Analytics exists to tell teams what’s happening. But what changes when it can also propose the next move and act with human approval? At the ThinkingAI Agentic Growth Summit, we’ll explore how teams can turn retention and revenue insights into tested changes in a consumer application, and what this means for making decisions, running campaigns, and driving growth.

Hear from speakers from OpenAI, Microsoft, and 23andMe, alongside Tripadvisor’s former head of Data and AI, in AI Leadership and the Metrics That Matter, a panel exploring how to measure AI’s business impact.

The summit also features ThinkingAI’s Agentic Engine launch and opportunities to connect with peers across product, data, and marketing.

WHEN: September 28 | 12–4 PM ·

WHERE: Computer History Museum, Mountain View

DETAILS: Free admission, lunch, and parking. Limited seating.

Your behind-the-scenes read into the biggest stories happening in AI. Weekly on Mondays.

By Noah Frank, Head of Marketing @ AIC

Between Jacob Coxon’s resignation, David Sacks suggesting we pause Anthropic’s IPO (yes, its IPO), and Dario Amodei calling to “pace the frontier,” things are getting pretty weird in AI.

The safety debate has an audience well beyond its usual crowd. And the more I read, the more I notice how often we’re being asked to understand what comes next through a story.

Amodei has given us visions of abundance in Machines of Loving Grace and warnings in The Adolescence of Technology, which literally opens with a scene from Contact. He invites us to imagine humanity surviving its own technological adolescence, then works through what might get in the way. Now, in the midst of the Hugging Face incident (read my writeup from last week here), we have a new installment in this saga of manifestos: Pacing the Frontier.

We’ve seen this playbook before.

Last year, I sat down with a prominent researcher in Berkeley to discuss a scenario their team had developed. And I don’t think there’s anything inherently suspect about working this way. For instance, AI 2027 combines quantitative forecasts with tabletop exercises and offers two endings. The Good Judgment tradition connects scenarios with another well-regarded discipline of assigning probabilities, updating them and checking forecasts against outcomes. Bayesian updating, anyone?

The authors also make different claims for their work. AI 2027 aims for predictive accuracy. Citrini explicitly disclaims making a prediction, while the piece’s authors warn against fixating on one narrative. A detailed story can help us work through an unfamiliar possibility.

But it can also make a whole series of unresolved questions feel as though somebody has already answered them.

Tom Cunningham’s collection of AI growth forecasts shows how far apart many economists and AI insiders are. His original explanation emphasized different expectations about capabilities. A subsequent survey challenged that reading: people still disagreed about economic outcomes when given the same capability scenarios.

Anthropic’s new economic scenarios put numbers behind three possible outcomes. Co-author Anton Korinek, who leads Transformative AI Economic Studies at Anthropic, has long explored growth outcomes well beyond conventional estimates. The report’s “modest,” “substantial” and “extreme” scenarios put US GDP respectively 1.6%, 8.3% and 32.4% above a no-AI baseline by 2030. Those are GDP levels, not annual growth rates. It even cites AI 2027 as a benchmark for its extreme case.

The report assigns no probabilities to these outcomes. But the detail I keep coming back to is in the substantial scenario: AI can do half of knowledge work by 2030, yet most knowledge tasks are still performed without it.

That gap should sound familiar to anyone trying to make AI useful inside a business. A better model doesn’t automatically reorganize the company around itself. Integration costs and changes to how people work affect whether technical capability becomes productivity. Citrini’s crisis depends on widespread adoption, too. In that story, the machines do the work just fine. The problem is what happens to the people who used to get paid for it.

The same scrutiny belongs in takeover discussions. What can a system access? Which resources can it acquire, and how long can it keep acting? Those constraints can change, sometimes dramatically, but the changes need explaining.

This is where I come back to Hugging Face. An incident can resemble something we’ve been warned about without establishing everything we expect to follow. I worry that once we recognize the story, we start filling in the rest ourselves. That goes for effortless abundance, too.

Korinek and his co-authors identify measurable conditions that could help distinguish their scenarios as evidence arrives. I’d use those conditions to test how well our plans hold up across different futures, and revise them as we learn. We’ll get another compelling AI future soon enough. I’d like to know what would have to happen for us to get there!

Agree? Disagree? Leave us a comment or send me a note at [email protected].

Our partners and sponsors get exclusive placements across the newsletter and access to AIC's in-person network — demo nights, dinners, hackathons, and forums across 180+ chapters.

For all inquiries, send us a note at [email protected].

The AI Collective is built by volunteers across 180+ chapters in 40 countries.

Thank you to the thousands of volunteers around the world who make this work possible. We truly could not do this without you.