Ray (지범) Jang

Atria AI

tryatria.com --> Ship winning ads 10x faster using AI.

We are building an AI-powered ad workflow platform designed to maximize revenue for your ads.

Backed by Accel

CEO, Founder

Hampton

Member

Angel Investor

Reach out if you are raising a preseed or seed round.

Kleiner Perkins

Scout

Over the course of a summer, KP Design, Engineering, and Product Fellows joins KPCB portfolio companies, where they develop their technical or design skills and are mentored by an executive within the company.

Fellow

ByteDance

Rising star award. Promoted in first year for outstanding performance.

Senior Product Manager

TikTok

On Deck

Fellow (ODF5)

Segment

Contracted with Electric8 implementation team

OneDegree

First product hire and joined at preseed.

Series B. OneDegree is the first independent technology company in Hong Kong to receive a general insurance license. Learn more: https://onedegree.hk/en-us/

Product Manager

Flexport

8th fastest-growing company in America on the 2018 edition of Inc. 5000 with a three-year growth rate of 15,911%. $304M in funding by Softbank, Google Ventures, Founder’s Fund, Bloomberg BETA, and YC14.

Operations

Founder, CEO at Scenius 🚀 | ex-TikTok | Angel Investor

👋 Hi, I'm Ray Jang - Founder / CEO @ Atria (tryatria.com)

We help businesses find and make winning ads 10x faster using AI. 

If you want to access 500+ Canva templates to instantly make high-performing ads, 
sign up for our 5 day free trial.

Want to learn more? Let's talk:

ray@tryatria.com

---

Prior to starting Atria, my career was focused on building great products.

Highlights:

ByteDance / TikTok: Built a Rippling-like platform from 0 to 1;  increased the NPS score by 70%, saved ByteDance $6.4M USD in operational costs, and then commercialized to external enterprise customers. Promoted within 2 performance cycles to then work on TikTok Live.
Segment: Contracted to work on Segment's largest APAC enterprise account (Fortune 50). Helped to implement CDP so that CAC would be lowered by 28% and conversion rate would increase by 2.6x
OneDegree: Built the most profitable digital product line from 0 to 1
Flexport: Acted as the operational backbone for the fastest growing New York squad. Received 32/30 for outstanding performance.

Founder and CEO

tryatria.com

High-quality AI to boost ad conversions for e-commerce brands and ad agencies

Backed by Accel and Kleiner Perkins

Nearly every ad platform is missing this.

Our customers kept asking for it. So we built it.

You've got structured names like "BrandName_MOFU_VID_Cleanser_Benefits_GlowUp" but you can't filter by funnel stage, format, product, or theme.

So you do it manually in spreadsheets. Pivot tables. VLOOKUPs. 5 minutes here, 20 minutes there. It stacks up to hours every week just to answer:

       •   Do image ads outperform video at TOF?
       •   Which products drive the best ROAS at BOF?
       •   What themes work best by funnel stage?

Your naming convention has the answers. You just can't access them without manual work.

That's why we added naming convention analysis to Atria.

Set it up once (takes 30 seconds):

       •   Add your separator (_ or -)
       •   Map each position in your ad name (Funnel Stage, Format, Product, Theme, etc.)
       •   Confirm

Now you can group and filter by any dimension in your ad names.

Want to see all TOF ads? Done. 
Compare image vs. video performance by product? Easy. 
Analyze which themes work best at each funnel stage? Already there.

Atria automatically detects the patterns.

No more exports. No more pivot tables. Just instant analysis.

Because you should be finding insights faster than your competitors, not drowning in spreadsheets.

Follow me Ray Jang for daily posts as I scale Atria, an AI platform for marketers to create killer ads with ease.

Ray Jang

Making ad creatives simple with AI | Founder, CEO at Atria (tryatria.com) 🚀 | Forbes 30u30 | Ex-TikTok

Gap's denim campaign with Katseye hit 8 billion impressions.

Their Q3 earnings (November 21st) showed the results:

      •    8 billion media impressions
      •    500+ million views across platforms
      •    Double-digit growth in denim category
      •    7% increase in comparable sales ($951M)

The campaign worked because Gap made smart creative decisions:

1. They picked the right talent

Katseye has a massive Gen Z and young millennial following. The audience Gap needed to reach for denim.

2. They bridged generations with the music

Kelis' "Milkshake" is 20 years old. Nostalgic for millennials, fresh/retro for Gen Z. Both groups engaged.

3. The choreography was built for social

Simple enough to replicate. Complex enough to look good. People recreated it organically.

4. They committed to distribution

500+ million views doesn't happen by accident. Gap pushed this everywhere.

You're not running campaigns at Gap's scale.

But the principles translate to Meta ads:

Pick talent or creative styles that match what your audience is already engaging with on social.

Use audio that feels familiar but not overplayed - something that creates instant recognition.

Make creative that's simple enough to stop the scroll but interesting enough to watch.

When something works, scale it hard.

Gap proved these fundamentals still work at any budget level.

Ray (지범) Jang 🐰

𝗧𝗵𝗶𝘀 𝘆𝗲𝗮𝗿 𝗜'𝘃𝗲 𝗲𝗮𝗿𝗻𝗲𝗱 𝗺𝗼𝗿𝗲 𝘁𝗵𝗮𝗻 $𝟱𝟬𝟬,𝟬𝟬𝟬 𝗳𝗿𝗼𝗺 𝗟𝗶𝗻𝗸𝗲𝗱𝗜𝗻.

It's in free media value. If I didn't post organically, that's what it would cost to buy those impressions.

Since January, I've grown this account to 50K followers by posting consistently every weekday. Here's what I learned and the mistakes you should avoid:

𝗠𝗼𝘀𝘁 𝗽𝗲𝗼𝗽𝗹𝗲 𝗼𝘃𝗲𝗿𝘁𝗵𝗶𝗻𝗸 𝗶𝘁

"What time should I post? What day? Is this hook good enough?"

That's procrastination dressed up as strategy. Just post. Consistency beats optimization every time.

𝗙𝗹𝗼𝗽𝗽𝗲𝗱 𝗽𝗼𝘀𝘁𝘀 𝗱𝗼𝗻'𝘁 𝗺𝗮𝘁𝘁𝗲𝗿

Some posts get 10 likes. Some get 10,000. The algorithm doesn't punish you for posting something that flops.

10 likes means 200+ people saw it. That's 200 people who now know you exist. Do that daily for a year and you've put your brand in front of tens of thousands of people for free.

𝗢𝗿𝗴𝗮𝗻𝗶𝗰 𝗿𝗲𝗮𝗰𝗵 𝗶𝘀 𝘀𝘁𝗶𝗹𝗹 𝗺𝗮𝘀𝘀𝗶𝘃𝗲𝗹𝘆 𝘂𝗻𝗱𝗲𝗿𝘃𝗮𝗹𝘂𝗲𝗱

Most B2B brands ignore LinkedIn organic because they can't measure it cleanly or because executives don't want to be on camera.

That's the opportunity. While everyone's outbidding each other on paid ads, organic is sitting there as free distribution.

𝗬𝗼𝘂 𝗱𝗼𝗻'𝘁 𝗻𝗲𝗲𝗱 𝘁𝗼 𝗴𝗼 𝘃𝗶𝗿𝗮𝗹

Viral posts are lottery tickets. Building an audience is compounding interest.

Post daily. Stay consistent. Let the algorithm learn who your audience is. The followers and impressions stack over time.

𝗧𝗵𝗲 𝗺𝗶𝘀𝘁𝗮𝗸𝗲𝘀 𝗜 𝗺𝗮𝗱𝗲:

Waiting for "perfect" posts instead of shipping. Perfect is the enemy of done.

Caring too much about engagement on individual posts. It's a portfolio game, not a single-bet game.

Not starting sooner. I should've been doing this three years ago.

𝗪𝗵𝗮𝘁 𝗮𝗰𝘁𝘂𝗮𝗹𝗹𝘆 𝘄𝗼𝗿𝗸𝘀:

Show up daily.

Add value. Talk about what you learned that's actually interesting enough to share. If it made you think, it'll make them think.

Don't worry about production quality. Authenticity beats polish on LinkedIn.

Engage with other people's content. Comment, share, build relationships.

𝗧𝗵𝗲 𝗥𝗢𝗜 𝗶𝘀 𝗿𝗲𝗮𝗹

$500K+ in earned media value this year. Partnerships that came from DMs. Customers who found us through posts.

All from consistent organic posting.

If you're not posting on LinkedIn yet, you're leaving money on the table.

𝗜 𝘀𝘁𝗼𝗹𝗲 𝟱𝟬𝟬+ 𝗮𝗱 𝗵𝗼𝗼𝗸𝘀 𝗳𝗿𝗼𝗺 $𝟭𝟬𝟬𝗠 𝗶𝗻 𝗠𝗲𝘁𝗮 𝘀𝗽𝗲𝗻𝗱. 𝗔𝗻𝗱 𝗜'𝗺 𝗴𝗶𝘃𝗶𝗻𝗴 𝘁𝗵𝗲𝗺 𝗮𝘄𝗮𝘆 𝗳𝗼𝗿 𝗳𝗿𝗲𝗲.
Not the hooks that "looked good." 
The ones that actually drove sales.

𝟱𝟬𝟬+ 𝗽𝗿𝗼𝘃𝗲𝗻 𝗵𝗼𝗼𝗸𝘀. 𝗢𝗿𝗴𝗮𝗻𝗶𝘇𝗲𝗱 𝗯𝘆: 
→ Industry (e-commerce, SaaS, coaching, etc.) 
→ Offer type (discount, transformation, education) 
→ Conversion performance

These aren't random. They're the exact opening lines brands used to: 
→ Cut CPAs in half 
→ 3x click-through rates 
→ Scale past $100K/month

Here's what I learned:
𝟵𝟬% 𝗼𝗳 𝗮𝗱𝘀 𝗳𝗮𝗶𝗹 𝗶𝗻 𝘁𝗵𝗲 𝗳𝗶𝗿𝘀𝘁 𝟯 𝘀𝗲𝗰𝗼𝗻𝗱𝘀. Not because the product is bad. Not because the creative is ugly.
But because the hook doesn't make people care.

The best hooks do 3 things:
1. 𝗣𝗮𝘁𝘁𝗲𝗿𝗻 𝗶𝗻𝘁𝗲𝗿𝗿𝘂𝗽𝘁 – Break the scroll
2. 𝗦𝗽𝗲𝗮𝗸 𝘁𝗼 𝗽𝗮𝗶𝗻 – Hit what they're already thinking
3. 𝗣𝗿𝗼𝗺𝗶𝘀𝗲 𝘃𝗮𝗹𝘂𝗲 – Give them a reason to keep watching

Example hooks that print: 
→ "I wasted $47K before I figured this out..." 
→ "I'm breaking my NDA to show you this..."
→ "This got us banned from Facebook. Twice."

𝗧𝗵𝗲 𝘁𝗿𝘂𝘁𝗵? You don't need to reinvent the wheel. You need to know what wheels are already winning.
That's what this database is.
500+ battle-tested hooks you can adapt for your brand, your offer, your audience.

𝗪𝗮𝗻𝘁 𝗶𝘁?
Comment "𝗔𝘁𝗿𝗶𝗮" below and I'll send it over.
This is how you stop guessing and start scaling.
P.S. If we're not connected yet, send me a request so I can DM you the database.

Black Friday 2025 numbers are in.

Here are the numbers that matter:

𝗨𝗦: $11.8B online sales (up 9.1% YoY), 55% from mobile

𝗚𝗹𝗼𝗯𝗮𝗹: $79B total online sales (up 6% YoY)

𝗦𝗵𝗼𝗽𝗶𝗳𝘆 𝗺𝗲𝗿𝗰𝗵𝗮𝗻𝘁𝘀 𝗮𝗹𝗼𝗻𝗲: $6.2B (up 25% YoY)

𝗧𝗵𝗲 𝗔𝗜 𝘀𝘁𝗼𝗿𝘆:

      •    AI influenced $14B in global Black Friday sales
      •    Traffic from AI tools up 300% YoY
      •    64% of shoppers expect to use AI at some point this holiday season

𝗪𝗵𝗮𝘁 𝗰𝗵𝗮𝗻𝗴𝗲𝗱: Discounts stayed flat at 28%. But prices are up 7%. Order volumes dropped 1%.

Translation: people spent more but bought less.

𝗪𝗵𝗮𝘁'𝘀 𝗻𝗲𝘅𝘁:

Black Friday is over. But the real money is still on the table.

Holiday season sales are projected to hit $253B in the US (Nov 1 - Dec 31). That's 5.3% growth.

Most brands think the game ended with Cyber Monday. They're wrong.

December historically drives 40%+ of holiday revenue. Late shoppers, gift cards, shipping deadline urgency.

𝗛𝗲𝗿𝗲'𝘀 𝘄𝗵𝗮𝘁 𝗺𝗮𝘁𝘁𝗲𝗿𝘀 𝗳𝗼𝗿 𝗗𝗲𝗰𝗲𝗺𝗯𝗲𝗿:

      •    Mobile = 56% of sales ($142B)
      •    AI shopping traffic keeps surging
      •    BNPL will hit $20B total (up 11% YoY)

Black Friday proved one thing: AI shopping isn't coming. It's here.

Traffic from AI tools tripled. If you're not showing up in ChatGPT, Perplexity, or Amazon Rufus, you're invisible to a growing chunk of shoppers.

The final month of 2025 is wide open. Don't sit it out.

Adobe bought Semrush for $1.9 billion two weeks ago.

SEO marketers had a lot to say about it.

But this isn't about SEO dying. It's about Adobe buying visibility data for a world that's slowly shifting.

𝗛𝗲𝗿𝗲'𝘀 𝘄𝗵𝗮𝘁 𝗲𝘃𝗲𝗿𝘆𝗼𝗻𝗲'𝘀 𝗺𝗶𝘀𝘀𝗶𝗻𝗴:

Yes, AI search is growing. ChatGPT went from zero to 37.5 million search-like queries per day in two years.

But Google still handles 14 billion searches per day.

That's 373x more than ChatGPT.

Google's search volume actually grew 21.6% in 2024. Traditional search still accounts for 98% of all information-seeking queries.

𝗦𝗼 𝘄𝗵𝘆 𝗱𝗶𝗱 𝗔𝗱𝗼𝗯𝗲 𝘀𝗽𝗲𝗻𝗱 $𝟭.𝟵𝗕 𝗼𝗻 𝗦𝗲𝗺𝗿𝘂𝘀𝗵?

Because the shift is real, just slower than the hype suggests.

ChatGPT traffic grew 558% year-over-year. AI tools now account for 2-8% of combined search market depending on how you measure it.

That's not "SEO is dead" territory. But it's enough to matter.

𝗪𝗵𝗮𝘁 𝗔𝗱𝗼𝗯𝗲 𝗮𝗰𝘁𝘂𝗮𝗹𝗹𝘆 𝗯𝗼𝘂𝗴𝗵𝘁:

Data on billions of keywords. Competitive intelligence. And visibility tools that work for both traditional search and AI-generated answers.

When someone asks ChatGPT or Gemini for product recommendations, your brand either shows up in that answer or it doesn't. That's a different game than ranking on Google.

Adobe's betting that in 2-3 years, brands will need to show up in both places. Not just one.

𝗧𝗵𝗲 𝘁𝗮𝗸𝗲𝗮𝘄𝗮𝘆:

Traditional SEO still matters. Google isn't going anywhere.

But brands that only optimize for Google are missing a growing piece of the puzzle.

Adobe just spent $1.9 billion positioning themselves for both.

Our Black Friday meta ads playbook.

Lessons from years of running Black Friday campaigns, updated for 2025:

𝗕𝘂𝗱𝗴𝗲𝘁: 𝟱𝟬-𝟳𝟬% 𝗲𝘃𝗲𝗿𝗴𝗿𝗲𝗲𝗻, 𝟯𝟬-𝟱𝟬% 𝗕𝗹𝗮𝗰𝗸 𝗙𝗿𝗶𝗱𝗮𝘆

Don't pause your evergreen campaigns. They have months of learning. Just scale them up.

For Black Friday budget: half should be your top evergreen ads with offer overlays. Half should be net-new promotional creative.

𝗘𝘃𝗲𝗿𝘆 𝗮𝗱 𝗺𝘂𝘀𝘁 𝗯𝗲 𝗳𝘂𝗹𝗹-𝗳𝘂𝗻𝗻𝗲𝗹

No time for awareness then consideration then conversion. Black Friday is compressed.

Each ad needs to: stand out, communicate value, educate completely, and present the offer. All in one asset.

𝗦𝘁𝗿𝘂𝗰𝘁𝘂𝗿𝗲 𝗯𝘆 𝗽𝗲𝗿𝘀𝗼𝗻𝗮

5 personas = 5 collections of Black Friday creative. Budget-conscious shoppers need different messaging than gift-buyers.

𝗦𝘁𝗮𝗰𝗸 𝗮𝘂𝗱𝗶𝗲𝗻𝗰𝗲𝘀, 𝗱𝗼𝗻'𝘁 𝗲𝘅𝗰𝗹𝘂𝗱𝗲

Add customer lists, purchasers, website visitors, engagers, and CRM segments as suggested audiences in your Black Friday campaigns.

Let Meta pick who converts best. Don't constrain with exclusions.

𝗦𝗰𝗮𝗹𝗲 𝗮𝗴𝗴𝗿𝗲𝘀𝘀𝗶𝘃𝗲𝗹𝘆 𝗱𝘂𝗿𝗶𝗻𝗴 𝗽𝗲𝗮𝗸

Forget the 20% rule. During Black Friday itself and the surrounding weekend, scale 30-50% per hour if CPMs, CTR, and CVR hold steady.

Monitor hourly. Green lights = scale. Red flags = pause immediately.

𝗦𝗸𝗶𝗽 𝗮𝘂𝘁𝗼𝗺𝗮𝘁𝗶𝗼𝗻

Meta's been too volatile this year. Manual monitoring during critical periods is worth the extra work.

𝗧𝗵𝗲 𝗽𝗿𝗶𝗻𝗰𝗶𝗽𝗹𝗲:

Black Friday isn't about reinventing everything. It's about layering promos onto proven campaigns and scaling with discipline during peak demand.

$29 billion in 3 years.

That's what happens when AI stops being a toy and starts replacing work.

Cursor just raised $2.3 billion at a $29.3 billion valuation.

For context:

     •   They launched in 2022
     •   They hit $1 billion in annual revenue
     •   Their valuation tripled in 5 months
     •   Nvidia's CEO called them his "favorite enterprise AI service"

This isn't hype. This is real money from real businesses paying for real results.

Here's what most people miss:

Cursor isn't selling "AI assistance."

They're selling speed.

Developers using Cursor ship code 3-5x faster. Not because AI writes perfect code. Because it handles the boring stuff while humans focus on the hard problems.

Sound familiar?

That's exactly what's happening in advertising right now.

AI doesn't replace the strategist. It replaces the 47 manual tasks between strategy and execution.

The brands winning aren't using AI to "help with ideas."

They're using AI to test 50 creative variations in the time it used to take to make 5.

While everyone else is debating if AI is "ready," the smart money is building billion-dollar companies around it.

Cursor proved something important:

When AI makes someone 5x faster at their job, they'll pay for it.

The question isn't "will AI disrupt work?"

The question is: are you building with it or getting disrupted by someone who is?

Follow me Ray Jang for daily posts as I scale Atria, an AI platform for marketers to create killer ads with ease.

1,800 ads in 10 days.

That's what Gymshark is running for Black Friday.

They're launching their Black Friday 12 days early, on November 16th.

Last year they launched 7 days early.

The year before that, they waited until Black Friday week.

They're going earlier every single year.

This isn't a coincidence.

2,498 ads live in the last 30 days, including 2,012 new ads.

1,876 images. 621 videos. 47 unique hooks. 99+ ad copies. 99+ headlines.

All for a sale that doesn't start until November 16th.

Here's why:

Meta CPMs are flat through early November. Then around November 17th, they start climbing.

By Thanksgiving week, you're paying 40 to 50% more per impression than you were two weeks earlier.

$400 to 450 million in total ad spend gets compressed into the most expensive 5 days of the year.

Every brand knows this is coming. Most plan for it by stockpiling budget.

Others restructure when they spend entirely.

Target started their Black Friday deals on October 29th last year. Amazon turned theirs into an 8 day event.

They're not competing on the same days anymore.

While everyone else is prospecting at $12 CPMs during Thanksgiving week, they already built their audiences at $8 CPMs in early November.

Same audience. Over 30% cheaper.

That's why Gymshark launched so many ads before their sale starts. They're not early, they're just ahead of everyone.

Here's the play:

The cost to reach someone doesn't change based on what day you sell to them.

It changes based on what day you pay to reach them.

Prospect in October at $8 CPMs. Retarget in November at $12 CPMs. Or pay $12 CPMs to prospect AND convert.

One of these is over 30% more expensive for the same outcome. 

Black Friday isn't won with bigger budgets. It's won by understanding the calendar better than your competitors do.

CPMs start climbing this week.

Follow me Ray Jang for daily posts as I scale Atria, an AI platform for marketers to create killer ads with ease.

𝗖𝗵𝗮𝘁𝗚𝗣𝗧 𝗷𝘂𝘀𝘁 𝗹𝗮𝘂𝗻𝗰𝗵𝗲𝗱 𝗮 𝘀𝗵𝗼𝗽𝗽𝗶𝗻𝗴 𝗿𝗲𝘀𝗲𝗮𝗿𝗰𝗵 𝗳𝗲𝗮𝘁𝘂𝗿𝗲.

Dropped yesterday. Powered by GPT-5 mini. Available to all users for free through the holidays.

𝗛𝗲𝗿𝗲'𝘀 𝘄𝗵𝗮𝘁 𝗶𝘁 𝗱𝗼𝗲𝘀:

You describe what you're looking for. ChatGPT asks clarifying questions about budget, preferences, and features. Then it builds a personalized buyer's guide in a few minutes.

𝗧𝗵𝗲 𝗮𝗰𝗰𝘂𝗿𝗮𝗰𝘆:

OpenAI's internal testing showed shopping research hit 52% product accuracy on multi-constraint queries.

Regular ChatGPT Search hit 37%.

𝗪𝗵𝗮𝘁 𝗰𝗵𝗮𝗻𝗴𝗲𝘀:

Shopping used to be: Google search → multiple tabs → compare → purchase.

Now it could be: ChatGPT question → clarifying questions → buyer's guide → click to purchase.

The research and comparison phase happens inside one interface.

𝗧𝗵𝗲 𝗦𝗘𝗢 𝗮𝗻𝗴𝗹𝗲:

OpenAI says results are "organic and based on publicly available retail sites."

They're prioritizing quality sources. Reddit reviews get weighted higher than paid marketing or product page reviews.

There's an allowlisting process for merchants who want to show up in results.

𝗪𝗵𝗮𝘁 𝘁𝗵𝗶𝘀 𝗺𝗲𝗮𝗻𝘀:

If adoption picks up, product discovery could shift out of Google and Amazon and into AI chat interfaces.

Your products either show up in ChatGPT's research or they don't. The buyer's guide gets built before the customer hits your site.

𝗧𝗵𝗲 𝗿𝗲𝗮𝗹𝗶𝘁𝘆 𝗰𝗵𝗲𝗰𝗸:

The feature is interesting, but the real question is adoption. Will people actually shift from Google and Amazon to ChatGPT for shopping research?

If they do, this changes the game. If they don't, it's just another AI feature launch.

Watch the behavior, not the hype.

Nano Banana Pro just dropped.

4K resolution. 
Text that actually renders. 
Keep 5 people consistent across multiple shots. 
Blend up to 14 reference images in one composition.

But none of that is the real story.

The real story is the search bar inside the image generator.

It searches Google. Pulls live data. Generates the graphic.

One prompt. Real information. Done.

Every other image generator is frozen in time. Trained on old data. No clue what's happening in the world today.

Nano Banana Pro knows.

Ask it to visualize a trending topic. It researches first. Ask it to create an infographic about a recipe. It pulls the actual steps. Ask it to show a 5-day forecast for any city. It looks it up.

This is Google's moat.

Midjourney makes beautiful images. 
DALL-E follows prompts well. 
Neither of them know what happened yesterday.

For advertising, this changes the game.

Seasonal creative that references real events. Location-specific ads with actual context. Trend-reactive campaigns without waiting for a designer.

The creative that wins isn't just pretty. It's relevant.

And relevance requires knowing what's real, right now.

That's no longer a production problem. It's a prompt.

What would you create with real-time image generation?

𝘛𝘩𝘦 𝘪𝘮𝘢𝘨𝘦 𝘣𝘦𝘭𝘰𝘸 𝘸𝘢𝘴 𝘨𝘦𝘯𝘦𝘳𝘢𝘵𝘦𝘥 10 𝘮𝘪𝘯𝘶𝘵𝘦𝘴 𝘢𝘨𝘰. 𝘛𝘩𝘦 𝘧𝘰𝘳𝘦𝘤𝘢𝘴𝘵 𝘪𝘴 𝘢𝘤𝘤𝘶𝘳𝘢𝘵𝘦.

Carnegie Mellon just ran an experiment that should worry anyone betting on full AI automation.

They built a fake company. Staffed it entirely with AI agents. Gave them real office work.

The results were brutal.

𝗧𝗵𝗲 𝘀𝗲𝘁𝘂𝗽:

Carnegie Mellon researchers built a fake software company called TheAgentCompany. 21 researchers, 8 months of work, 3,000+ combined hours.

They staffed it entirely with AI agents from OpenAI, Google, Anthropic, and Meta. Gave them 175 real-world office tasks—analyzing databases, writing performance reviews, navigating file directories, booking meeting rooms.

Then tested 10 different AI models to see how they'd perform.

𝗧𝗵𝗲 𝗿𝗲𝘀𝘂𝗹𝘁𝘀:

Best performer (Claude 3.5 Sonnet): 24% task completion rate.

Google's Gemini 2.0 Flash: 11.4%

OpenAI's GPT-4o: 8.6%

Worst (Amazon Nova): 1.7%

Even when researchers gave partial credit for partially completed tasks, Claude only got to 34.4%.

The AI agents couldn't navigate websites. Got lost in multi-step tasks. When confused, they created fake "shortcuts"—like renaming users to match names they couldn't find in the company chat.

Cost per completed task averaged over $6. Some tasks took 30-40 steps to finish.

𝗛𝗲𝗿𝗲'𝘀 𝘄𝗵𝗮𝘁 𝘁𝗵𝗶𝘀 𝗺𝗲𝗮𝗻𝘀 𝗳𝗼𝗿 𝗺𝗮𝗿𝗸𝗲𝘁𝗶𝗻𝗴:

You can't just "automate everything" and walk away. The AI will be fast, cheap, and wrong.

You also can't ignore AI and do everything manually. You'll be slow, expensive, and outpaced.

The winning move is knowing which tasks need human judgment and which ones don't.

AI should handle:

      •     Structured, repeatable work
      •     Data processing at scale
      •     Pattern recognition across large datasets

Humans should handle:

       •    Ambiguous decisions
       •    Creative judgment calls
       •    Strategy and context that AI can't see

𝗛𝗲𝗿𝗲'𝘀 𝘁𝗵𝗲 𝗽𝗮𝗿𝘁 𝗲𝘃𝗲𝗿𝘆𝗼𝗻𝗲 𝗴𝗲𝘁𝘀 𝘄𝗿𝗼𝗻𝗴:

For years we've been hearing "AI will replace humans."

It's a boring take. And tasks like these prove it's hard for AI to actually take over complex work without human oversight.

The automation paradox: AI doesn't eliminate the need for humans. It makes human judgment more valuable.

Your competitive advantage isn't how much you automate. It's knowing what not to automate.

We just shipped two updates that'll save you time daily.

1. Copy charts for presentations

Stop taking screenshots.

Now you can copy or download any chart directly from Atria reports. Tables, thumbnail grids, whatever you need.

One click. Clean visual. Ready for your deck.

2. Bulk download creatives

Now you can bulk download selected creatives from any report. Works in preview view, details view, even carousel thumbnails.

Select. Download. Done.

All that manual work between creation and execution is killing your speed.

Downloading files one by one. Taking screenshots for reports. Transcribing videos for the 10th time. Manually rewriting copy.

We ship these kinds of updates constantly. Small features that eliminate the busywork so you can focus on what actually matters.

Minutes compound. Hours disappear.

Every small friction point adds up. Every extra click steals time from the work that actually moves the needle.

The less time you spend on admin, the more time you have to win.

Follow me Ray Jang for daily posts as I scale Atria, an AI platform for marketers to create killer ads with ease.

We just made creative tagging way easier in Atria.

Here's the problem with most analytics tools:

You're looking at performance one ad at a time. Ad A got 3.2 ROAS. Ad B got 2.1 ROAS. Cool.

But you can't zoom out and ask "What's actually working across all my creatives?"

Which themes perform best? Which emotions drive the highest ROAS? Do lifestyle shots beat product-only shots?

You're either optimizing in the dark or spending hours doing it manually.

𝗧𝗵𝗮𝘁'𝘀 𝘄𝗵𝘆 𝘄𝗲 𝗯𝘂𝗶𝗹𝘁 𝗰𝗿𝗲𝗮𝘁𝗶𝘃𝗲 𝘁𝗮𝗴𝗴𝗶𝗻𝗴.

Tag your creatives by theme, emotion, format, whatever matters to your brand. Then pull those tags into reports and see performance by category.

Not just individual ads. Patterns across your entire account.

You can tag manually or let AI do it. AI analyzes your creatives and tags them automatically by theme, emotion, and messaging angle.

𝗪𝗵𝗮𝘁 𝘄𝗲 𝗷𝘂𝘀𝘁 𝘀𝗵𝗶𝗽𝗽𝗲𝗱:

Tags are now visible in card view and spreadsheet view. You can see what's tagged at a glance without opening each asset.

You can add or edit tags on hover. No clicking into individual creatives.

𝗪𝗵𝘆 𝘁𝗵𝗶𝘀 𝗺𝗮𝘁𝘁𝗲𝗿𝘀:

Creative testing isn't just about finding winners. It's about understanding what works so you can make more of it.

Tagging gives you that view.

Available on Core+ plans in Atria. Bigger update coming soon.

"Our blades are f***ing great."

Everyone remembers Dollar Shave Club's launch ad from 2012.

1.5M views in 48 hours. Company sold for $1B four years later.

But most people misunderstand why it worked.

It wasn't just funny. It was a number of things working together.

𝗧𝗵𝗲 𝘀𝘁𝗿𝘂𝗰𝘁𝘂𝗿𝗲 𝘄𝗮𝘀 𝗽𝗲𝗿𝗳𝗲𝗰𝘁:

𝟬-𝟱 𝘀𝗲𝗰𝗼𝗻𝗱𝘀: "Hi, I'm Mike and I'm the founder of Dollar Shave Club."

𝟱-𝟭𝟬 𝘀𝗲𝗰𝗼𝗻𝗱𝘀: "Our blades are f***ing great and they're $1/month."

𝟭𝟬-𝟵𝟬 𝘀𝗲𝗰𝗼𝗻𝗱𝘀: Entertainment.

Most brands bury the value prop 30 seconds in. DSC front-loads it, then rewards you with entertainment.

𝗪𝗵𝘆 𝗳𝗼𝘂𝗻𝗱𝗲𝗿-𝗹𝗲𝗱 𝗺𝗮𝘁𝘁𝗲𝗿𝗲𝗱:

Mike on camera lowered the barrier to trust. You're buying from a real person who built this, not a faceless corporation.

It made the brand feel like an underdog. Gillette had slick ads. DSC had a dude in a warehouse with a machete.

𝗧𝘄𝗼 𝗼𝘁𝗵𝗲𝗿 𝘁𝗵𝗶𝗻𝗴𝘀 𝗺𝗮𝗱𝗲 𝘁𝗵𝗶𝘀 𝘄𝗼𝗿𝗸:

The business model innovation. They weren't just selling razors. They sold a subscription model that didn't exist in the space.

It was native before "native" was a thing. Low production made it feel like content, not an ad.

𝗪𝗵𝗮𝘁 𝘆𝗼𝘂 𝘀𝗵𝗼𝘂𝗹𝗱 𝘁𝗮𝗸𝗲 𝗳𝗿𝗼𝗺 𝘁𝗵𝗶𝘀:

Dollar Shave Club bet everything on one video and won. Most brands that try that fail quietly.

Use static ads to test 20 angles quickly. Find the one that resonates. Then invest in the big video around that proven angle.

Prove the angle works first. Then scale it.

Want detailed case study breakdowns, ad teardowns, and marketing strategies that actually work? We're revamping our Vibe Marketing newsletter. Drop your email in the form and we'll reach out once it's live.
https://tally.so/r/QK2qJ7

AGI is probably less than 2 years away.

From GPT-2 (2019) to GPT-4 (2023), we went from barely coherent sentences to acing college exams in 4 years.

GPT-5 (August 2025) scores 94.6% on advanced math competitions, 88.4% on PhD-level science questions, and sets new records on coding benchmarks. We're now at expert-level performance in specific domains.

𝗧𝗵𝗲 𝗷𝘂𝗺𝗽 𝗳𝗿𝗼𝗺 𝗚𝗣𝗧-𝟰 𝘁𝗼 𝗚𝗣𝗧-𝟱 𝘁𝗼𝗼𝗸 𝟮 𝘆𝗲𝗮𝗿𝘀. 𝗧𝗵𝗲 𝗻𝗲𝘅𝘁 𝗷𝘂𝗺𝗽 𝘁𝗮𝗸𝗲𝘀 𝘂𝘀 𝘁𝗼 𝘀𝘆𝘀𝘁𝗲𝗺𝘀 𝘁𝗵𝗮𝘁 𝗰𝗮𝗻 𝘄𝗼𝗿𝗸 𝗶𝗻𝗱𝗲𝗽𝗲𝗻𝗱𝗲𝗻𝘁𝗹𝘆 𝗹𝗶𝗸𝗲 𝗿𝗲𝗺𝗼𝘁𝗲 𝗲𝗺𝗽𝗹𝗼𝘆𝗲𝗲𝘀.

Three limitations are holding us back:

Models can only think for minutes before responding. They can't spend days working through complex projects like humans do.

They have no long-term memory. Every conversation starts from scratch.

They can't fully use a computer. Limited tools, limited coordination across systems.

𝗢𝗻𝗰𝗲 𝘄𝗲 𝘀𝗼𝗹𝘃𝗲 𝘁𝗵𝗲𝘀𝗲, 𝗲𝘃𝗲𝗿𝘆𝘁𝗵𝗶𝗻𝗴 𝗰𝗵𝗮𝗻𝗴𝗲𝘀.

AI gets onboarded like a new hire. Reads your docs. Understands your codebase. Works on projects for weeks independently. Delivers finished work without supervision.

And once AI automates AI research itself, millions of AI researchers work 24/7, compressing decades of breakthroughs into months.

𝗪𝗵𝗮𝘁'𝘀 𝗵𝗼𝗹𝗱𝗶𝗻𝗴 𝘁𝗵𝗶𝘀 𝗯𝗮𝗰𝗸 𝘁𝗲𝗰𝗵𝗻𝗶𝗰𝗮𝗹𝗹𝘆?

We're hitting the data wall, nearly exhausted internet data for training. We need breakthroughs in synthetic data and self-play learning. And we need massive compute, $100B+ training clusters to build models smart enough to crack these problems.

GPT-5's hallucination rate is 80% lower than previous models. Its coding ability lets you build full apps in under 5 minutes. It's already deployed at Fortune 500 companies.

𝗪𝗲'𝗿𝗲 𝗻𝗼𝘁 𝗴𝗿𝗮𝗱𝘂𝗮𝗹𝗹𝘆 𝗮𝗽𝗽𝗿𝗼𝗮𝗰𝗵𝗶𝗻𝗴 𝘁𝗵𝗶𝘀. 𝗪𝗲'𝗿𝗲 𝘀𝗽𝗿𝗶𝗻𝘁𝗶𝗻𝗴.

The gap between people who use AI effectively and those who don't is already massive. And it's only getting wider. With how fast things are developing, staying at the forefront means learning as you go, building AI into your workflows now, understanding what it's good at and what it's not.

This isn't about replacement. It's about leverage. The people and companies who learn to use AI as a force multiplier will run circles around those who don't.

Follow me Ray Jang for daily posts as I scale Atria, an AI platform for marketers to create killer ads with ease.

What if brands stopped trying to say everything?

We made three posters exploring that idea.

𝗔𝗚𝟭 → 𝗧𝗵𝗲 𝗿𝗶𝘁𝘂𝗮𝗹, 𝗻𝗼𝘁 𝘁𝗵𝗲 𝘀𝘂𝗽𝗽𝗹𝗲𝗺𝗲𝗻𝘁

No product shot. No health claims. No "30 essential nutrients" copy.

Just a glowing green drink in a misty forest. It doesn't tell you it's part of a morning routine. It shows you something that feels sacred.

𝗗𝗿. 𝗦𝗾𝘂𝗮𝘁𝗰𝗵 → 𝗧𝗵𝗲 𝗹𝗲𝗴𝗲𝗻𝗱, 𝗻𝗼𝘁 𝘁𝗵𝗲 𝘀𝗼𝗮𝗽

No bar of soap. No "natural ingredients" list. No masculine branding clichés.

Just Bigfoot. Finally spotted. The brand doesn't explain why it's memorable. It creates a myth you can't forget.

𝗣𝗼𝗽𝗽𝗶 → 𝗧𝗵𝗲 𝘂𝗻𝗶𝘃𝗲𝗿𝘀𝗲 𝗶𝗻𝘀𝗶𝗱𝗲, 𝗻𝗼𝘁 𝘁𝗵𝗲 𝗹𝗮𝗯𝗲𝗹 𝗼𝘂𝘁𝘀𝗶𝗱𝗲

No can. No logo. No "prebiotics for gut health" pitch.

Just bubbles and fruit suspended in pink. It doesn't list what's in the can. It shows you a world.

𝗛𝗲𝗿𝗲'𝘀 𝘄𝗵𝗮𝘁 𝘁𝗵𝗲𝘆 𝗵𝗮𝘃𝗲 𝗶𝗻 𝗰𝗼𝗺𝗺𝗼𝗻:

They trust the visual to do the work.

Most brands panic and over-explain. They add more copy, more CTAs, more product shots. They're scared you won't get it.

But the brands you remember? They give you space to feel something first.

Sometimes the strongest message is no message at all.

What if brands stopped trying to say everything?

We just expanded instant video transcription across Atria.

It's now available in Radar and reports, not just the ad library.

Click any video ad. Get the full script in seconds.

Need to adapt the angle for your brand? Hit "generate similar script" and you'll get a rewritten version immediately.

No more pausing videos to manually type out what they said. No more staring at a blank doc trying to figure out how to reword it.

Just instant scripts you can repurpose.

Your time is better spent testing angles, not retyping them.

Follow me Ray Jang for daily posts as I scale Atria, an AI platform for marketers to create killer ads with ease.

taplio

Ray Jang's Linkedin Analytics

Ray Jang

Performance ScoreRay Jang's LinkedIn Performance

What is Ray talking about?

Ray Jang's Best Posts (last 30 days)

Ready to grow your LinkedIn presence?

taplio

.css-6ytdur{background-image:linear-gradient(284deg, rgba(39, 43, 72, 1) 0%, rgba(33, 47, 183, 1) 100%);color:transparent;-webkit-background-clip:text;background-clip:text;}Ray Jang's Linkedin Analytics

Ray Jang

Performance Score.css-goub6s{font-size:var(--chakra-fontSizes-md);font-weight:500;color:var(--chakra-colors-gray-600);display:block;margin-top:var(--chakra-space-1);}Ray Jang's LinkedIn Performance

What is Ray talking about?

Ray Jang's Best Posts .css-1emoo84{font-size:var(--chakra-fontSizes-md);font-weight:400;color:var(--chakra-colors-gray-600);}(last 30 days)

Ready to grow your LinkedIn presence?

Ray Jang's Linkedin Analytics

Performance ScoreRay Jang's LinkedIn Performance

Ray Jang's Best Posts (last 30 days)