Automation isn’t waiting for an intelligence threshold — it arrives task by task, wherever the agent costs less than the person. Which means the AGI debate matters less than everyone thinks, and understanding your own work deeply enough to redesign it matters more. The report: https://www.goldmansachs.com/pdfs/insights/goldman-sachs-research/an-ai-job-apocalypse/report.pdf We’re building a course on working with AI while staying the author of your own work — link below. And join us at the Artificiality Summit, October 22–24 in Bend, Oregon. #aiandjobs #tokeneconomics #aiagents #stayhuman #futureofwork

I just read a token economics report from Goldman Sachs. The idea is pretty simple. Tokens are getting cheaper, so more human work clears the ROI of barf automation. The numbers may get sort of really concrete, like a coding agent costs about $13 a day, but the code that it replaces costs $300. Corescentra agent costs $93 a day and the human costs $90. So the Corescentor numbers are close, and the coding numbers are not close at all. That gaps with understanding, 'cause it means if you're expensive and your work is made of text, this is something companies want to automate. So why did coding tip first? Well, three things lined up. Codes are expensive, $300 a day is a big prize. Codes text and text is what these models are made of. The tools were ready. Coding has a mature stack for running agents and checking their output and catching failures. Voice work has less of that, yet, which is why the call center agent still costs more than the person. But notice what's protecting the call center worker. Partly, that voice is hard and partly that they're cheap. Being paid $90 a day is a kind of protection. They should tell you where this goes. It's not very comfortable. Because token prices fall every year, work that wasn't worth automating last year becomes worth automating this year. The economics just keeps climbing the pay scale towards more expensive people. But there's another insight that I thought was really interesting here. Agents are the new AGI. The whole AGI debate was really a proxy for one question. When will AI be capable enough to replace a human work? And I think that question just got replaced. Automation isn't waiting for some intelligence threshold. It kind of arrives task by task now, whenever the agent costs less than the person and works reliably enough. $13 a day versus $300 is not an intelligence claim. It's actually a price claim. So the AGI timeline arguments don't matter much anymore. The labor impact is being priced in tokens one workflow at a time. And that intelligence is no longer scarce. If you look at the leaderboards right now, Google's flagship sets behind Anthropic Open AI and the Open Weight Chinese models. Open models trade blows with closed ones, frontier advantage losses, weeks. Demosus services said for a while now that the future is smaller, more specialized models. I've always thought that, not the giant ones. So we're gonna see smaller things at the edge. That's a good thing. So when something stops being scarce, it stops being where the value is and where does the value go now? Now, value accrues to whoever owns this scarce resource. We know that if AI expertise is commoditized, the value moves to that scarce complement, proprietary data, domain knowledge, knowing how stuff actually works. The Golden Report even says this with unique data sets becoming more valuable as AI improves, not less. Everyone permits to agents that drives more specialization everything including data, including the knowledge of the work itself. What this means for you is that if intelligence is cheap, that scarce understanding about your work, especially how to redesign it for agents becomes valuable. Because that's happening in every company. All the workflows are redesigned around agents. Every redesign is where the human sits. Design is vital and it's vital to be engaged and stay connected. Because what's happening is we're deciding where the judgment actually stays. And it's the time to use it and the authority to use it. So that's the question our research is about how people stay the authors of their own minds while they work with AI. So be the person who really does understand the work, the person who designs that workflow, thinks about new ways to do it, and how to put these more specialized models to your advantage and launch a new certification in this. And how to work with AI in a way that keeps your judgment yours that allows you to craft and create work. Two days in person, four cities in the US during the fall. So stay posted and DM me if you want more information on this soon and we can connect. I'm Hannah from Artificiality. Follow me for more on Staying Human with AI.