@angus.sewell

What they said

Use cases mentioned (in order), with what the frames show:

1. Competitor ad analysis: tweet by Matthew Berman (@TheMattBerman, dated Sep 17 2026, 1M views): "jev is INSANE. in 40 seconds it broke down 724 live ads from 37 brands. every hook. every format. offer. stage. landing page mismatch. used 9 cents of tokens. (will be avail in @stealads)". Frame headline "steal the whole category" with a mosaic of hundreds of ad thumbnails. Speaker: "analyze all of their competitors' ads, which you can do with this prompt, by the way." On-screen prompt "COMPETITOR AD TEARDOWN" (read from frame): paste into Claude Code; 1) install the TypeSafe skill (claude plugin marketplace add typesafe-ai/skills, then claude plugin install typesafe@typesafe-ai/skills) with a key in TYPESAFE_API_KEY, call POST https://api.typesafe.ai/v1/systemone with model jev-1.13.0; 2) pull every active US ad for [competitor 1, 2, 3] from the Meta Ad Library with the Apify actor apify/facebook-ads-scraper (200 per brand; "the official Ad Library API won't return these ads"); 3) drop ads whose copy is an unfilled template placeholder, compute days_running from each start date; 4) send each ad as its own request (copy, format, destination, days_running); 5) ask five questions in one call: pain_point, bold_claim, curiosity, offer (choice), awareness_level (problem/solution aware?), cta_strength (score) ...; 6) Jev reads text only; for video ads, take the first few seconds with transcription; 7) save everything to ads.csv, then show ads running 60+ days grouped by hook, with tokens and cost; flag ... instead of counting it.
2. Email classification: tweet from vogel (@ryanvogel): "this model is actually insane at email classification. i tested it on 1500 of my own emails to see how well it works and i am blown away." Speaker: "analyzing emails, really cool"; frame shows an email table with priority (Low/High), spam %, reply %, time (ms).
3. "Adaptable ad blocker. Wait, that's kind of sick actually" (Fandom page demo).
4. Website visitor -> follow-up: "when your analytics platform detects somebody new on your landing page, it can take all the information of where they clicked, their IP address information, whether we have a contact for them in our CRM, and feed it all to Jev in a second and figure out which of your follow-up email templates would make the most sense to send to this person after they leave the page. You can set all that up with this prompt." Frame: "WEBSITE + CRM ROUTING" prompt: every hour pull identified PostHog persons with events in the last 7 days (pages, steps, docs opened); match each to a [HubSpot] contact and add CRM fields (role, last touch, open deal); send that as the state, ask Jev route (sales_..., onboarding_help, nurture, ... blocked_by_setup); write jev_intent and jev_route; never email... Frames show PostHog analytics, IP geolocation lookup, a HubSpot contact (Dana Reyes), and Gmail drafts "15 minutes this week?" and "getting hubspot connected" signed Angus.
5. Live Zoom coaching: "that's actually good - build an actually good Cluely, dude?" (Frame: "Zoom Webinar Backstage" with prospect/SDR lines; "Transcript -> Jev", "A sales call, recording on: what the prospect just said / what they said earlier on this call / who is on the call / your offers and specialists -> Jev picks what to say while they talk".)
6. Cold outreach system: "imagine a cold outreach system like this": frame "A new lead hits the CRM: what they wrote in the form / what they read before they asked / what the company looks like / your candidate messages -> Jev matches lead to message".
7. Close: "Get ready for some Jev posts. We'll unlock so many sick use cases, bro. Follow your boy. I will keep shitposting all of my Jev ideas for free."

• Caption: "If you want the prompts for 13 different Jev use cases comment 'prompt' and I'll send em your way. #ai #buildwithai #vibecoding #jev"

Try this

• Marketers and agencies: Try these use cases with any strong AI model: a competitor ad teardown (hooks, formats, offers, landing pages), email triage, and choosing which follow-up email a website visitor should get.
• Sales teams: Prototype live call prompts that suggest what to say based on what the prospect just said, who is on the call and your offers.
• Cold outreach: Feed the lead's form answers and pages they read to an AI and have it pick the closest message from your tested set.
• Test small first: Run any prompt on 20 to 50 real cases, check the output by hand, and track the time it saves.
• Privacy: Do not send customer, CRM or email data to a third-party model without checking the terms and your consent rules.
• Treat the hype as hype: The results come from social posts, not tests. The same flows usually work with other models.

Links

• Jev (LLM; model id jev-1.13.0) via TypeSafe skill: https://api.typesafe.ai/v1/systemone, claude plugin marketplace typesafe-ai/skills (read from frame; unverified; I did not install or call anything).
• Claude Code (used to run the prompts), Apify actor apify/facebook-ads-scraper, Meta Ad Library, PostHog, HubSpot, Gmail, Zoom, Fandom (demo site), "stealads" (@stealads, a product Matthew Berman says will have it), Cluely (AI meeting assistant mentioned).
• Accounts mentioned: @TheMattBerman, @ryanvogel.