Shipping Fast

Shipping Fast: A Field Manual From The Trenches

Based on my live talk at Station F. Everything below reflects what I said, including the messy parts, the wins, the lawsuits, and the near-faceplants.

Why speed mattered

• Lovable shipped and raised fast to outrun Figma’s PR and product push. Figma launched “Make” right before their IPO. We had to get ahead or get erased. Speed created oxygen to compete.
• AI evolves weekly. New models break your stack and your prompts. If you don’t rewire fast, you decay. Fast shipping is survival, not theater.

What “MVP” actually means

• It solves the core problem for a narrow persona. Nothing more. If one feature solves it, that’s your MVP. Everything else is gift wrap.
• Iterate visibly. If your UI is unchanged after two weeks, you didn’t talk to users. Treat the surface like the sea: always moving.
• Bucket features: Must-have / Nice-to-have / Don’t build. Stay ruthless.

Launches, timing, and reality

• Launch 1 (open-source, local): dev-only, tiny traction. Launch 2: spike with no retention. Launch 3: renamed, re-positioned, priced; timing finally matched the market, and it hit. Same team, different moment. Timing isn’t everything, but it’s the multiplier.
• Don’t mythologize “stars aligning.” We earned it by changing the product, branding, mission, and pricing between launches.

Experimentation: the cadence

• Run 3 tests/week minimum. Over 3 months, that’s ~36 shots. You don’t win by being right; you win by getting more at-bats.
• What to test: personas, value props, channels, pricing, onboarding, video, CTA, features. Measure with PostHog/Mixpanel/HubSpot; but also talk to humans, constantly.
• Sample sizes: site H1s → 50/50 across all traffic; cold B2B email → 100–200 is a sane floor; enterprise cycles → fewer conversations, deeper reads.

The growth stack that lets a non-coder ship in hours

1. Brain dump → PRD: 20 minutes of voice to OpenAI, then “generate a PRD.”
2. User journey: FigJam for flows. Keep it ugly and fast.
3. Frontend prototype: Build pixel-perfect mock-data in Lovable. Clickable, no backend. Show it to users the same day.
4. If validated: Wire Supabase, APIs, auth. Automate glue with n8n.
5. Harden: Export to Cursor, fix bugs, security pass. Stripe for payments. PostHog or Plausible for analytics.
6. Go-to-market assets: love.art (or your model of choice) fed with your PRD/context via n8n.
7. Second brain: Notion to track decisions, tests, and assets.

Result: hours to market, not months. The constraint is your taste and nerve, not engineering throughput.

What actually drove users

• FFF: Friends, Family, Fools. Your first believers buy belief, not features. Use them.
• Be loud: Daily presence across X, Reddit, YouTube, Discord. Opinionated content wins.
• Cold outreach: LinkedIn + email + phone. I’ve closed enterprise via cold calls. It’s unglamorous and it works.
• Paid: When scaling, 60–80% of traffic became paid (ads, influencers, affiliates). It’s fine. Just track CAC by cohort and kill what doesn’t earn back.
• Community: Discord went from a few geeks to ~200k. Champions answered faster than we could. Subreddit was user-run. Community is a force multiplier if you feed it with shipping and access.
• Hackathons: We seeded teams globally to ship with Lovable, creating developer proof onstage.

Wins worth copying

• AI Free Weekend / AI Shutdown: Negotiated free unlimited model credits across providers, ran a public bake-off, and harvested real usage data to negotiate pricing and model choices. Outcome: LLM cost dropped from ~$1.35M/week to ~$800k with similar consumption. Bonus: 750+ creator videos, free.
• Referral program (“Spread the love”): Converted word-of-mouth into a repeatable engine and gave users a path to free credits.
• Hiring signal: Bringing in Elena Verna (Head of Growth) and Notion’s marketing lead as CMO unlocked credibility with investors and candidates. People believe experts; VCs love them.

Fails you should learn from, not repeat

• Used a creator’s comparison video in Google Ads without consent: He’d partnered with a competitor, found out via X, threatened suit. Even if you “win,” you lose time and goodwill. Do the rights work.
• Figma integration via Builder.io: Needed perfect auto-layout hygiene; signups fell; competitive product beat us to a native launch; partner relationship torched. Right target (designers), wrong timing and dependency.
• “Dev Mode” naming: Figma trademark friction. Name features like a lawyer is in the room. Because they are.
• Times Square ad: Expensive vanity. Minimal impact given short rotations. If it doesn’t move signups or qualified trials, it’s theater.
• Public Webflow duel: 10k live viewers, we lost the on-stream build sprint. We proved speed but not polish. Choose battles where the judging rubric favors your advantage.
• Lovable 2.0: Built for collaboration and pricing changes users didn’t ask for. Within a week we made collaboration free and rolled credits. Talk to power users before you bet the roadmap.
• Partner program: Smart structure, wrong for the mission (we exist so anyone can build, not to send them to agencies). Strategy must align with story.

Outages and turning losses into narrative

• GitHub repo meltdown: A support agent killed our mega-repo because it stressed infra. We woke up to projects gone, cost ~$2M that day. We escalated to the CEO, got it restored, then declared a free weekend plus build challenge. We bled money and gained mindshare. Note: ~70% of that surge churned later. Hype ≠ retention.

B2C vs B2B, churn, and the truth about ARR

• AI tools churn. Heavily. Ours, competitors’, everyone’s. Many “ARR” claims are extrapolated from monthly revenue, not contracted. If you don’t fix activation, time-to-value, and repeatable use cases, paid trials evaporate. Be honest about the base.
• B2C gives you volume for experiments. B2B gives you depth and larger checks but fewer data points. Set your sample sizes to reality, not fantasy.

EU vs US reality

• Research parity is real. APIs are global. The gap is capital and risk appetite. US checks are bigger and earlier; Europe demands more proof. We created a US C-Corp shell to sell enterprise and make investors comfortable. It’s a paperwork hack, not an identity crisis.

Messaging and brand

• We moved from “for engineers” to “for everyone” to “zero friction” homepages. When famous enough, the best homepage is a prompt box. Kill copy. Let the product start. Change pages weekly: layout, color, onboarding, price. Static equals stale.
• Execution beats ideas. With GPT-level parity, output quality across tools is similar. Brand, distribution, and cadence decide winners.

Technical debt without drowning

• Early: bias to shipping. Late: introduce brakes. Legal, security, marketing, finance should veto dumb speed. Don’t spend $1.2M/week on models without a plan to renegotiate or swap. Ship fast until you know; then ship smart.

Sales notes founders don’t like but need

• If you can’t sell your own product, no one can. Founders must own the first closes before hiring. I’ve done 9–12 month enterprise cycles in past roles. It’s part stamina, part system.
• Cold calls work. I closed Christie's by phone in a previous startup, then later sold the company to them. Pick up the phone.

The five-step path for your next 30 days

1. Ship a clickable prototype this week. Mock data is fine. Show ten people. Change it daily.
2. Run 3 experiments/week. Keep a ledger. Kill underperformers fast.
3. Instrument everything. PostHog or Plausible. NPS in support. Record calls. Read transcripts.
4. Explode distribution. FFF + Cold + Be Loud. Add a referral with real value.
5. Negotiate your stack. Bake-off models with live traffic. Pay only for what wins.

Closing conviction

Shipping fast isn’t reckless. It’s disciplined urgency. Ruthless scoping, public iteration, honest metrics, and unapologetic distribution. Do the work in hours, not quarters. If it fails, pivot. If it hits, double down. The market rewards motion and taste. Everything else is noise.