It was 2 a.m. on a Wednesday. I had just finished reviewing a quarterly roadmap for an agentic AI initiative that had taken four months of planning, three workstreams and more alignment meetings than I care to count. Then YouTube served me a video of a 24-year-old using an AI agent to set up a working ad agency in 45 minutes.
I watched it twice, because I wanted to think through what it meant for companies like the ones I work with.
What the video shows
Brett Malinowski's video demonstrates Claude's computer-use capability building an advertising agency for car dealerships. In the sequence shown, the agent opens a browser, researches competitors' dealership ad strategies, creates a brand identity guide with logos, colour palettes and typography, and builds a storefront website. It then pulls real vehicle inventory from a dealership, including VINs and prices, and generates ad creatives with legal disclaimers and geo-targeting.
It goes further. The agent logs into Meta Ads Manager and deploys the campaigns, prospects 20 dealerships in the target market, finds their Instagram accounts and sends personalized outreach messages offering advertising services. The whole run takes 45 minutes, with no API integrations or custom code, just an agent controlling a web browser the way a person would.
The context was Y Combinator's announcement that it would fund AI-fulfilled agencies, which puts institutional capital behind the idea.
Why the thesis matters more than the demo
An advertising agency sells strategy, creative production, media buying and campaign management, bundled into billable hours multiplied by headcount. Each unit of revenue needs people, so the margin structure is constrained by labour.
The thesis behind AI-fulfilled agencies is that services may go through a compression similar to the one SaaS brought to on-premise software, with outcomes replacing labour instead of tools replacing tools. An entity that delivers comparable campaigns at a marginal cost approaching zero would be a different kind of business rather than a better agency.
The skill is workflow design
What struck me most is that the person in the video was designing a workflow more than writing code or clever prompts. Each step had clear inputs and outputs, the AI handled execution, and the person handled the sequence and the judgment.
This is the pattern I wrote about in The Orchestration Era. The valuable skill is designing the workflows that connect autonomous agents to business outcomes, which sits where technical understanding meets strategic thinking. The person in the video decomposed agency operations into sequential steps an agent could execute end to end. The AI did the work, and he designed how the work was organized.
That lowers the barrier to building a services business from hiring ten specialists to designing one good workflow.
What this breaks
The traditional agency equation is simple: revenue equals headcount times billable rate times utilization. AI-fulfilled agencies decouple output from headcount, so one person with the right workflow can produce what used to need a team.
I explored the wider version of this in The Autonomous Enterprise. When AI can execute the routine work, value shifts to whoever designs the system. If one person can serve fifty clients, prices fall, and a ten-person agency can no longer compete on cost. It has to compete on judgment, relationships and strategic insight, which are much harder to automate.
Many agencies sell execution presented as judgment. Strategy decks are often templated, the creative process follows predictable patterns, and media buying is already algorithmic. The agencies that survive will be the ones that always sold genuine strategic thinking, and others may find that their moat was friction rather than expertise.
The governance gap
The ads in the video looked good, the copy read as compliant and the targeting was logical. That made me more concerned, because quality at speed without governance is a liability.
The video does not show brand-safety review, legal review beyond basic disclaimers, optimization after launch, client communication or the relationship work that turns a campaign into a partnership. It also leaves regulatory questions open. The EU AI Act's Article 50 transparency obligations, which apply from 2 August 2026, cover situations such as people interacting with AI systems and certain AI-generated content, and a real business would need to check whether its creative and outreach fall within them. I cover how to build that evidence in EU AI Act Article 50: The Label Is the Easy Part.
An agent can deploy a campaign in minutes that a compliance team would take days to review. That speed gap is a risk factor, and the organizations that do well will be the ones that move fast with guardrails in place.
The strongest objection
An agency leader will say that a 45-minute demo is not a business. Clients buy accountability, a relationship, months of performance optimization and someone to call when something goes wrong, and none of that appears in the video.
That is fair, and it is where the governance gap lies. What the demo does show is that the cost of the execution layer is collapsing. Pricing pressure will hit execution first, which is why agencies and internal teams need to be clear about how much of their value is judgment and relationship, and how much is production that an agent can now do.
What I would tell executives
From the agentic AI work at Holcim to the autonomous pipeline behind Pipesignal, three recommendations keep coming back.
- Map your exposure. Many functions are agency-shaped: marketing, sales development, content production, customer support and financial analysis all follow a sequence of research, creation and distribution. Ask which of your internal operations follow the same pattern. The AI Readiness Assessment is a good place to start that mapping.
- Invest in orchestration as well as automation. Training should focus less on how to use AI tools and more on how to break a business process into steps an agent can execute, which is systems thinking applied to autonomous agents.
- Build governance before you need it. The demo impressed partly because it skipped governance, and in a real enterprise that is how a company ends up in the news for the wrong reasons. Brand guidelines, compliance checks and human review for high-stakes decisions are the trust infrastructure that lets AI scale without damaging your reputation.
The bottleneck moves to judgment
For decades, the bottleneck in business was labour. Companies had more ideas than people to execute them, hiring and training were slow, and whole industries sold other people's labour, from consultancies to agencies to outsourcers. That bottleneck is loosening quickly.
The new bottleneck is judgment: what to build and what to leave alone, which outcomes matter, which risks are acceptable and how to keep trust at machine speed. Those are questions about values, priorities and accountability, and they belong with people who can be held responsible for the answers.
Monday move
List three functions in your business that follow the research, creation and distribution pattern. For the one with the largest external spend, write down each step, mark where judgment is exercised and which checks a regulator or client would expect, and estimate how much of the work an agent could execute today. Bring that one-page map to your next planning discussion.