Idea to Live Ads in 3 Hours. Then the Token Bill Came.

The Briefing

Five things this week that matter if you're shipping with AI.

Meta Advantage+ is writing your creative for you now. AI dubbing, AI-generated music, and persona-based creative adaptation rolled into Advantage+. Sixty-five percent of advertisers are already scaling through it. If you run Meta ads, stop producing 14 versions of the same video for 14 markets. Meta does it.

Google killed Dynamic Search Ads. AI Max is default. Out of beta. Rolling globally. Google's own numbers: about 7% more conversions at similar CPA with the full feature suite on. Your Search campaigns are migrating whether you scheduled it or not. Audit your accounts this week.

TikTok Symphony now generates full branded video. ByteDance's Dreamina 2.0 model is inside Symphony for advertisers. Text in. Branded in-feed video out. If you've been putting off TikTok because production cost was too high, this is your green light to run a test budget.

Anthropic launched Claude Design. Prompts become prototypes, slides, one-pagers, ad creative — applied to your design system. Figma's stock wobbled on the news. For solo operators this removes the "I need a designer" bottleneck for landing pages and ad creative.

ChatGPT Ads rolled into Canada, Australia, New Zealand. Cost-per-click. Small channel right now. But it's the first generative-native ad product from a major platform. This is where part of your search budget is going in 18 months.

The Build

Phew, a lot this week.

I went from idea to concept to website to operational site running ads in about 3 hours. Conversions hit within two days.

I also had multiple "F**K you" moments. Swearing at my screen. Wondering why I ever opened the AI pandora box.

Both things are true at the same time. That is the job now.

The wins were real. A live site shipping in an afternoon. Cold traffic converting inside 48 hours. That compression used to take me two weeks. It used to take a team.

The pain was the token burn. I'd hit a flow, batch through iterations, look at the usage meter, and feel my stomach drop. So I switched. Tried DeepSeek. Tried other platforms. Ran the same job across a few models to see what actually held up. I'm still pondering which one is best for what.

The deeper problem isn't tokens. It's that there's no one to ask.

When I started my product launch career I came home to Australia and realized I was lightyears ahead of the local market. Nobody to learn from. You had to figure it out, test it, write it down, and eventually become the person others would ask.

It feels exactly the same right now with AI. The experts who do exist are high-level — people like Jacob Bank, who has been super helpful to me. But most of what shows up online is marketers pretending to be experts. They teach the what. They hold back the how — because the how is what they're trying to sell you.

So we are all figuring it out on our own. Burning tokens. Breaking things. Writing it down.

Here is the prompt I've been using to compress the old "idea to live ads" cycle down to a single working session. It is not polished. It is the one that actually works for me this month.

You are my launch partner. I have an idea. Help me get from idea to live ads today.

Here is the raw idea:
[2-3 sentences]

Here is the target buyer:
[Specific avatar — title, #1 pain, desired outcome]

Here is the offer:
[One sentence — what they get, price, guarantee]

Produce in this exact order:

1. One-line positioning:
   "[Audience] who want [outcome] without [pain]
   use [thing] to [result]."

2. Three hook angles: direct, curiosity, contrarian.
   One line each.

3. A 6-section landing page outline
   (hero, pain, proof, offer, FAQ, CTA)
   with one sharp line per section.

4. Three ad concepts for the strongest hook.
   Format: Hook → 3-line stack → CTA.

5. One "what would break this" stress test.
   The 3 most likely failure modes and how
   to spot them in the first 48 hours of traffic.

Rules:
- Specific numbers, not vague claims.
- Short lines. Rhythm over polish.
- No corporate language. Write like I'm shipping tonight.
- If you need info I didn't give you, ask. Don't invent.

What worked: Speed. Brutal, uncomfortable speed. Three hours from a blank page to a site with traffic. Two days to conversions. That compression changes what's possible for a solo operator. It also killed my instinct to polish — the stuff that converted was the ugly v1, not the imaginary v4 I never would have shipped.

What didn't work: Token economics. You burn real money learning where a model gets stuck and where it shines. DeepSeek was cheap but shallow in the spots I needed depth. Claude was deep but expensive when I iterated hard. Other models sat in between. The honest answer after a week of testing: there is no single "best" model. There are right-model-for-right-job decisions, and you only learn them by burning tokens.

Which is the actual insight. The edge right now is not the tools. It is having done the reps. And almost nobody has.

The Play

If you're reading this and wondering if you're behind — you are not. You're the 1%.

Most people are still on the sidelines, watching YouTube videos from marketers teaching the what and selling the how. You're actually in the work. Burning tokens. Breaking things. Shipping ugly v1s that convert.

Keep going. Lean in hard. Become the expert that doesn't exist yet.

Try this today. Two moves.

One: set a token budget that stings — $200, $500, pick the number that hurts. Spend it this week building one real thing. Not a demo. Not a course prototype. A working agent that moves a number in your business.

Two: open a blank doc on Sunday. Write one paragraph on what that build taught you that no YouTube guru is saying. Hit publish. Do it again next Sunday.

In 12 weeks you'll have shipped 12 agents and 12 posts. That's the stack of an expert. That's how the 1% compounds.

The 1% are leaning in? So are we. → AgentCMO.io