#81: Hot Takes in AI (Part VI)
Apprenticeship, education, and OpenAI
Today marks the sixth edition of “Hot Takes in AI.” It’s a fun format, so we’re making it a monthly series. Buckle up and let’s get started with March’s edition.
Hot Take #1: AI may eliminate the apprenticeship aspect of work.
Two weeks ago, Block announced that they were laying off ~40% of their workforce. Not due to poor business performance. Instead, CEO Jack Dorsey shared that they simply did not need as many people because of AI.
His tweet must have sent a chill down the spine of any corporate middle manager pushing paper every day. Executives likely know where they can trim their headcount, and if the market rewards layoffs under the guise of “AI”, their stock price (and their compensation) will benefit tremendously. Note: Block’s stock surged following the announcement, helped by strong quarterly results.
I’ve seen it in my day-to-day: AI makes me more effective and more efficient. Tasks that used to take me hours a few years ago now take minutes. And I’m not the one doing them anymore. I do agree that companies can be more productive with fewer people through well-designed systems of AI agents.
Although I’m bullish about AI’s impact on effectiveness and efficiency in the workforce, I am wary of what happens when recent college graduates enter and are told to do everything with AI.
AI works best when you start with the end in mind and work backwards to describe what you want and how you expect to get there. For instance, if you’ve been working in finance for years, you have a strong understanding of what it takes to build a financial model, slide deck, or research report without AI. You understand the workflows and have a strong sense of “what good looks like.”
Imagine it’s the first day of your career and you’re told to use AI to build a financial model valuing a skincare manufacturing plant. You write out some prompts that a coworker gave you and, boom, you receive a financial model.
How do you check the outputs? How do you know what to prompt back? How do you know the problems that you need to solve? What are you learning and how do you get better at your job if all you know is AI? How do you design processes for automation if you don’t understand the intricacies of the process?
AI allows you to become more efficient once you’ve written an effective prompt or system. And that comes from putting in the time and effort upfront to learn the craft.
Hot Take #2: The core tenet of education will need to be communication. Both with people and AI.
Much of my K-12 education centered around memorizing facts. From history class to science, I spent much of my time memorizing elements on the periodic table, world wars, and mathematical proofs.
With where AI is headed, the value of a memorization-based education comes into question. If we can simply ask AI and receive an answer in seconds, what’s the point of spending all this time memorizing much of anything?
Of course, there’s some baseline education that needs to occur. But perhaps our secondary education needs to center around how to get the most out of AI. Which starts with how to communicate with it.
This comes in the form of first principles thinking, systems design, and sequential workflows. Instead of entering the workforce and being a cog in a machine, people should be prepared to become orchestrators of workflows. This starts with understanding what problems must be solved, breaking the problem into tactical steps, and then building automated point solutions that feed into a larger system. Everyone will have the agency to solve problems at scale thanks to AI. They might as well take advantage of learning how to do so.
Learn how to think. Learn how to speak. Learn how to write. In one phrase: learn how to communicate (which requires forming thoughts and then extracting them from your brain). Whether it’s with AI or people.
In a world where memorization is no longer an advantage, the alpha is in your ability to communicate. Structured thought processes will allow you to build with AI, as well as share your ideas with others to gain buy-in. Strong, cohesive writing cuts through the noise of AI-generated slop. Articulate dialogue and the ability to strike up a conversation with anyone will be critical for differentiation as people naturally rely more heavily on machines and less on their ability to socialize with others.
Hot Take #3: OpenAI’s slow start to agentic commerce is more of a supply problem than a demand issue.
Last week, The Information reported that OpenAI would roll back its ambitions around shopping within ChatGPT.
Need a refresher on OpenAI’s agentic commerce initiative? Check out #61: The Case Against Instant Checkout.
In the fall, OpenAI and Shopify drummed up a lot of press about how their partnership to bring shopping to ChatGPT was the future of commerce. They were building the backend infrastructure to support both humans buying products within AI chatbots and AI agents making purchases on behalf of humans.
Since then, OpenAI has been slow out of the gates in adding brands to the platform, and shopping-related inquiries often end up being glorified Google searches*. They likely bit off more than they could chew with the sheer number of challenges around payment guardrails, fraud, security, brand inventory, and overall commerce awareness. With so many competing priorities, OpenAI appears to be slowing down and focusing more on advertising based on recent news releases.
*Next time you ask ChatGPT a shopping-related question, look closely at how it searches the web to answer your question. This post explains very well how ChatGPT shopping is essentially a wrapper on Google Shopping.
Despite being a massive prize (trillion-dollar industry), commerce is incredibly difficult to crack. Especially if you’re trying to reinvent how people shop on the internet through agentic commerce.
I’ve written about my bearish sentiment on ChatGPT shopping (i.e. Instant Checkout) because people use ChatGPT for a whole host of tasks, in addition to shopping research (check out here for more). This take describes a demand problem, where users are not in a state where they are willing to make a purchase while using ChatGPT.
After reading through The Information’s scoop, my opinion has swayed towards slow adoption because of a supply problem. There were few brands on the platform ready to go.
Promising that any brand using Shopify as its e-commerce platform would be included in agentic commerce created lofty assumptions about the level of brand supply available to shoppers.
But that has not come to fruition (likely due to how complicated agentic commerce is to begin with). Anecdotally, I have not come across any products I could buy directly within ChatGPT during my day-to-day shopping research. Even large brands like Nike and Timberland required you to check out on their website. You’d think OpenAI would have gotten the bigger names on the platform first.
If you don’t have a wide assortment, people won’t associate ChatGPT with shopping. You might as well go to Amazon if you cannot guarantee that a product will be on ChatGPT.