 2 — Node compatibility as the headline

Deno 2 shipped on October 9, 2024 and its January 2025 adoption curve was the real story. The release dropped the anti-Node purity that had capped Deno’s addressable market. Deno 2 understands package.json, node_modules, npm workspaces, bundles @types/node v22, adds a Node-style process global, and made this the first Deno release with formal LTS guarantees.

Practical implication: Deno 2 is now a reasonable default for new server-side TypeScript projects that don’t depend on native Node add-ons. Migrating an existing Express app mid-flight still isn’t worth it.

Vite + Rolldown — Rust under the hood

The January 2025 news was the Vite team’s announcement that Rolldown, their Rust-based bundler, was entering beta. The full timeline matters in 2026 terms: Rolldown-Vite became a technical preview on May 30, 2025, Vite 7 shipped with Rolldown support on June 24, 2025, and Vite 8 Beta made Rolldown the default bundler in late 2025 — unifying esbuild and Rollup into a single Rust tool with plugin back-compat.

Real numbers: GitLab moved their web frontend to Rolldown and reported a 2.5-minute build dropping to 40 seconds, with roughly 100× memory reduction. For large monorepos that figure matters more than any single framework choice.

Ladle — the quiet Storybook alternative that survived

Ladle hit the digest in January 2025 as a leaner, faster React component workbench. It held up. Ladle cold-starts in ~1.2s against Storybook 10’s ~8s; Uber runs 335+ projects and ~15,896 stories on it internally. The catch: no full Storybook addon ecosystem — pick Ladle when you want fast visual regression and speed over plugin breadth.

What shipped in PHP — property hooks are the headline

PHP 8.4 — property hooks, array_find, asymmetric visibility

PHP 8.4 released November 21, 2024 and January 2025 was the month teams actually started upgrading. The feature everyone will remember is property hooks — custom get/set logic lives on the property itself, so the boilerplate getter/setter pair vanishes. Combined with asymmetric visibility (public read / private write), PHP finally reaches feature parity with modern C# and Kotlin on data-class ergonomics.

Four new array functions — array_find(), array_find_key(), array_all(), array_any() — replace about half the foreach loops in a typical Laravel codebase.

Phuzz — fuzzing PHP web apps

Phuzz is a research-grade fuzzing framework for PHP web applications, presented at 38C3. It introduces precision beyond black-box scanners by tracing request parameters against sink invocations. Worth trying for any PHP app handling payments, PHI, or auth.

Voyager vulnerabilities — a cautionary tale

Multiple critical vulnerabilities in the PHP Voyager admin package were disclosed in January 2025 — remote code execution, arbitrary file upload, XSS — and left unpatched for months despite responsible disclosure. If your team uses Voyager, the rational move is still to replace it or run it behind a hardened reverse proxy with strict role ACLs.

AI’s big moments — DeepSeek, Gemini, Imagen 3

DeepSeek-R1 — the 95% price cut heard around the world

DeepSeek-R1 released January 20, 2025. It scored 79.8% on AIME 2024 against OpenAI o1’s 79.2% and priced at $0.55 per million input tokens and $2.19 per million output tokens — versus o1 at $15 and $60. That’s a ~95% reduction on output cost for comparable reasoning quality.

The open-weights release forced every frontier lab to cut prices within 90 days and reset every AI product startup’s unit economics. If you’re pricing a SaaS around LLM inference, recheck the line item — the 2024-era model mix is now 3–10× too expensive.

Google Gemini 2.0, Imagen 3, Project Mariner

Gemini 2.0 Flash went GA in early 2025 with strong agentic-workload tool use. Imagen 3 kept the #1 slot on the Image Generation Leaderboard for most of the year. Project Mariner — Google’s browser-control agent — stayed early-access for Google AI Ultra subscribers and steadily added tab-control, form-fill and long-horizon browsing.

The broader arc: Gemini 3.0 launched publicly on November 18, 2025 hitting a 1501 Elo score and is now Google’s default model across consumer Search for 2B+ users. For product teams, the practical consequence is that the “one model, one vendor” era ended in January 2025 and the sensible default is now multi-model routing.

The laws of the web — Oracle vs Deno, Google vs scrapers

Oracle faces Deno Land over the JavaScript trademark

Deno Land filed the cancellation petition on November 22, 2024 at the USPTO. The petition runs three counts: genericness (“JavaScript” is the language name, not a trademarkable product), abandonment (Oracle has not commercially used the mark on a JavaScript product), and fraud on the USPTO (Oracle allegedly included a third-party Node.js screenshot in its 2019 renewal).

Oracle filed a motion to dismiss on February 4, 2025. The TTAB granted partial dismissal of the fraud count on June 18, 2025 but let the genericness and abandonment claims proceed. Discovery opened September 6, 2025 and is expected to run through early 2027.

Practical implication for product teams: nothing changes in 2026 — you can still say “JavaScript” on your careers page. If Deno wins the genericness claim, expect a minor uptick in products using “JavaScript” in branded names; if Oracle wins, everything stays as-is.

Google starts requiring JavaScript for Search

Announced January 17, 2025: Google began requiring JavaScript execution to render search results, deliberately blocking curl-style scraping used by a generation of SEO tools. Rank trackers, keyword-volume estimators and SERP feature detectors had to switch to headless browsers (Puppeteer, Playwright) to keep working.

Concrete consequences in 2026: SEO tooling bills went up ~15–40% industry-wide as vendors passed on the cost of headless-browser farms. Free rank-tracker browser extensions largely died. Google’s own JavaScript-SEO docs were updated December 18, 2025 and March 2025 to cover non-200 status codes and AI-Mode results.

Practical implication: if you’re building anything that scrapes SERPs, budget for the headless-browser overhead up front. For regular site SEO, server-side rendering (SSR) or hybrid SSG is now the table stakes — client-only React apps with minimal pre-render will lose ground to competitors that render on the server.

The geeky side — Doom in a PDF

ading2210 released DoomPDF around January 15, 2025. The trick: Emscripten compiles Doom to asm.js, the PDF’s JavaScript engine runs it, output is ASCII-rendered into PDF text fields in a 6-color monochrome palette, keyboard input comes from focusable form fields, mouse input from button widgets. It runs at ~80ms per frame on Chromium-based PDF viewers only.

This is a what can we actually do in a document format demonstration, not a shipping feature — but it’s a useful reminder that PDFs still run arbitrary JavaScript in Adobe Acrobat and Chromium. If your product lets users upload PDFs, sandbox them properly.

Web development’s next big moves — trends that aged well

AI’s impact on web development — not hype anymore

The January 2025 prediction that AI would transform web dev workflows was already under-sold. By Q4 2025, agentic coding tools were shipping production code at ~40% of the output volume of senior engineers on well-scoped tasks, and spec-driven methodologies were the default in fast-shipping teams. We wrote the full playbook for our own delivery in this piece on spec-driven agentic engineering.

HTML6 and CSS5 — hype versus reality

The original digest mentioned HTML6 and CSS5 as on the horizon. In 2026 the honest answer is: neither exists as a formal version. HTML is maintained as a Living Standard by WHATWG with incremental additions (last update April 22, 2026); CSS is captured in the CSS Snapshot 2026, a W3C Group Note, not an endorsed standard.

What actually shipped in the meantime: CSS nesting, :has(), container queries, subgrid, @scope, view transitions, anchor positioning, relative color syntax, and native CSS layers. For HTML: popover API, <dialog> stabilized cross-browser, declarative shadow DOM.

Don’t wait for “HTML6.” Just adopt the Living Standard features that already have 90%+ browser support — that list is longer than most teams realize.

PWAs, voice search and serverless — the hard numbers

Progressive Web Apps. The PWA market is ~$2.47B in 2025 and forecast to hit ~$3.14B in 2026 at ~30% CAGR. Desktop PWA installations are up roughly 400% since 2021. PWAs are the right default for anything that needs offline, low-friction installability, or lightweight mobile distribution without App Store friction.

Voice search. A majority of users now prefer voice over typing for fast-answer queries. If your site makes money from local or informational queries, schema markup + conversational-phrasing optimization is no longer optional.

Serverless. The serverless market sits near $17.78B in 2026 on ~22.5% annual growth. For bursty B2C traffic or unpredictable tenant scaling (booking platforms, chat apps, AI inference), serverless compute is the cost-optimal tier — the inflection point versus dedicated VMs lands around 35–50% steady-state utilization.

A decision framework — how to act on a web-dev news cycle in 5 questions

1. Does this news change cost by >20%? DeepSeek-R1 did (inference cost). Rolldown did (CI cost). Most framework-of-the-week stories don’t.

2. Does this news change risk by >20%? Google requiring JavaScript changed SEO risk for scraper-dependent businesses. The Voyager CVEs changed risk for teams running it unpatched.

3. Is the feature available in your current stack? Property hooks require PHP 8.4. array_find() requires PHP 8.4. If you’re on 8.2 or earlier, the feature list doesn’t exist for you until you upgrade.

4. What’s the migration cost? Deno 2 for greenfield: free. Deno 2 for an existing Node codebase: 2–6 weeks depending on native add-on count. Vite 8/Rolldown in an existing Vite monorepo: hours.

5. Will your customers notice? Build-time wins never reach end-users. Inference-cost wins can. SSR-for-SEO wins can. Prioritize accordingly.

Five pitfalls when acting on web-dev news

1. Chasing the new bundler mid-sprint. Rolldown is default in Vite 8 — upgrade with the next major framework bump, not during a feature sprint. Migrations done under schedule pressure ship bugs.

2. Assuming DeepSeek is always the right pick. DeepSeek-R1 beat o1 on math benchmarks at 27× lower cost, but on multilingual conversational tasks and safety-sensitive outputs, the routing question is empirical. Benchmark on your own prompts.

3. Ignoring the SSR shift. Google’s JavaScript requirement is about blocking scrapers, but the parallel message is that server-side rendering is no longer optional for SEO-reliant products. Client-only SPAs keep losing ground.

4. Upgrading PHP without running the deprecations report. PHP 8.4 removes some legacy behaviours and deprecates others. Run phan or PHPStan at level 8+ before the upgrade PR.

5. Taking trend articles as roadmap input. “HTML6 and CSS5 on the horizon” wasn’t true in January 2025 and isn’t true now. Cross-check every trend article against the actual spec status before it hits your RFC doc.

KPIs — what to measure on a web build in 2026

Quality KPIs. Lighthouse performance ≥ 85 on mobile. INP < 200ms at the 75th percentile. Escape defects < 2% per release. PR-to-prod lead time < 48 hours.

Business KPIs. LLM cost per active user < $0.05/day for conversational features after multi-model routing. PWA install rate ≥ 8% on marketing traffic that sees the install prompt. Time-to-first-paid-action < 5 minutes.

Reliability KPIs. Monthly uptime ≥ 99.9% on user-facing paths. Payment gateway success ≥ 98% per provider. CI build time < 120s for incremental, < 8 min for full rebuild (Rolldown makes both achievable).

When you should not act on a web dev news story

Not every news item is a call to action. A bundler shipping in beta is not a migration trigger; wait for the stable default release (Rolldown crossed that line with Vite 8). A framework’s GitHub star count is not a production signal; engineering hiring rate and issue-close velocity are.

The most expensive mistake in 2025 was teams pivoting backend stack mid-build because the timeline article ranked their framework 6th. Keep the stack you picked for the product reasons that picked it; only migrate when the cost or risk threshold is clearly breached.

What we actually changed in production after reading January 2025

Rolldown-Vite by default on every new monorepo we opened after Vite 8 stabilized.

Multi-model LLM routing with DeepSeek for bulk reasoning, Anthropic Claude for long-context and nuanced output, Google Gemini for vision-heavy tasks, OpenAI for specific tool-use patterns. Fallback routing on rate limits.

SSR-first React on any SEO-facing page. Next.js, Remix, or our own RSC setup depending on hosting constraints.

PHP 8.4 minimum on new backend projects; mandatory property-hook refactor on any legacy PHP codebase we take over.

Headless-browser budget added to the line-item cost of any product we build that depends on live SERP data.

FAQ

What to Read Next

Ready to ship against the 2026 stack, not the 2024 one?

January 2025 looks in hindsight like the month the web ecosystem reshuffled. Frontier AI went 95% cheaper overnight. The bundler went Rust by default. Google quietly rewrote the SEO industry’s cost structure. PHP finally caught up to modern data-class ergonomics. Deno picked a fight with Oracle that’s still in court.

If your product was scoped before April 2025, half of that list is a concrete line item you’re overpaying on. Book a 30-minute call and we’ll walk through the audit — inference routing, SSR-for-SEO, bundler cost, LLM tier — and give you a short list of changes ranked by payback period.