#36: Hot Takes on AI (Part I)
Thoughts and predictions
Mixing it up from the usual format today. Here are a few of my takes on the current state of AI.
Heat Level: 🔥
AI probably won’t take your job in the short-term, but finding a new role may be challenging as companies stall hiring plans and rely on AI instead.
Earlier this week, the WSJ released an article interviewing various CEOs about how AI will affect broader employment trends. The headline is click-bait, but the general ethos of the read is reasonable: AI will affect employment growth, which I agree with.
Companies will be a lot pickier on approving new roles if AI fills the void effectively. In fact, Shopify’s CEO Tobi Lutke recently sent a company-wide memo that detailed how employees are now expected to prove that AI can’t do the job before requesting additional headcount.
Spoken like a true fiduciary of shareholder value, Lutke knows that effective implementations of AI will streamline workflows and allow Shopify to grow revenues with a leaner workforce. As dystopian as it may sound, AI doesn’t ask for salaries or health insurance. Why wouldn’t companies consider process automation over hiring more people that are very likely more expensive than technology? It’s a no-brainer.
Unless you’re working in a call center or have a data entry position, your role is probably safe for now. Plenty of companies are slow to adopt AI and are not sure to start in terms of implementation. But switching companies may get harder if you’re behind the curve on using AI.
I find the following quote from Andrew Ng (co-founder of Coursera) quite fitting for this topic: "AI won’t take your job. But someone who knows how to use AI will.”
As companies rely increasingly on AI instead of humans, it’ll be critical to employ people who know how to get the most out of AI for their industry. Learn how to implement relevant use cases of AI in your industry (or the industry you want to enter) and you’ll be a hot commodity.
…
Heat Level: 🔥🔥
Too many AI companies do the same exact thing.
If I had a dollar for every AI customer service (CX) or AI personalization (think A/B testing on e-commerce websites) company, I’d have a lot of dollars (though still not enough to pay New York City rent).
Many AI companies have fancy websites but are essentially GPT wrappers. A GPT wrapper is a software layer in between user input and a large language model (LLM), such as OpenAI or Perplexity. It’s a translator between user and model.
Translators can be essential for getting specific, helpful answers from the LLMs because without a curated prompt, it’s unlikely that the LLM will be able to read your mind and provide the output that you’re looking for. These large language models are excellent at recognizing patterns; however, they need guided instructions to provide relevant answers.
In the example of an AI CX, a GPT wrapper pulls in details related to your brand’s customer list, style, and tone to ultimately influence the output from the LLM. The wrapper helps tailor the LLM’s responses to match the brand’s voice.
There are differences among these AI CX companies, but I have yet to hear a spiel that boasts technology head and shoulders above the rest. Most AI CX firms are GPT wrappers with little-to-no differentiation.
The winners will be determined by niche industry expertise. Truly understanding your customer and tweaking your prompts and processes to develop defensible intellectual property will be table stakes to stay alive.
I predict consolidation in AI sub-industries where the barrier to entry is low. Whoever acquires the most market share first will gobble up the slower growing platforms for the sake of their client lists. If this paragraph was a painting, its title would be “capitalism”.
…
Heat Level: 🔥🔥🔥
Just because everyone can create software doesn’t mean they want to, or that they’re skilled enough to.
Every time I peek at LinkedIn, someone is bragging about how they created their own AI agent to automate a process or even an entire industry.
AI agents are software programs that execute tasks based on pre-defined parameters and decision-making frameworks. For example, that customer service rep named Rufus that you chatted with on Amazon is an AI agent. Amazon’s data scientists have trained Rufus on large amounts of data to be able to effectively answer your questions.
In the past few years, a new crop of AI companies has popped up to give the AI agent development power to the user. Challengers like Replit and Lovable allow users to develop their own AI agents, no matter their technical skill level. It’s as easy as typing out instructions in a prompt bar, and tweaking things in a chat interface until you’re satisfied. Veteran software engineers and those who don’t know the reference “Hello World” can both build impressive digital products and processes. This beautiful process is called “vibe coding”.
When I first started vibe coding, I thought I was seeing magic before my eyes. I’ll write a future piece on some of the AI agents I’ve built and tips on how to get started.
AI agent builders level the playing field for non-technical users. However, that doesn’t mean suddenly everyone will be creating their own custom software.
Vibe coding is accessible, but being effective still requires solid prompt engineering skills, patience to iterate, and a clear vision for your product. Not everyone has the desire to put in time and energy to vibe code. You need to be relentlessly curious.
Also, vibe coding isn’t for everyone. Without a technical foundation, it’s easy to miss edge cases or create tools that lack proper security encryption. Take this example.
Say you want to create a computer game like Frogger. An AI agent builder can help you vibe code it. But if you want to monetize this game, you’ll need to integrate a payment processor and potentially store user data. That’s a problem, as AI agents are only as good as the prompts they get. If you can’t adequately look through the code and understand where the weak points are, you’re opening yourself and your users to a lot of risk.
For non-technical people, AI agent builders can be very helpful regarding automating tedious tasks like data analysis and visualization, summarizing documents, and creating computer games. But when it comes to building scalable, production-grade software, it’s unlikely to become the standard for everyone’s day-to-day needs as not everyone has the skillset or desire to build a truly reliable product.