Hermes Revenue Wins, Meta Ad Safety, GPT-Image-2 — AI Daily Jun 20

350 messages · 63 active members

350

messages

active members

@rmktg, @mb29266, @WhiskeyATX

top contributors

Overview

Today's chat was dominated by a Hermes reality check — @samarh90 called it a token guzzler and shiny object, prompting @rmktg and @Coybh to share concrete revenue use cases including $8k/day ad spend at decent ROAS and a sales pipeline booking 30 calls/week. Consensus: Hermes wins on rapid prototyping speed, not raw capability, and pairs well with custom infrastructure replacing tools like GHL. Romain also surfaced that OpenAI 5.5 exposes a 272k context window inside Hermes via OAuth, though extra usage fees are tripping people up. Meta ad safety was the other big thread, with @cdxxx asking for ban-safe bulk uploaders for a big client. AdManage, AdsUploader, and Adnova got endorsements, while @rmktg stressed that system-user API access, rate limiting, and accuracy monitoring matter more than tool choice. @ferchonaso shared a 2-hour AI-scraped Meta-flagged words list, with @c_1media warning against trusting LLM-identified flags without rep confirmation. On the build side, @tounano shared a StoryBrand + gpt-image-2 + Codex workflow to escape generic SaaS landing pages, with agreement that iterating designs as images is more token-efficient than HTML. Grok OAuth integrations are fragile — @geilt shipped a custom harness, but @sam reported SuperGrok breaking on OpenCode and Hermes. GLM-5.2 is being seriously evaluated as a Codex replacement, the Fable hunt continues (Opus 4.8 falling short), and builders are stacking 3+ Claude subs to scale parallel agent work.

Topics

Hermes Revenue Proof + 272k Context via OAuth30 msgs

@samarh90 challenged the group to prove Hermes drives real revenue. @rmktg shared running $8k/day ad spend at decent ROAS, built in 2 weeks without GitHub. @Coybh runs his entire sales pipeline on Hermes booking ~30 calls/week. Separately, @romainpetitjean confirmed OpenAI 5.5 in Hermes exposes a 272k context window via OAuth, though extra-usage fees on top of subs are catching people out.

Meta Ad Ban-Safe Bulk Uploaders for Agencies22 msgs

@cdxxx asked for the lowest-ban-risk bulk ads uploader for a big client. AdManage got endorsements from @filiuser and @rutty123; @realcrischico vouched for AdsUploader across 100+ ad accounts plus a Claude Code CLI. @rmktg emphasized system-user API access, rate limiting, and accuracy monitoring matter more than tool choice — bans correlate with aggressive verticals, not tooling.

GPT-Image-2 + StoryBrand for Non-Generic Landing Pages15 msgs

@tounano shared a workflow to escape cookie-cutter SaaS designs: run StoryBrand in Codex, generate top-fold image variations in ChatGPT, then Codex implements from screenshot. Key insight: iterating on images is more token-efficient than iterating on HTML (DeepSeek research backs this). Codex's native gpt-image-2 tool also generates site imagery in-flow.

Grok OAuth Breakage & GLM as Codex Alternative10 msgs

@geilt shipped a custom harness with Grok OAuth, secure credential storage, and integrated flow, while @sam reported his SuperGrok sub stopped working on OpenCode and Hermes, falling back to failing API calls. @carfarah probed whether the group has switched from Codex to GLM-5.2, with guardrail behavior and browser automation support as the open questions.

Fable Replacement Hunt + Multi-Sub Claude Stacking13 msgs

@zippi101 and others continue searching for a Fable-equivalent for creative/copy work, with Opus 4.8 falling short. @Max_Bernstein runs an LLM cocktail: 5.5 extra-high-fast + Grok CLI + Codex with 4.8. @tz1888 framed GPT-5.5 as the OpenAI equivalent of Fable (~10T params). @mlsmdm polled who's running 3+ Claude subs to scale parallel agent work.

Key Takeaways

Hermes ROI is real but speed-driven — @rmktg hit $8k/day ad spend in 2 weeks, @Coybh books 30 calls/week; value is rapid prototyping, not raw model quality. OpenAI 5.5 also exposes 272k context via OAuth.
Meta ad ban risk lives in API behavior, not the uploader — use system user auth, rate limit, monitor accuracy, avoid aggressive verticals. AdManage and AdsUploader vouched for at agency scale.
Iterating landing page designs as images (gpt-image-2) is more token-efficient than HTML iteration — pair StoryBrand framing with image variants, then Codex implements from screenshots.
Grok OAuth integrations are fragile right now — working in custom harnesses but breaking SuperGrok subs on OpenCode/Hermes. GLM-5.2 is the next Codex alternative under evaluation.
Don't trust LLM-identified Meta flag word lists without rep confirmation — @c_1media warned about the 'all-is-flagged' false positive trap. Builders are also stacking 3+ Claude subs for parallel capacity.