Last week Y Combinator said it would start investing in AI-fulfilled agencies. Agencies where AI does most of the client work.
Not SaaS. Not AI tools. Agencies.
Most people scrolled past that announcement. Brett from The Brett Way did the opposite: he gave himself 24 hours to test whether it was actually true, by building an ad agency for local car dealerships from scratch.
His goal: a $10K/month revenue-ready agency. Brand, strategy, ads, storefront, outreach. All of it.
The core of it took him about 45 minutes.
I watched the whole video and took notes, because this is the clearest demo I've seen of what YC is betting on. Here's the full breakdown, with the numbers.
The setup: no APIs, no logins, just a browser
One detail matters before anything else.
Brett didn't wire up APIs. He didn't hand the AI his passwords. He didn't build some custom integration stack.
The agent simply used his browser. It sees what's on screen, takes control, and does the work in the same tabs he'd use himself.
That's the unlock. The old version of "AI automation" meant weeks of Zapier chains and API keys. This version means you type a sentence and watch your cursor start moving.
He picked local car dealerships as the niche for a simple reason: they already spend heavily on ads. Every dealership runs promotions, every promotion needs creative, and most of them are paying an agency for it right now.
Good niche selection is half the business. Pick people who already pay for the thing.
Minute 1: a full brand style guide
First prompt: build me a custom brand style guide so our ads look unique and consistent.
In under a minute he had a premium automotive brand guideline. Color palette, typography, tone, visual rules for every ad going forward.
No Canva. No designer. No $2K brand package from a freelancer.
Is it as good as a top brand studio? No. Is it better than what 90% of local agencies hand their clients? Honestly, probably yes.
Minute 10: a strategy deck built from 20 competitors' ads
This is the part that made me sit up.
Brett asked the AI to scan 20 different car dealership accounts in the Meta Ad Library and extract the tactics and strategies they're actually running.
The agent opened the Ad Library in his browser, searched dealership after dealership, indexed thousands of live ads, and came back with a six-page strategy doc:
- Seasonal sales events
- Stacked financing offers
- 0% APR patterns
- Trade-in bonuses
- Public-notice style creatives
In less than 10 minutes he had a playbook built from what top dealerships in America are running right now.
Most real agencies never do this research. They recycle the same three templates. The AI did competitive analysis at a depth most shops skip, in the time it takes to make coffee.
The offer: $2,500/month, with a checkout link
Next he asked for a store page and a checkout link so he could actually collect money.
A few minutes later: a storefront positioned as "the AI-powered automotive ad engine", clear pricing, a booking calendar, and a checkout link for $2,500 per month.
Do the math with me.
→ 4 clients at $2,500/month = $10K/month. That was the goal.
And the fulfillment cost is close to zero, because the same agent that built the brand also makes the ads.
That detail (a real checkout link and a calendar, not "let's hop on a call sometime") is what turns a demo into a business. When you reach out to a dealership, you're sending a page with a price and a book-a-call button. That's a real offer.
The real test: ads for an actual dealership
A brand guide and a strategy doc prove nothing if the ads are bad.
So Brett took McCarthy Auto, a dealership from his childhood hometown in Kansas, dropped their website link into the AI, and asked it to create custom ads using the brand guide and the strategy doc.
The agent pulled actual inventory from the site. Real vehicles, real VIN numbers, real pricing, real sale info. No hallucinated cars.
The output was proper direct response: urgency, scarcity, concrete offers, all following the brand guidelines.
Brett's a marketer. His verdict: "These are ads I would absolutely run, and they're probably better than what I would come up with this quickly."
That's the sentence agencies should be nervous about.
Then it set up the campaign in Ads Manager
The step everyone assumes still needs a human: campaign setup.
Brett typed one sentence. "Now upload these into my Meta ad account and set up the ads."
The agent opened Ads Manager in Chrome and filled everything in. Uploaded the creative, wrote the headline and copy, added the correct link. It even knew the dealership was in Kansas and set geo-targeting to a 25-mile radius around the actual location without being told.
All Brett had to do was hit publish.
If you run any kind of service business, this is the moment to pay attention to. Fulfillment, the expensive part of every agency, just became a workflow. I wrote about this shift in my piece on starting an AI services business, and the gap between that article and this demo is maybe six months of model progress.
Then it did the sales outreach too
YC's thesis is about AI doing fulfillment. Brett pushed further: can it do sales?
He asked for a spreadsheet of 20 car dealerships within a 20-mile radius, with their Instagram handles. Two minutes later, done.
Then: "Open my Instagram, search each dealership, and send them a message pitching my services."
He sat there, hands off, watching his own monitor as the agent searched each account, hit the message button, and sent the pitch. 20 DMs, zero manual work.
One honest caveat Brett flagged himself: automating cold DMs likely brushes against Instagram's terms of service, so treat that part as a proof of concept, not a growth channel. The spreadsheet part alone (lead list plus handles in 2 minutes) is worth stealing.
So in under 45 minutes: brand, competitive strategy, storefront with a $2,500/month checkout, real ads for a real dealership, campaign set up in Ads Manager, and 20 outbound pitches sent.
End to end. One person with a browser.
What this means if you're a bootstrapped founder
Brett's line at the end stuck with me: "The bottleneck is no longer labor. It's thinking."
Here's how I'd translate that for the founders I talk to every week:
→ Service businesses just got margin. If fulfillment drops to near zero, a solo operator can run what used to need a 5-person team. That's the same pattern as the systems-thinking approach Brett showed for automating your own job, applied to client work.
→ The playbook is nicheable. Car dealerships were Brett's pick. Swap in dentists, gyms, roofers, med spas. Any local niche that already buys ads works, and the same workflow rebuilds itself per niche from a single website link.
→ Speed is the moat, briefly. Right now, almost nobody is doing this. In 18 months, everyone will be. The founders who build business automations into their operations this year get the compounding head start.
→ It's a wedge, not just an agency. Land dealerships at $2,500/month for ads, then expand into their CRM, their follow-up, their reviews. The agency is the entry point; the operating system is the business.
"Can AI run a business" is settled. Brett showed it can. The open question is which founders redesign their business around that fact first.
FAQ
What is an AI-fulfilled agency?
An agency where AI handles most of the client deliverables (research, creative, campaign setup, reporting) while a human owns the relationship and the strategy. Y Combinator announced it's investing in this model, which is what triggered Brett's experiment. The economics look like an agency's revenue with software-like margins.
Did Brett really build it in 45 minutes?
The core workflow, yes: brand guide, competitor research across 20 Meta Ad Library accounts, storefront with a $2,500/month checkout, ads for a real dealership, campaign setup in Ads Manager, and 20 outreach DMs. He compressed the footage for the video, but the tasks themselves each took minutes, not days. What it doesn't include: signed clients. Revenue-ready is not revenue.
What tools did he use?
A browser-controlling AI agent (Claude driving his screen) plus his normal accounts: Meta Ad Library for research, Ads Manager for campaign setup, Instagram for outreach. No APIs, no custom code, no logins handed over. The agent works inside the browser he's already using.
Is this a real business or just a demo?
The fulfillment is proven; the business isn't, yet. Brett built a revenue-ready agency, not a revenue-generating one. The missing piece is the oldest one in the book: getting a dealership to say yes and hand over $2,500. That still takes a human who can sell. The difference is that once one client says yes, delivery scales without hiring.
Could a non-technical founder copy this?
Yes, and that's the point. Every step was a plain-English prompt in a browser. You don't need to code. You need enough taste to judge the output: what a good brand guide, a good offer, and a good ad actually look like.
Steal the playbooks before they're obvious
Every week on the Profitable Founder Podcast I sit down with bootstrapped founders doing $100K to $10M a year and pull out exactly how they did it. The AI-agency wave is creating a new batch of these stories right now.
If you want the playbooks while they're still early, listen in.
