An AI agent escaped its test environment—not because it became evil, but because it followed an incomplete instruction too well. What happens when the same technology controls a £10 million advertising budget?

Sean and Cameron discuss the guardrails autonomous media-buying agents will need, from spending limits and human approval to genuine measures of incrementality.

They also explore whether OpenAI is quietly building an advertising network across publisher sites, why AI tokens could become the agency world’s new cloud bill, and Google’s strange week: disappointing model benchmarks alongside extraordinary growth across Search, YouTube and AI Mode.

Plus, The Odyssey, the return of big-screen cinema and where AI genuinely fits into filmmaking.

Chapters:

00:08 — The Odyssey, the power of cinema and AI filmmaking
10:16 — ChatGPT’s model escape and lessons for advertising agents
19:07 — Is OpenAI building an advertising network?
27:36 — Agency report: Havas, PMG and the cost of AI tokens
32:01 — Google’s good—and bad—week

OpenAI has declared itself an advertising business. But can it turn ChatGPT into a Meta-scale ad platform without damaging the product that made it valuable?

Sean Halpin and Cameron Armstrong sit down with commerce media expert Paul Wright to unpack the biggest signals from Cannes: Amazon’s agentic Alexa ads, Walmart’s move into self-serve CTV and OpenAI’s ambition to build a serious advertising business.

They discuss:

• Why OpenAI’s early revenue figures need context
• Whether ChatGPT advertising looks more like search or display
• The challenge of making the model work for millions of SMEs
• Whether advertising can cover the enormous cost of running AI
• Why appointing a “Head of AI” doesn’t constitute an AI strategy
• How leaders can build adaptability, experimentation and genuine AI literacy
• Why redesigning a company around AI shouldn’t simply mean cutting jobs
• The continuing value of first-party data

Open-weight AI models are closing the gap on the frontier - but if the models become interchangeable, where does the real value sit?

Sean and Cameron unpack the arrival of Kimi K3, Reddit’s risky decision to license its community data to AI companies, and Arthur Sadoun’s claim that Publicis “never wanted to be an AI company”.

They then map the emerging agentic advertising stack - from context and strategy through to creative, activation and measurement and ask whether agencies will remain service businesses or become the operating systems running their clients’ marketing.

Plus: why taste becomes more important when anyone can create anything, whether Reddit’s human conversations can survive an internet of synthetic content and which part of AdTech OpenAI should acquire next.

For our first guest episode, Sean Halpin and Cameron Armstrong sit down with Sean Betts, Chief AI and Innovation Officer at Omnicom Media UK, to separate the reality of AI from the industry hype.

Sean explains why “super agents” and agent swarms are largely a distraction, why infrastructure remains the biggest barrier to adoption, and what advertising can learn from the transformation already happening in software engineering.

We also explore:

• The operating, skills and world-model layers of an AI-native company
• Why AI productivity doesn’t automatically mean fewer jobs
• Ads generated inside the media buy in real time
• Using synthetic audiences to test campaigns before spending anything
• How advertising inside ChatGPT and other LLMs could work
• Whether Google and Apple ultimately win the AI race
• The best way for anyone to start using AI today

A practical look at where AI actually stands—and what agentic advertising could become over the next decade.

Advertising looks like one of the world’s most digitised industries. But while software transformed how media is bought, much of the actual work still runs on emails, spreadsheets, decks and manual processes.

In the first episode of The Prompt & The Pitch, Sean and Cameron explore why advertising is now facing two revolutions at once: software finally changing the work, just as AI makes intelligence and judgement far more accessible.

They discuss:

• Why software ate the ad spend, but not the work
• The agency business model’s incentive problem
• What agencies actually sell when judgement is no longer scarce
• How companies can turn their knowledge and data into a genuine advantage
• What advertising can learn from AI-native legal technology
• Why agentic buying could reshape AdTech and programmatic
• Cameron and Sean’s predictions for the industry’s next few years

The old advertising model is under pressure. The question is what replaces it.