Atlas expedition

Vibe-coding tool stack for a 2-3 hour workshop on shared wifi

Lovable as the default with Bolt.new as the backup, hosted hotspot for wifi, $0–$50 host spend for 20 attendees, and signup done the night before.

76 sources ~11 min read #56 vibe-coding · workshop · lovable · bolt · v0 · replit · claude-artifacts

Decision. Pick Lovable as the primary tool — chat-only UI, server-side rendering, one-click publish to *.lovable.app, and the only vendor with a real K-12 / no-account classroom track [2] [7] [38] [55]. Keep Bolt.new warm as the backup (faster cold start, no account, but WebContainers hate locked-down laptops and Lovable has logged 314+ outages since May 2025) [4] [5] [31] [71]. Have attendees sign up 24–48 h before the room, off the venue wifi, and bring a phone-hotspot kit. Host budget: ~$0 if you stretch free tiers, ~$50 if you pay for one shared Pro seat as fallback; the per-attendee marginal cost is zero [27] [76].

Layer Primary Backup Why
Build tool Lovable (free) Bolt.new (free) Lovable hides the file tree, ships SSR HTML (low bandwidth); Bolt boots in seconds and needs no account [2] [3] [4] [38]
Identity Continue-with-Google Pre-provisioned host account Google SSO sidesteps Claude’s SMS verification gate and ChatGPT’s phone-country check [42] [45]
Hosting Built-in *.lovable.app *.bolt.host Both are one-click on free tiers; no Vercel / Netlify / Cloudflare account required at workshop time [55] [58]
Sharing QR code on screen Short link Phone-camera scannable, works for grandma in the room [67] [68]
Network plan A Venue wifi Phone hotspot 100+ laptops with previews + npm + Git saturate venue APs; a single instructor hotspot is the canonical fallback [39] [76]
Network plan B Pre-warmed accounts Pre-built starter repo Shared NAT triggers Cloudflare Turnstile escalations and “too many sessions from 1 IP” rate limits [50] [69]

1. Tool shortlist — what survives contact with a non-coder

The 2026 vibe-coding field has six tools you can plausibly hand to a stranger. They are not equivalent.

Tool Layman-friendly Time to first working app File tree in face? Notes
Lovable ~60 min curriculum proven for grade 6-12 [7] Polished UI on first prompt; auth+DB+hosting auto-wired [2]
Bolt.new 2-4 min landing page, 8-12 min to-do app [5] Lowest signup friction in the field; demo-grade output [4]
v0 Works, but assumes Vercel mental model [1] UI-first, weaker on full-stack [1]
Replit Agent IDE overwhelms non-coders [3] Free tier near-unusable (10 checkpoints/mo, 512 MB RAM, sleep at 5 min) [17]
Cursor Reviewers say “start somewhere else” [6] Full IDE; useful later, wrong first tool [6]
Claude Artifacts Runs in chat, ~70 % to shippable [9] Public share lands viewers on a Claude login wall [61]

In a head-to-head non-developer test, Lovable and v0 both shipped a working photo-booth app from a single prompt while Bolt.new failed across six attempts — the camera feature never loaded [1]. One facilitator who has run multiple workshops moved from Bolt to v0 to Lovable as her default classroom tool — non-coder friction was the driver [78]. Lovable’s own docs concede Bolt has the lowest barrier for a complete novice (no account, no config) — but Bolt’s output is also the most demo-grade [4].

The trade. Lovable produces things that look done; Bolt produces things that exist. For a 2-3 h workshop where the deliverable is a working URL to show a friend, Lovable wins; for a one-prompt-and-done warmup, Bolt is the quickest path.

2. Free-tier math — does this break the bank?

Short answer: no, if you plan it. Long answer:

Tool Free-tier limit (May 2026) Reset Pro price Burnable in 3 h?
Lovable 5 credits/day, 30/month [11] Daily 00:00 UTC $25/mo, 100 cr [12] ⚠ Yes; one landing page = ~2 cr
Bolt.new 1M tokens/month, 300K/day hard cap [15] 1st of month, no rollover [15] $25/mo [16] ⚠ Possible; debug loops eat tokens [28]
v0 7 messages/day (variable), $5 monthly credits [13] [14] Daily / monthly $20/mo ✓ Yes; 7 msgs is one bug fix
Replit Agent 10 checkpoints/month, 512 MB RAM, repls sleep 5 min [17] Monthly $25/mo Core [18] ✓ Yes; effectively unusable
Cursor 2,000 completions + 50 slow premium reqs/month [20] Monthly $20/mo ✗ Generous for one session
Claude.ai ~15-40 messages per rolling 5 h [22] Rolling 5 h $20/mo Pro ⚠ Tight on heavy iteration
ChatGPT 10 GPT-5.5 msgs / 5 h, then mini [23] Rolling 5 h $20/mo Plus ⚠ Tight; auto-falls back

Host-pays math. A solo vibe-coder spends $100-250/month unnoticed [27]. For 20 attendees on Pro for one month, expect ~$500 (e.g. 20 × $25 Lovable). No vendor publishes a workshop SKU — group rates are extrapolated from per-seat. Cheaper paths exist:

  • Students (verified .edu). Replit Core: $10/mo for six months [26]. Cursor Pro: free for one year (~$240 value) [21]. Verified educators get Replit free with student credits [25].
  • Multi-account chaining. Stacking free credits across alt Google accounts violates OpenAI ToS §4(e) explicitly [24] and equivalent clauses on the others. Don’t.
  • Host shares one Pro seat. $25 covers the demo, the room watches the screen, attendees stay on free. Marginal per-head cost: $0.

3. Shared-wifi survivability

The tools split into three bandwidth tiers. Pick by your venue.

Tool First-load weight After boot Survives wifi drop?
Lovable Server-rendered HTML chunks (TanStack SSR) [38] Thin client, retries are cheap ✓ Each request is independent
v0 Server-rendered, similar to Lovable Light ✓ Same profile
Cursor Streams over HTTP/2 to *.cursor.sh [33] Persistent connection ✗ Chronic weekly disconnects, IDE restart often needed [34]
Claude Code (CLI) Tokens to Anthropic API; 5.5× lighter than Cursor [36] Slim ⚠ Silently hangs spinner on drop, no error [35] ⭐ 126k
Bolt.new <1 MB WASM runtime [29], + npm dependencies Runs offline once booted [30] ⚠ 30 s hard boot timeout [32] ⭐ 16k
Replit Full cloud IDE, continuous sync Heavy ✗ Lose access during outages; tethered mode burns mobile data [37]
GitHub Codespaces Remote VM Server-side ✓ Preserves session across disconnects [40]

The Bolt trap. Bolt’s troubleshooting page admits that ad-blockers, VPNs, and non-Chromium browsers — all common on locked-down work/conference laptops — break the cross-origin-isolation handshake outright [31]. On a slow venue wifi the 30-second boot timeout will fire before the WebContainer initialises, with no partial-load fallback [32] ⭐ 16k. If you bet on Bolt, screen-test laptops before the room.

The venue-wifi reality. 100+ laptops doing previews, npm, Git pushes routinely saturate APs not provisioned for power users; backup hotspot kits are standard event-ops advice now [39]. A teacher’s mobile hotspot used as a wifi-failure fallback is a pattern borrowed from VIPKid-era ESL workshops and still works [76].

4. Signup friction (the silent workshop killer)

Most attendee minutes lost in a vibe-coding workshop are lost at the signup screen, not the prompt.

Tool Account required? Phone SMS? Age gate Credit card? Workshop-day verdict
Bolt.new ✗ (free tier) None at signup [5] ✓ Walk-in friendly
Lovable ✗ (Google SSO) None visible ✓ Walk-in friendly
v0 / Vercel None at signup [47] ✓ Google/GitHub/Apple/GitLab/Bitbucket SSO [46]
Cursor None at signup ✗ Hobby [48] ⚠ Student Pro needs 3-7 day SheerID review [49]
Claude.ai ✓ (email signup) [75] 18+ enforced [41] ⚠ Use Google SSO to skip SMS [42]; VoIP rejected [43]
ChatGPT 13+, parental for 13-17 [44] ⚠ IP & phone-country dual-blocked in ~20 countries [45]

Shared-NAT hazard. Cloudflare Turnstile escalates challenges on shared-IP wifi; the most common silent failure is device clock skew rejecting the token [50] [51]. On a school NAT, OpenAI returns “too many sessions from 1 IP” and offers no fast support channel [69]. Schools also block ChatGPT at the firewall regardless of the ToS [53].

The 24-48 h pre-flight. Established AI-coding curricula explicitly tell learners to create their accounts 24-48 hours before the room, off the venue wifi, citing 30+ minute downloads and trial-clock concerns [52]. Do this — it’s the single highest-yield mitigation.

Institutional escape hatch. Universities like Harvard front a pre-provisioned “AI Sandbox” that wraps GPT/Claude/Gemini behind SSO, removing every per-student phone/age/payment gate [54]. If your venue has anything similar, use it.

5. Deploy & share — fastest path to a public URL

Every recommended tool has a built-in publish button on free. Use it. Do not route attendees through Vercel/Netlify dashboards on workshop day.

Tool Publish click count URL pattern Custom domain on free? Permanence
Lovable 1 [name].lovable.app [55] [56] Persistent
Bolt.new 1 *.bolt.host [58] Persistent + “Claim URL” path to your own Netlify [59]
v0 1 Vercel production URL [57] ✗ Hobby Persistent; Hobby is non-commercial only [65]
Replit Starter 1 *.replit.app ⚠ Repls sleep at 5 min, only 10 Agent checkpoints/mo [17]
Claude Artifacts 1 claude.ai/public/artifacts/<id> [60] ⚠ Viewers see Claude login wall [61]
ChatGPT Canvas 1 chatgpt.com/canvas/shared/<id> [62] Persistent

Generic hosts as fallback when you need to migrate the build off the generator:

Host Free-tier limit Source
Cloudflare Pages 500 builds/mo, unlimited bandwidth, 20-min build cap [63]
Netlify 300 credits/mo (15/deploy, 20/GB), pauses on overrun [64]
Vercel Hobby 100 GB bandwidth, non-commercial only [65]
GitHub Pages 1 GB site, 100 GB/mo soft, public repos + static only [66]

Cloudflare Pages wins for workshop fallback — unlimited bandwidth means a viral demo can’t bankrupt you [63].

Sharing in the room. Generate one QR per URL with QRStuff or Canva — both free, static, unlimited scans, native iOS/Android camera readable [67] [68]. Display on the projector; attendees scan with their own phones. Don’t email links — captive-portal browsers eat them.

6. Failure modes seen in the wild + mitigations

These keep showing up across multiple host write-ups [73] [78]:

Failure Mitigation
Venue wifi saturates / drops [39] Instructor phone hotspot, cellular plan B [76]
Shared NAT → “too many sessions from 1 IP” [69] Pre-warm accounts off-venue [52]; split between two tools
Lovable outage (314+ logged since May 2025, recent incidents up to 10 h+) [71] Bolt.new ready as backup; pre-built starter repo
Cursor SheerID “Verification Limit Exceeded” when many .edu signups [74] Don’t depend on Cursor day-of; queue verifications over a week
Claude SMS verification rejects VoIP / no-phone attendees [43] Continue-with-Google bypass [42]; pre-provisioned host accounts
Free-tier credit exhaustion mid-build [2] Constrain scope to ≤1-2 prompts on Lovable; switch to Bolt for iteration
Non-coder hits paywall on first prompt (Cursor, Replit) [70] Don’t pick those for first-touch
Demo collapses under real testing (no planning step) [77] Force 10 min of “what does the app do” before the first prompt
Beautiful prototype, broken on second prompt [72] Frequent commits + a forked starter template so the LLM can pin bugs to a commit

The most consistent cross-host recommendation: keep two tools warm and stay flexible [73]. One facilitator who has run many sessions describes converging on Lovable as default after starting with Bolt and v0 — the friction with non-coders is what drove the migration [78].

7. The runbook (TL;DR for the host)

T-48 h.

  1. Email attendees: create Lovable + Bolt.new accounts using Continue-with-Google. Include screenshots [52].
  2. Build one reference app in Lovable yourself; save the share URL.
  3. Fork a public starter repo (any minimal Vite/React) — backup if both generators fail [72].

T-2 h.

  1. Charge your phone. Enable hotspot. Tell venue IT how many SSIDs you need [39].
  2. Print 3 QR codes: Lovable login, Bolt login, the workshop’s shared starter repo.

T-0.

  1. 10 min: “what does your app do?” — forced planning step, ungoverned by LLM [77].
  2. 90 min: build in Lovable. One prompt → preview → one revision prompt. Discourage debug rabbit holes.
  3. 20 min: hit Publish. Generate QR. Show your app to the person next to you.

If wifi melts → flip to hotspot. If Lovable is down → flip everyone to Bolt and shorten scope to “landing page”. If both fail → the starter repo plus Claude Artifacts (runs inline in chat) is enough to ship something visible [8].

Total host spend.

  • Free path: $0. Lovable + Bolt free tiers cover one app per attendee with no iteration room.
  • Comfort path: $25 (one Lovable Pro for the host, attendees on free) [12].
  • Insurance path: $50 (Lovable Pro + a Cloudflare Pages-backed static repo).

The $100-250/mo “vibe-coding budget” reported for solo builders [27] does not apply at workshop scale, because three hours of guided activity is well below a single user’s monthly burn.

8. What this report did not answer

  • No named-host, dollar-itemised post-mortem of a vibe-coding workshop (“I taught 20 strangers Lovable for $X”) was found — workshop budgets here are extrapolated from per-seat pricing [27].
  • No first-person Reddit/HN thread of a vibe-coding workshop killed specifically by conference wifi; the wifi-fail evidence is industry-vendor framing [39].
  • Bolt.new’s total first-load weight including npm dependencies is not publicly documented — only the <1 MB WebContainer runtime is confirmed [29].

If you run this workshop, write up your spend, your wifi survival rate, and what broke. Future Scout runs would pick up your post and tighten these numbers.

Citations · 76 sources

Click the Citations tab to load…