Grok 4.6, Gemini 3.8 Flash, Meta Ad Nukes — AI Daily Sep 2
410 messages · 77 active members
Overview
Topics
@thewildzeno made the case for Grok 4.6 as a fast, terse code executor — great as a subagent when direction is clear, weaker on design/UI — with speculation Cursor's usage data positions xAI for an 'Auto' routing mode. Gemini 3.8 Flash shipped with ~66% token reduction and improved agentic video, but @jonmacofficial found it slower than 3.7 in his Orca orchestrator and still not standalone-ready for coding without an Opus 5 or Sol advisor.
@arielletolome, @ltv_1, and @scalingfrog reported ad accounts nuked mid-scale with CPLs jumping from $5 to $12+ and checkout costs doubling day-over-day. Multiple builders confirmed August was their worst ecom month while July was the best. Fixes discussed: fresh ad account + pixel + page swaps, moving spend via Meta API across agency assets, and running multiple accounts past $50k/month.
@arielletolome argued native statics are saturating as every YouTuber pushes them — her approach is to rip top Meta creatives and rerun them on Taboola, and she shared a Newsbreak scaling Notion doc noting NB's whitelabel exchange lets you scale vertically in one account. She also open-sourced an overlay text video editor for TikTok-style verticals and announced a UGC/VSL AI editor beta with early takers.
@sibunting ran a stable 8-hour overnight session with Fable 5.1 alongside 5 other agents including Opus and Sol, calling stability a meaningful upgrade, though @scalingfrog and @jonmacofficial saw 20% of weekly quota gone in 20 minutes of low-effort scans. Astra is on release watch for tomorrow following the post-Claude cadence, and @roki1122 shared allegedly leaked GPT 2.1 image samples that impressed the group.
@jasonakatiff hit $423 in Vercel build costs in four days on a NextJS lead router and migrated to a $300/month Hetzner box promising 2-5x faster builds — traffic costs were only ~$10, the pain was CPU-heavy NextJS builds. He then argued this is the largest opportunity of our lifetime: every company needs AI but has no idea how to implement it, opening a market for custom harnesses, OCR, Slack tooling, and staff retraining.
Key Takeaways
- Grok 4.6 shines as a fast code-writing subagent when direction is clear; Gemini 3.8 Flash cuts tokens ~66% but is measurably slower than 3.7
- Meta ad account wipes are sweeping ecom verticals — keep pixel and BM clean and be ready to rotate accounts via API past $50k/month spend
- Newsbreak runs on a whitelabel exchange, so native statics scale vertically in one account — no need for Meta-style account stacks
- Fable 5.1 holds stable across 8-hour multi-agent runs, but weekly quota burn is notably faster than 5.0 despite cheaper pricing
- Vercel build costs can spiral fast on NextJS; a $300/mo Hetzner box often pays for itself in days and the real AI market is SMB implementation, not models
Hot Threads
Meta ad nukes, CPL spikes, and native static scaling on Newsbreak
Vercel build costs, Hetzner migration, and the AI SMB opportunity
Grok 4.6 use cases and 'Auto' routing speculation