Lições reais de quem opera SaaS próprios em produção. Preços, erros, decisões arquiteturais e números que ninguém publica.
A terminal AI agent in Node grows to V8's ~2GB heap ceiling in hours. The 3 crashes I hit and the compaction, cleanup, and redaction pattern I ship now.
Read post →Mockito's real cost isn't writing the mock — it's build_runner. I cut it from a 40k-line app and local test time dropped from 45s to 8s. Here's the math.
Read post →Stripe and Meta deliver the same webhook more than once. Without dedup, ~0.5% of customers get charged twice. How I caught it in an 11pm stress test, and the pattern I use.
Read post →I ran 50 real Brazilian-Portuguese clips (WhatsApp voice notes + podcasts) through all three. Verdict: Deepgram for streaming, Whisper for batch, and I skip Google STT. The numbers below.
Read post →Animated heroes don't convert. Social proof in the first viewport and an LCP under 2.5s do. The numbers, the right budget, and what I'd do today.
Read post →A 4-stage pipeline to auto-publish on Instagram with AI — caption, image, human review queue, Graph API. Three stacks from $0 to $40/month, with the real numbers.
Read post →v0 ships a website, Bolt ships code that won't compile for the stores. For a native app that survives review, vibecoding stalls. Here's where, and the fix.
Read post →If 2+ of these 7 signals show up in your vibe-coded SaaS, stop iterating and hire a dev. Closed matrix with scoring and action band per tier.
Read post →WhatsApp bots generated by Lovable, Bolt, or v0 break in 8 predictable ways once they hit production. Here's the checklist I run before any bot ships to a paying client.
Read post →The 12-month invoice for scaling a Lovable-built SaaS from 0 to 1,000 active users: $3,000–$5,000 across infra, AI credits, and the rewrite nobody warns you about.
Read post →