AI Agencies vs SaaS, Veo3 Ad Pipelines, Hermes Agents — AI Daily May 09

404 messages · 65 active members

404
messages
65
active members
@arielletolome, @mb29266, @tounano
top contributors

Overview

Today's chat orbited the question of where the real money is for AI builders: ai-native agencies vs SaaS vs rapid-fire ecom plays. @Kieran argued proprietary AI workflows form a compounding moat that PE will eventually buy out, while @tounano countered that ai-native is a head start, not a long-term moat — once standardized, margins revert. @arielletolome and @javiernjmj added that SaaS only makes sense with VC fuel, and that marketing beats product virtue almost every time (Cal.ai's $100M sale, endless quiz/horoscope/supplement playbooks). On the production side, @arielletolome detailed a fully automated Seedance pipeline (~1hr passive per video, throttled lately) plus a Veo3 system targeting 50–60 one-minute videos in ~10 minutes. Media buying ran deep: 1-1-4/1-1-5 adset structures, broad targeting, max CPA, kill at 1.5x spend with no conversions. @mb29266 walked through a 50–300 ads/day routine with Claude pre-planning batches overnight and advertorial→PDP funnels with auto-checked subscription bumps for instant MRR. @kennyaronson teased a 24-hour AI advertorial workflow hunting the right monetization angle. On the tooling side, builders are migrating from OpenClaw to Hermes for agent orchestration, citing better reliability and sub-agent spawning, while flagging GPT-5.5 copyright refusals vs more permissive Opus behavior. @romainpetitjean shared a Claude Code + Codex workflow turning markdown transcripts into shareable HTML artifacts via surge.sh, and @andreilunev dropped the code.claude.com/docs/llms.txt compliance doc that several builders immediately plugged into their stacks.

Topics

@Kieran argued ai-native agencies (proprietary workflows, lean teams, compounding data moats) beat SaaS on ROI and churn, citing YC's pivot. @tounano countered that AI is a head start, not a moat — once standardized, margins revert. Consensus: vertical B2SMB SaaS still wins long-term LTV, but the window to extract outsized margins from ai-native agencies is now.

@arielletolome shared his automated Seedance pipeline (~1hr/video passive) and a Veo3 system targeting 50–60 one-minute videos in ~10 minutes. Media buying details: 1-1-4/1-1-5 structures, broad targeting, max CPA, TCPA only on destabilization, kill at 1.5x spend with no conversions. He's building a creative agency rather than SaaSing it.

@mb29266 detailed a 50–300 ads/day routine with Claude pre-planning batches and advertorial→PDP funnels with auto-checked subscription upsells for instant MRR. @kennyaronson teased a 24-hour AI advertorial workflow with unused capacity. @Coybh casually mentioned $40k/mo MRR from a shut-down store's residual subs — a reminder of how durable subscription ecom can be.

Multiple members are moving workloads from OpenClaw to Hermes, citing better reliability, self-healing, and the ability to spawn specialized sub-agents. Some flagged GPT-5.5 copyright refusals on skills (e.g., debrid downloads) where Opus remains permissive, plus the need for platform diversification after April's Claude Max issues.

@andreilunev shared code.claude.com/docs/llms.txt as a compliance check that materially improves Claude Code rule-following, plus a distilled Russian-channel knowledge dump. @romainpetitjean shared a Claude vs Codex workflow turning .md transcripts into standalone HTML pages via surge.sh — community preferred Claude's output, sparking ideas for human-in-the-loop artifacts agents can update live.

Key Takeaways

  • AI-native agencies offer a real but temporary moat — extract margin now while the window is open; PE will likely buy these out as SaaS appetite cools.
  • @arielletolome's media buying stack: 1-1-4/1-1-5 structure, broad targeting, max CPA, kill at 1.5x spend with zero conversions or 3x spend with CPL above payout.
  • Subscription ecom MRR is sticky — auto-check the subscription box on checkout, instant-refund complainers, and use a dispute app to defend chargebacks.
  • Hermes is winning mindshare over OpenClaw for agent orchestration; GPT-5.5 is noticeably more restrictive than Opus on copyright/skill execution, so model choice matters.
  • Drop code.claude.com/docs/llms.txt into your Claude Code workflow as a compliance check, and keep .md as canonical source while publishing HTML artifacts via surge.sh for humans.

Hot Threads

@Kieranstarted

AI-native agencies as the new SaaS — moat, margins, and PE exits

30 replies6 participants
@arielletolomestarted

Seedance/Veo3 ad pipeline + media buying structure walkthrough

28 replies5 participants
@weslindquiststarted

Moving from OpenClaw specialist agents to Hermes sub-agents

8 replies6 participants

Linked Items