jev is INSANE. in 40 seconds it broke down 724 live ads from 37 brands. every hook. every format. offer. cta. awareness stage. landing page mismatch. used 9 cents of tokens. (will be avail in @stealads + mcp)

each AI shopper got a name, age, gift list, budget and deal attitude. they scrolled real ads. then went back to buy a gift for everyone on their list. Dana's husband says he wants nothing. olive oil gift box. Priya shops at 3am. same Blissy sale for her sister, mom and best friend. Hannah has 3 rescue dogs. free octopus bed from BarkBox.

founders argue this every november. "50% off works." 19 of the 100 were discount hunters. they picked 50%-off ads 3x more than ads with no discount. everyone else picked them 31% less. drop to 50% and you train them to only buy on SALE.

jev ran black friday. in 3 minutes it ran 100 AI shoppers on 3,092 real US meta ads from 669 brands. top 100 ads they picked: only 2 had 50% off. (avail in @StealAds)

i built our new onboarding @StealAds around a HUGE problem. no one wants to upload all their photos and brand assets. now you get your whole brand pulled in 60 seconds.

went out with a DTC brand owner spending $60k/mo on Meta ads this weekend. he's frustrated bc his 4 person creative team is burning $25k/month shooting 30 ads that die in 2 days. i showed him @t3dotcodes on my phone. connected to linux boxes running claude agents w/ skill loops. prompt: run our competitor ad extraction pipeline on his top 10 competitors. filter collation_count > 5, active > 75 days. decompose the 3 second hook mechanisms with vision. cross reference against @exa_search Reddit reviews. write 5 finished UGC briefs. run time: 4 minutes. the agent pulled 14 long running competitor ads with heavy scale flags. it broke down the exact visual anchors in the first 45 frames. it mapped those mechanisms to the exact friction phrases his customers use on Reddit. i slid the phone across the table. 5 ready to go video scripts with frame by frame visual cues ready to render in Seedance. he stared at the screen, put his fork down, and texted his head of growth: "cancel the video agency retainer tomorrow morning." the gap between guessing what ad concepts to film and autonomous market extraction is just nuts.

okay grok bot is insane for ads my content bot: → scours x + tiktok for formats that are actually working → steals the structure → writes the script → generates all the shots → edits it → qa's its own cuts and remakes the bad ones → texts me all done videos this one's a fake conference q&a for saas

i got tired of youtube recommending other people. so i built metube. a chrome extension that uses @krea_ai 3 to put me in every single thumbnail. @johnternus? me. @nateherk? me. a 4 hour podcast i'll never watch? also me. finally the internet is about the only thing i really care about. me. comment "ME" and i'll send it to you so it can be about you instead (worse, but fine)

how to steal the top 1% of viral carousels in your niche and turn them into finished slides: 1/ maintain a dynamic watchlist of 50 competitor profiles with @scrapecreators. pull all carousel posts across Instagram and TikTok photo mode into raw JSON on a nightly run. 2/ run multi-signal outlier filtering against each account's own distribution. kill anything under 2.0x median likes. filter for 1.5x median comments to measure real depth. drop carousels under 5 slides. 3/ run multimodal vision across surviving slides in claude. extract the exact text hook and visual anchor on slide one. map the curiosity escalation on slides two and three. extract the exact conversion CTA on the final slide. mine the comment section for raw buyer objections. 4/ auto-bundle full-res slides into a searchable metadata vault. index them by arc type (ie mythbust. teardown. transformation. 6-step framework). 5/ feed the top ranked structural pattern into your carousel generator. keep the proven visual pacing and tension points. swap in your own product proof and brand assets. 6/ publish and track your own 2.0x outlier winners. pipe your top performers back into the library to train the next batch. stop saving screenshots on your phone. build an automated visual pipeline that extracts winners and produces 100x finished creative.

i (finally) gave the @StealAds marketing site a facelift. → autopilot ads → ad breakdowns for your whole niche (in seconds) → rival DEEP DIVES (angles/formats/etc) if you're on the waitlist + not in yet (thank you) i have a fresh round of invites coming next week.

how to run an autonomous meta ad loop in 2026 (that never burns budget): here's the exact 5 step closed loop architecture: 👇 1/ pull ad metrics every 15 to 30 minutes via Meta API. → track spend, CPA vs target, frequency, CTR vs baseline, hours live, learning status. → watch 6 hour velocity slopes to catch CPA spikes BEFORE 48 hour averages move 2/ jev (@typesafeai) returns a single JSON decision. → decisions: KILL, WATCH, SCALE, PATCH_PAGE_FIRST. → hard rules (tweak for your brand): 0 conversions + spend ≥ 2x target = KILL CPA ≥ 2.5x after 24h = KILL frequency ≥ 3.5 + decaying CTR = WATCH or KILL CPA ≤ 0.8x with ≥ 8 conversions = SCALE 3/ gate every SCALE with a message match check. → jev compares ad hook, offer + CTA against landing page H1, offer + CTA → 'mismatch' blocks 'SCALE' and triggers PATCH_PAGE_FIRST. → don't kill a winning ad because your landing broke the promise. 4/ run 3 hook alignment checks with jev 1.14 → check the opening frames: is copy big enough for mobile? do the visual hook, text overlay, and verbal hook (transcript) line up 100%? → flag ANY badly aligned creative + cut off text before your CTR crashes 5/ call claude ONLY after Jev says go. → when jev triggers SCALE claude pulls winning hook structures. → draft + run 3 to 5 variations for new ad sets (one URL per set) don't murder winning ad hooks because your landing page can't close. run a decisions loop at ridiculous speed.

jev blitzing through 193,355 hooks so growth engineers can drop bangers