Products got easy. Customers did not.

More than 80% of the code Anthropic merges is now written by its own AI. At the start of 2025 it was a few percent.

For as long as there have been startups, building the product was the slow part. It is not anymore.

So the scarce thing has moved. When anyone can build, the product stops being rare. The customer becomes rare. And the work of winning customers is still done the way it was before any of this happened: by hand, in pieces, one handoff at a time.

Is customer acquisition getting harder? The free part of it is, sharply. Far more gets built, and on searches where Google now shows an AI answer, clicks on the top result have fallen from 7.3 in every hundred searches to 1.6.

Everyone can build now

On Lovable, a tool that turns a description into a working app, people started about 41,000 projects a day in its first months. This year the figure is 148,000 a day.

Figure 1

More gets built. Fewer clicks are free.

AI-built projects a day

202541,000

2026148,000

Clicks per 100 searches

20237.3

20251.6

Every one of those projects needs the same thing next. It needs someone to notice it.

And being noticed just got harder

Look at the right side of that chart. The cheapest customer a company could win was the one who searched, saw a result and clicked. Where Google now answers the question itself, that click has mostly gone.

More gets built. Fewer clicks are free. That is the squeeze.

Winning customers was always the bigger bill

It is tempting to treat marketing as what a company does after the real work. The accounts say otherwise.

Figure 2

Four of five spend more on selling than building

Sales and marketing against R&D, latest year

SellingBuilding

Salesforce$14.3B$6.0B

Adobe$6.5B$4.3B

ServiceNow$4.4B$3.0B

HubSpot$1.4B$0.9B

Datadog$1.0B$1.5B

Salesforce spends more than twice as much on winning customers as on building what it sells. Four of these five companies spend more on selling than on the product.

The bigger bill is the one AI has not rebuilt yet.

It runs on a formula

Strip away the tools and the job titles, and paid acquisition comes down to three things.

Figure 3

Winning customers is a formula

QualityAds worth running

×

VelocityFind the winners

×

OptimizationCompound what works

Each one capped by people's hours

Quality. Is the ad worth running? The right buyer, the right message, and the right signal telling the platform what a customer looks like.

Velocity. How fast are the winners found? Nobody can pick a winning ad in advance. Winners are found by testing, so the team that tests more finds more.

Optimization. Does each result make the next round better? That means knowing which results are real, moving the money to them, and carrying what was learned into the next campaign.

Whatever a team buys, builds or hires for, it is trying to move one of these three.

Each part has the same ceiling: hours

Good research takes days a team does not have. Every extra ad to test costs a designer's afternoon. And the results pile up faster than anyone can read them.

Those hours are already spoken for. DoubleVerify's 2025 research found campaign managers spending 26% of their time, more than ten hours a week, on manual changes: adjusting bids, moving budgets, resetting limits.

That is a day and a quarter every week of optimization done by hand, one setting at a time.

AI lifted the ceiling on building. It has barely reached this.

Figure 4

AI money went to coding first

Company spending on AI in 2025

Coding$4.0B

IT operations$0.70B

Marketing$0.66B

Customer success$0.63B

In 2025, companies spent $4.0 billion on AI for coding and $660 million on AI for marketing. A sixth.

Marketing does have AI. Every ad platform has automated its own corner, and each one reports on its own results. There is an assistant on sale for almost every single task.

What is missing is the thing that changed coding. Coding did not change because every step got its own assistant. It changed when one system could take a job from request to working result, check its own work, and go again.

From a relay to a loop

Today the work is a relay. An idea is handed from person to person and tool to tool, and every handoff is a wait.

The argument of this series is that AI can raise quality, velocity and optimization together, and that it happens only when the three run as one loop: make the ads, test them, read what came back, and start the next round from what was learned. A person approves the decisions that spend money. The rest runs without waiting.

That is the direction we are building in at Dynamic Ad.

The next nine articles go through it piece by piece, with the evidence:

• How the work is done today, handoff by handoff (2).
• Quality: what the machine is told to chase (3).
• Velocity: why finding winners is a numbers game (4).
• Optimization: how to tell a real result from a lucky one (5).
• The technology: what AI needs to see, how agents run a whole job, and which decisions stay with a person (6, 7, 8).
• What it means for agencies, and for anyone choosing tools (9, 10).

What to do on Monday

1. Name your weakest part. Quality, velocity or optimization. The answer usually comes fast.
2. Count the hours. How much of last week went to keeping campaigns running, and how much to making them better?
3. Count the handoffs in one campaign. From the idea to the first result: how many people, how many tools, how many days of waiting. That count is the next article.