Why this is in the vault
Part 5 of Technically's "Software Eng for Vibe Coders" series — the scaling installment, teaching non-engineers what happens when a vibe-coded app goes viral and outgrows its hosting tier — kept alongside Parts 1-4 as the same plain-language reference set, and notable because the byline rotated to a third author mid-series.
The core argument
Continuing the Art Vandelay (potato-chip import/export) narrative, this piece picks up right after Part 4 deployed the app: a trade newsletter sends 4,000 users at once, the app slows from 2 seconds to 10 seconds to dead, and Art loses the spike because his Railway hobby-tier plan had a fixed ceiling (CPU, memory, bandwidth) nobody read closely at signup. The piece frames cloud capacity as a rented apartment: get the slice too small and traffic overwhelms it, too large and you pay for unused capacity ("right-sizing"). It draws the standard vertical-vs-horizontal scaling distinction — vertical is a bigger apartment (more CPU/RAM on the same server), horizontal is more apartments (replicas sharing the load) — and notes neither is universally better: vertical has a hard ceiling (no "7-bedroom apartments"), horizontal adds coordination overhead. The pre-cloud alternative was provisioning for peak demand and eating the idle cost the rest of the time; elastic scaling (Railway's pitch) grows and shrinks automatically, billing only actual utilization. Concretely: Railway's hobby tier caps at 48 vCPU/1GB RAM per service and 50 total services; upgrading to Pro removes the vertical ceiling (auto-scales to 1000 vCPU/1TB RAM, bills only for the blue line of actual use) but horizontal scaling (replica count, multi-region placement for latency and resiliency) stays a manual developer judgment call. The piece closes acknowledging Art's scaling was reactive — he found out his app was down because a friend called him — and teases Part 6 (observability) as the fix: dashboards and alerts instead of "listening for screams."
Mapping against Ray Data Co
The concrete gap: none of RDCO's live surfaces (Scribble Works on Cloudflare Pages/Workers, the credibility site, Squarely's backend) run on Railway, so the specific hobby-tier-ceiling story doesn't transfer literally — but the underlying failure mode does. Cloudflare Workers has its own hard ceilings (CPU-time-per-request, concurrent subrequests, KV/R2 rate limits) that are just as unread-at-signup as Art's Railway plan, and the credibility-site push (project_credibility_for_phdata_sales, explicitly aimed at demand generation and pipeline, not subs) is precisely the kind of surface that could get an unplanned traffic spike from a good post landing on HN or being shared by someone with reach. The piece's reactive-vs-proactive framing ("flying the plane by listening for screams") is the same gap as feedback_verify_static_asset_diffs_not_just_tests and the Scribble Works R2-binding failure (feedback_pages_r2_binding_breaks_function_publish): capacity and config limits that nobody checks until something breaks in production. Worth a five-minute pass: does anyone know Cloudflare's per-request CPU-time limit on the current Workers plan, or would a real spike be the first time that question gets asked?
Related
- [[2026-09-22-technically-deployment-vibe-coders]] — same series, Part 4, same sponsor
- [[2026-09-08-technically-auth-security-vibe-coders]] — same series, Part 3, same sponsor
- [[2026-08-25-technically-vibe-coders-databases-storage]] — same series, Part 2, same sponsor
- [[2026-08-11-technically-frontend-backend-vibe-coders]] — same series, Part 1, different author (Justin Gage)
- [[feedback_pages_r2_binding_breaks_function_publish]]
- [[feedback_verify_static_asset_diffs_not_just_tests]]
- [[project_credibility_for_phdata_sales]]
⚠️ Sponsorship
Railway sponsors this issue, confirmed via the same explicit "Thanks to our friends at Railway... for sponsoring this series" block plus referral-tagged railway.com/new CTA link. This is the fifth consecutive confirmed recurrence of the series-level Railway sponsorship (Part 1 Aug 11, Part 2 Aug 25, Part 3 Sep 8, Part 4 Sep 22, now Part 5 Oct 6) — pattern holds, no break. The byline, however, did break pattern: this installment is "a guest post by Will Raphaelson," a third author for the series after Justin Gage (Part 1) and David Krevitt (Parts 2-4). Note Raphaelson previously bylined the unrelated, unsponsored "What's going on with open-weight models?" issue (Aug 18) — same author, different series, no sponsor there, so his byline is not itself a sponsorship signal. Treat the elastic-scaling pitch and the Railway-specific tier numbers (hobby vs. Pro limits) as commercially motivated framing; the general vertical/horizontal scaling concepts are accurate and provider-neutral.