Aman Jain (@ajain) profile picture

Engineering at @OpenAI. Previously: Co-founder of @usenotchai (acquired by OpenAI), Meta, Twitter

San Francisco, CA
Joined February 2012
Photo by Aman Jain (@ajain)Photo by Aman Jain (@ajain)Photo by Aman Jain (@ajain)Photo by Aman Jain (@ajain)Photo by Aman Jain (@ajain)Photo by Aman Jain (@ajain)Photo by Aman Jain (@ajain)Photo by Aman Jain (@ajain)Photo by Aman Jain (@ajain)Photo by Aman Jain (@ajain)
Aman Jain retweeted
claim your business on ChatGPT and engage with customers in new ways:
Now in beta: Businesses in ChatGPT. Claim your business, manage your information, and get insights on how people engage with you. Eligible businesses can also build a Business Agent to answer customers’ questions and help them explore what you offer. openai.com/index/introducing…
165
71
14
1,523
270,048
Aman Jain retweeted
Notch’s founding team is joining @OpenAI. Two years ago, @ajain and I started Notch because we believed AI agents could help businesses grow autonomously. We built agentic experiences for advertising, worked with thousands of small businesses, and learned what it takes to turn fast-moving technology into something people use daily to run their ad budgets via Notch agents. I’m proud of what our team built. They brought care to every iteration, and our customers made the work sharper. To our teammates, customers, investors, and everyone who backed us: thank you. Aman and I are excited to bring what we’ve learned to the OpenAI Ads team.
70
19
12
333
32,573
Aman Jain retweeted
I reviewed 20 Meta ads for cleaning products launched since April. 15 used the mess to show who the product was for: soccer jerseys, school backpacks, pet dander, gaming crumbs, garden dirt. A spotless kitchen is a lazy brief. Shoot five messes before writing five more hooks.
1
1
73
Aman Jain retweeted
Your whole ad team is now a single hire. Meet Max by Notch. Your AI growth marketer. The ad grind never ends. Check the account. Refresh creative. Stalk competitors. Read the numbers. Again tomorrow. Forever. Max does the whole job (No CLI. No API keys. No Claude Cowork setup) - → Checks your ads every morning. Flags winners. Catches budget drains. → Studies competitor ads daily. Copies what works. Skips what doesn't. → Ships new creatives weekly. Statics, UGC, video. In your customer's voice. → Reports in plain English. Next tests already queued. → Kills losers. Scales winners. Works while you sleep. Max drops the drafts. You make the calls. He takes the grind, not your team. Top ecom business owners already trust him as their performance creative partner. Their teams ping him in Slack like a teammate. First of a family of agents. Day one. Comment "MAX" with your brand URL. First 50 brands get 1,000 free credits + founder-led onboarding. Me and the team, personally.
68
27
33
250
1,350,015
Aman Jain retweeted
the best ads from notch customers. with the prompts that made them. copy a prompt. paste it into notch. ship your brand's version tonight. the prompts are in the comments. here's how each ad was made. → drop the input. a brief. an idea. a winning ad to recreate. → notch ad agent loads brand memory. strategizes the angle. produces the ad. → on-brand version renders. live to meta in one click. study them. copy them. modify for your brand. ship yours.
23
11
9
98
414,563
Aman Jain retweeted
you've bookmarked 247 winning ads on twitter. you've shipped maybe 2. the gap was never strategy. it was the production cycle between bookmark and live. the workflow that closes it sounds fake until you try it once. here's the workflow. → paste the winning ad into claude. ask for the teardown. → claude returns the hook, angle, offer, and a ready-to-use prompt. → paste into notch ad agent. it doesn't generate. it strategizes and ships. on-brand version renders. → push to meta. same day. no designer. no notion. no waiting. while the teardown thread is still arguing why the ad won, your version is collecting data on meta. drop a winning ad link in the comments. we'll spin one up for you.
38
32
17
348
622,142
Aman Jain retweeted
Meet Notch Variants. Engineering 10 distinct ads from one winner used to take 6 weeks. Notch Variants generates 20 in 6 minutes. Each one passes Andromeda's clustering check. Same structure underneath. Distinct ads for different personas. → 20 sub-agents spawn in parallel. → Each varies one surface dimension. Avatar. Environment. Casting. Lighting. → Each inherits the structure lock and your brand memory. Zero drift. → All 20 build simultaneously. One session. 20 distinct ads. It solves two problems at once. Fatigue and clustering. Your top performer stops being a one-shot. Unfair advantage for brands sitting on a winner. Drop your brand URL and your top ad. We'll spin up 20 variants in your brand.
39
15
17
242
623,279
Aman Jain retweeted
i spent 7.5 years inside meta. built facebook lead ads to $1b. here's what i saw build, what no advertiser has adjusted to. before your ad hits the auction, andromeda narrows tens of millions of ads to 1,000 candidates. miss the cut, nothing else matters. andromeda rewards distinct volume. clones get clustered as one. that's why your 15-variant test ran as 1 ad. here's how to break past andromeda with notch. → one prompt in notch. 15 distinct concepts out. each varied on every dimension andromeda clusters on. → notch handles production. existing footage cast where it fits. new shots generated. brand memory keeps it all on-brand. → ship all 15 to meta. andromeda surfaces the survivors. you find winners in days, not weeks. → clone the winning structures with competitor clone. spin them into new variations. compound the wins. it's the unfair advantage for media buyers. feed andromeda diversity. or get filtered out before the auction even starts. drop your brand url. we'll spin up a distinct campaign in your brand. one andromeda actually rewards.
24
51
24
708
455,054
Aman Jain retweeted
Memory just landed inside Notch Agents. Brand colors. Voice clones. Do-not-say words. Taste. All read before every generation. On-brand ads. Faster. With brand context. Upload your brand context once. Or teach the agent inline as you chat. Each chat is no longer a re-introduction. It's training. Drop your brand URLs and instructions in the comments. We'll spin ads around your context.
44
22
35
167
649,205
Aman Jain retweeted
You sent your team a competitor's ad and said "make ours like this." That was nine days ago. In those nine days, their ad ran 100+ comments, 50 reshares, and never left rotation. That's not a viral fluke; that's a format that's scaling. The longer it runs, the more you know it's printing. Yours is still a brief. Here's how Notch competitor clone delivers the proven format in your brand style, fast. It opens your category's ad library in front of you. Every active ad, how long each has run, which ones are scaling. → Scroll the wall. Statics and video. → Pick the one working. The agent reads it the moment you select. → For statics it maps layout and copy. For video it goes deeper. Hook timing, pacing, script, cut rhythm, cta cadence. → Then it spins a variant in your brand. Structure preserved, your product, colors, and assets wrapped around every beat. The screenshotting doesn't make you money. Shipping does. Comment below your brand URL. We will share the competitors' insights and build an ad for you.
48
18
21
195
586,377
Aman Jain retweeted
Notch Clips. every other ad tool ignores the footage sitting in your dropbox and generates new clips from a prompt. notch opens your library instead. it watches every clip you have ever shot and understands what is inside each one. the mood. who is in frame. what they are doing. the quality. so when you ask it to build an ad, it does not start from scratch. it reaches into your past shoots, finds the clip that fits, and slots it into the exact moment that needs it. here is what that looks like. you had thirteen versions of one reaction shot. notch picked the one that worked and left the other twelve. not by filename. not by guessing from a reference image. it watched all thirteen and could tell the difference. that is the whole idea. the right clip was already in your library. every other tool was about to make a worse one. footage you already shot, turned into every ad you still need. drop your site and a few clips you have. notch builds the first ad from them.
50
26
45
274
407,937
Aman Jain retweeted
if we fuck up the ad, why should you pay? notch is the one place that doesn't charge for regenerations or edits. one price. one ad. however many tries it takes us to get it right. every "generate" is a slot pull. miss = re-roll = more credits gone. these tools don't sell you ads. they sell you chips. and the house wrote the odds. here's the math nobody runs before they load their card. a 30-second ai ad is never one generation. it's 30 to 40 pulls. → wrong face. re-roll. → hands melt. re-roll. → lip sync drifts a frame. re-roll. → b-roll doesn't match the brand. re-roll. every pull bills you. the winning clip and the garbage clip cost the exact same. you're not paying for the ad. you're paying for the misses on the way to it. the casino model (higgsfield, runway, kling, every per-credit tool): → buy credits up front → pull the lever → pay whether the clip is usable or trash → the agent loops to fix itself? your meter runs faster → you ration generations like chips at a table the finished-ad model (notch): → you direct the ad, not the generations → the agent re-rolls, iterates and fixes itself on its own dime → the misses are our cost, not your wallet → one price for the deliverable. however many pulls it took. per-credit pricing isn't pricing. it's a casino floor. the meter is the product. the ad is just the bait. you should pay for the thing you can actually run. not for the privilege of pulling the lever 40 times to find it. comment your brand below. i'll ship you 10 winning ad variations. on us. no credits. no meter. no catch. or try notch for free today.
55
32
45
224
300,681
Aman Jain retweeted
agencies leave. freelancers ghost. employees quit. this one stays. notch is an self learning ai ad agent that learns your brand permanently. your voice. your visuals. your audience. what converts. it studies your competitors weekly. remembers every session. gets sharper the more you use it. try finding that loyalty from a $4k/month agency that also handles 11 other accounts. you're not renting a tool. you're training a teammate that never leaves and costs less than lunch.
61
51
55
307
1,843,371
Aman Jain retweeted
a format just crossed 2.3 million views and most media buyers haven't seen it yet. pixar-style woman. bathroom. warm lighting. talking about her skincare routine. she was generated in 4 minutes. by Q4 every brand will run this. right now maybe 200 do. that's the window. you know this feeling. you felt it with ugc in 2021 before every brand had a creator program. you felt it with reels before the CPMs caught up. the format that feels slightly wrong at first is always the format that prints before everyone copies it. animated character ads feel slightly wrong right now. that's the signal. the accounts running them aren't posting about it. they're quietly scaling 47-day creatives while the rest of the feed refreshes every 14 days. same character. new script. new hook. new angle. the character never fatigues. the audience never tires of a face that was never overexposed. → one operator producing 15-20 per week → zero talent. zero shoots. zero "can you reshoot with better lighting" → cost per video: under $15 → cost per ugc video: $300-800 → the math isn't competitive. it's disqualifying. the brands who moved first on ugc owned 2022. the brands who move first on animated characters own 2026. the window is measured in months, not years. comment "EARLY" for the production setup (follow for dm)
28
6
34
2,321
Aman Jain retweeted
47 sales from a girl that doesn’t exist. zero ad spend. one photo. one product URL. one sentence: "make videos." the 23-year-old who built this has no marketing experience. and that’s exactly the point. here’s the setup: → dropped one image into the avatar builder → prompt: "an american woman. 27. straight blonde hair. energetic. warm. " → refined twice. locked in. she exists permanently now → video agent. selected avatar. selected product → "20 videos. different angles. you pick the formats." the agent researched hooks, wrote 20 scripts in her voice, matched formats the algorithm rewards, generated 20 finished videos → posted 3/day on a fresh account 2 weeks: 3 videos crossed 100k views 8 crossed 50k 47 organic sales those top 3 aren’t just content. they’re proven hooks — validated for $0. → "take the top 3. 10 ad variations each. same influencer. different openers." 30 ad-ready videos. cpa 40% lower than testing blind. the pattern: → create influencer once → agent generates daily content in her voice → algorithm surfaces winners for free → winners become paid ad creatives → only scale what’s already proven controversial take: scaling used to cost a media buyer, a strategist, UGC creators, and $10k/month to learn what works. now it’s one photo and "make videos." the skill isn’t writing hooks. it’s pressing scale on what the algorithm already chose for you. rt + comment "EASY" for the zero-expertise setup guide (follow for dm)
22
13
1
38
2,478
Aman Jain retweeted
Let me explain why 90% of people trying AI affiliate marketing are leaving money on the table They're using tools that give them raw clips and calling it content. A talking head with no hook structure, no b-roll, no music, no captions, and then they wonder why it gets 200 views The brands and offers on platforms like Glitchy: -> makeup -> skincare -> serums these are visual products. They need ads that hit pain points, build trust fast, and have a clean CTA. That's not a clip. That's a complete ad. This is where usenotch.ai genuinely different to everything else I've tested You don't touch a timeline. You describe what you want: -> the product -> the angle -> the vibe Notch agent handles everything. It scrapes your product page for key info, writes a script built around hooks and pain points, picks the right avatar for your target audience, layers in b-roll, music, captions. The output is a finished ad you can push straight to Meta or TikTok I made this Makeup Revolution serum ad in under 30 minutes. Not 30 minutes of editing 30 minutes total including writing the brief The play I'm running: -> Source high converting makeup/skincare offers from Glitchy -> Build 3-5 ad variations in Notch testing different hooks -> Post organically and run a ManyChat comment funnel sending people a free "skin guide" with my affiliate link baked in -> Let the best performing creative tell me what to scale One IG page. One funnel. Consistent creative output. That's a $20k-$30k/month business these kind of pages can be launched in a day and scaled in weeks The gap between people winning and losing at affiliate right now isn't the offer, it's the quality of the ad RT + Comment "Notch" and I'll send you the full playbook (must be following for auto dm)
60
36
80
5,915
Aman Jain retweeted
The $1B one-man company the The New York Times just profiled runs a system that got him to 833x growth. 300 users → 250,000. In 12 months. Everyone read the article and picked a side: “It’s fraud.” “It’s genius.” “It should be illegal.” I read it and saw a blueprint. Not the fake doctors. Not the AI-generated faces. Not the testimonials from people who don’t exist. I saw the layer underneath. A production system that turns one brief into hundreds of ad variants — no creative team, no studio, no media agency. One input. Hundreds of outputs. All running at once. That’s not marketing. That’s on-demand ad infrastructure. And it comes down to 5 steps. Step 1 is the architecture decision most brands skip because it doesn’t feel like “marketing.” Get it right → the other 4 become completely mechanical. That’s why he runs 800 ads while most brands struggle to ship 20 a week. Comment “Notch” and I’ll send you the full 5-step breakdown.
146
35
3
175
20,988
Aman Jain retweeted
There's a workflow that turns your winning ad URL into 50 variants. Most teams find out about it after they've already killed their winner. Here's what I kept watching happen. Brand finds a winner. Scales it. Frequency climbs. CTR drops. They brief a new creative round. Wait 3 weeks. By then, the audience is gone, and the budget is burned. The winning ad already had everything needed to prevent that. The hook that worked. The emotional trigger underneath. The pacing held attention. All of it is extractable. All of it is rebuildable into 50 variants in one sitting. Drop the URL. The agent reads the physics. Builds the rotation before frequency becomes a problem. → Winning URL in → Hook swaps generated → Background variations built → CTA rotations ready → 50 variants. One sitting. What took 3 weeks now takes one afternoon. The winner doesn't have to die waiting for the next brief. Comment "variants," and I'll send the exact workflow.
36
19
2
43
5,665
Aman Jain retweeted
2 weeks → 2 minutes. I've been building a system that creates 20 UGC video variations from a single direction. UGC works because it feels like a real recommendation But the turnaround was 2 weeks. So we tested this workflow for a brand. It turned out to be one of the best performing ads for them. All of the direction was given to agentic video that we've been building. This is what the system does: 1️⃣ You give it a direction → product, angle, format (UGC, problem-solution, testimonial) 2️⃣ It handles the heavy lifting → hooks, scripts, structure based on what’s working 3️⃣ It builds the videos → avatars, scenes, voiceovers, edits 4️⃣ You refine in chat → faster hook, new angle, different tone 5️⃣ It outputs multiple variations instantly So your loop becomes: → 1 idea → 20 variations → test → iterate → scale That’s why some brands suddenly dominate the feed. Not better creatives. Shorter feedback loops. In paid ads: Faster iterations → faster learnings Faster learnings → better spend allocation Better allocation → higher ROAS If you want the workflow, comment “UGC” 👇
151
26
4
131
13,621
Aman Jain retweeted
We gave 1 product to 4 different AI actors inside Notch. The data was hard to ignore. Actor 1 — 24-year-old with acne scars. Actor 2 — 38-year-old mom with tired skin. Actor 3 — Dermatologist in a white coat. Actor 4 — Gym guy. Post-workout recovery. The dermatologist had the highest CTR. The mom had the highest AOV. The gym guy unlocked an audience segment nobody was targeting. The 24-year-old had the lowest CPA on cold. Same product. Same offer. 4 completely different winners for 4 completely different buyers. Here's what that taught us: Meta doesn't find your customer. It finds the version of your customer that recognizes themselves in your ad. 1 actor = 1 identity signal = 1 audience pool. 4 actors = 4 identity signals = 4 audience pools Meta now knows how to fill. You're not running 4 ads. You're telling Meta your product is for 4 different people. And Meta goes and finds all 4. 1 product. 4 actors. 4 audiences unlocked. Same afternoon. Drop "Notch" in the comments. I'll send the exact workflow.
28
11
13
40
25,830