We tested four AI design tools on a real client brief. Only one shipped.
Same brief, four tools: Figma AI, v0, Galileo, Uizard. Which got us to a usable mockup the fastest, and which fell apart the moment the client had feedback.
Search Console is showing fewer impressions. The reasons aren't what you think.
Sites across our portfolio are seeing impression drops in 2026, but rankings are holding. AI Overviews are counted differently, deduplication tightened, and Bing is taking share.
We rewrote a client's onboarding after one usability session. Conversions doubled.
One 60-minute moderated session. Three users. Three changes shipped the same week. The numbers that came back surprised even us.
Edge computing for small business sites: when it's worth it, when it isn't.
Cloudflare Workers and Vercel Edge are marketed like everyone should use them. For most small business sites, the answer is no. The cases where it actually pays off.
AI coding assistants in production: what we let them do, what we don't.
Cursor, Copilot, Claude Code are now part of the day-to-day. Our written policy on where they're trusted, where they're audited, and where they're banned.
Microinteractions: where they earn their keep and where they just annoy.
The 'delightful microinteractions' era oversold itself. Sometimes a button click should be a button click. Three cases where animation helped, three where it slowed users down.
Remote unmoderated testing tools, compared head to head.
Useberry, Maze, UserTesting, PlaybookUX. Pricing, panel quality, what features actually move the work forward — based on six months of using all four.
The local SEO playbook for service businesses in 2026.
Google Business Profile, reviews, citations are still the holy trinity. What changed in 2026: AI Overviews surface local businesses differently, video reviews carry more weight, voice search keeps growing.
Web Components are finally good. What changed and what to use them for.
Web Components were clunky for a decade. Declarative shadow DOM and scoped CSS fixed enough of that to make them shippable. Where they now beat framework components.
RAG vs fine-tuning for small business use cases: a practical comparison.
Both sound technical, both have evangelists. For nearly every small-business workload we've seen, RAG wins on setup cost, maintenance, and drift.
Designing dashboards with color theory built for data, not landing pages.
Most designers learned color from landing-page work. Data viz needs different rules: perceptual uniformity, colorblind-safe palettes, distinguishing eight or more categories without losing meaning.
Core Web Vitals updates Google is enforcing this year (and what to fix first).
INP replaced FID for good. LCP thresholds tightened. Measuring with real user data matters more than ever. The fixes that paid off on our sites.
A/B test mistakes that cost us a client $40k last quarter.
Peeking at running tests, undersized samples, conflating clicks with revenue. The mistakes that look fine on the dashboard and bleed money in the bank.
Why we're moving small-business sites back to static generation.
Most small-business sites change once a quarter. Static is faster, cheaper, more reliable, and easier to back up than the SSR stack we kept reaching for.
Prompt injection in production: we got hit, here's what we learned.
A client's AI customer service agent leaked internal pricing through a cleverly worded support request. What broke, what we changed, what we now ship by default.
Accessibility audits are landing more clients for us than referrals.
ADA/WCAG lawsuits are way up. We started offering free 15-minute accessibility audits as outreach. Conversion to paid engagement is higher than anything else we've tried.
E-E-A-T after the latest helpful content update: what shifted, what didn't.
Google's spring update tightened the screws on AI-flavored content and author signals. The changes that moved rankings on real sites — and the changes that didn't.
Heatmaps vs session replays: which one actually changes how you build?
Most teams pay for Hotjar or FullStory and never act on the data. Heatmaps show patterns; replays show stories. Different uses; different price points.
CSS in 2026: container queries paid off, View Transitions delivered, subgrid is still confusing.
Three years after the big modern-CSS push, where we actually landed. Honest assessment of what we use daily and what we still avoid.
Local LLMs hit a tipping point this spring. Should you self-host?
Llama 4 and Qwen are now production-competent. The economics finally make self-hosting viable for some workloads, not for others. The breakeven math.
Programmatic SEO when AI writes everything: a defensible strategy.
AI content is everywhere and Google's helpful content update keeps tightening. Programmatic SEO still works — if the content rests on real data, not text variations.
The hamburger menu isn't dead. The way most sites use it is.
Hamburger-bad takes ignore that desktop nav is a different beast from mobile. The real issue is what's hidden inside, not the icon itself.
We replaced our design QA with AI vision models for a week. Here's what broke.
Five days of running Claude with vision against design mockups for spec compliance. Where it nailed it, where it missed obvious things.
Schema.org changes we missed in Q1 2026 (and the ones that actually matter).
Schema.org pushed a quiet update in Q1. Three changes matter for small business sites; the rest is noise. Here's what to add to your markup this month.
Server components, one year later: a small-agency report card.
After a year of shipping React Server Components on real projects, an honest assessment: where they paid off, where they made things worse.
AI agents replacing SaaS workflows: what actually works, six months in.
We replaced three SaaS subscriptions with AI agents this year. Two worked. One failed badly. Here's what separated them.
Dark mode fatigue is real. Most B2B products should ship light-only and stop apologizing.
Dark mode looks great in marketing screenshots. For data-heavy B2B work it makes charts harder to read and contrast harder to control. Why we now recommend light-first.
Five-user testing still beats every fancy research method we've paid for.
Jakob Nielsen's old finding holds up: five users surface most usability problems. Why we still use it instead of the panel studies clients keep asking for.
Google's AI Overviews are eating click-through rates. What we tell clients now.
Six months of watching Google's AI Overviews chew up informational-query traffic. The content strategy shift that's still working for our clients.
htmx is winning small-team frontends. We're not sorry to see React go.
After fourteen production sites on htmx, here's what we learned about replacing React on small-business projects: build pipelines, performance, where htmx breaks down, and the migration cost.