Intro: AI, the internet, and why you might not need all of it

Every cycle, a technology arrives that’s supposed to change everything. In the late ’90s it was the internet. Today it’s “AI.” The pattern is familiar. A handful of companies get a 10x advantage by aligning the tech with a business model the tech uniquely enables; a larger number of companies might only see a 10% lift. And a huge number of companies will just slowly see the tables stakes increase over the years and the bar simply raises as everyone uses the new technology and this is now a bare bones assumption

This is an operational reality. It's surprisingly easy for entire industries to be swept up in a moment and experience a collective fear of missing out. They key here isn't to automatically get to AI as quickly as possible. Your job is to decide whether you should invest in AI. You need to find out if it completely rewires your economics or simply improves your workflow, and to invest accordingly.

My goal in this piece is to help make decision easier. I’ll keep the useful parts of the history, correct a few popular myths, and give you a practical framework to find your Amazon moment (if you have one) and to run low-risk pilots if you don’t (yet).

I. The internet lesson (keep the good history, lose the hype)

We remember Amazon because it’s unforgettable. What we forget is survivorship bias: for every internet-native juggernaut there were thousands of businesses whose economics barely moved. The neighborhood florist that put hours on a website didn’t become a platform; they met an expectation. This is important because AI (and half a dozen advances since) followed the same arc.

The point isn’t to dismiss the web (or AI). It’s to look at what worked for who, and why:

• Amazon’s advantage was structural. National reach, searchable long-tail inventory, and logistics that scaled with traffic. The internet didn’t just make the old store cheaper, it made an entirely different store possible. The logistics of an internet-first bookstore work very differently than a brick and mortar store. They didn't move a bookstore to the internet, they created an entirely new product.

• Most businesses adopted deliberately. A website, maybe online ordering, better targeting, better analytics. All valuable. Rarely transformative. In the case of business like local restaurants, becoming an internet business might have taken 15 or 20 years to integrate with services like GrubHub, Uber Eats or Yelp, that finally made a little difference. They simply weren't impacted overnight.

Takeaway: Tech is transformative when it changes the constraints of your business model. If it literally rewires the dynamics of your current business it's transformative. If it enables entirely new modes of operation for you it's transformative. Otherwise it’s a tool. It’s important, sometimes urgent, but still a tool. Although it's entirely possible an entirely new business model might emerge that eats your lunch as a result. If that happens though it's not because you didn't "do AI", it's because you didn't change your model.

II. What actually changed in AI (and what didn’t)

“AI” is an umbrella that covers a huge family tree of technologies, some of which have been in use for years. A massive amount of business value today flows from machine learning (ML): algorithms that learn from data, drawing on applied statistics, optimization, linear algebra, probability, and systems. As of this writing (Q4 2025), a lot of businesses are still struggling to find value after sometimes massive investment in so-called generative AI and transformer/LLM derivative tech. The big winners seem to be pharmacy and logistics - the rest are wandering in the fog.

A few inflection points explain the current moment:

• 2012: AlexNet makes deep learning practical. Convolutional neural networks existed for years, but AlexNet (Krizhevsky, Sutskever, Hinton at the University of Toronto) showed that training on GPUs, changing the way neurons are emulated, adding new methods to enhance data and do more clever tuning could vastly improve image performance. This was a massive engineering and compute alignment step that unlocked the modern era.

• 2017: Transformers change language modeling. Attention Is All You Need (Vaswani et al., Google) replaced recurrence with self-attention, enabling models that turn messy text into dense vectors (embeddings) and predict likely continuations over long contexts. That architecture powers today’s LLMs (GPT, Claude, Llama, Gemini).

What this means for business boils down to few broad superpowers:

1. Turn unstructured text into structure. Businesses often have huge repositories of case notes, tickets, emails, and reviews. There may be poorly normalized data like patient notes or free hand information, or historical data such as drug testing results or literal corpus of text such as legal libraries. This type of data lends itself to the magic of embedding and reasoning (I use the term loosely). AI tools such as ChatGPT, Gemini, Claude, Llama, Deepmind and others excel at parts of this task. Toolchains to take advantage of this information often involve more than one of those tools, or even allow them to be swapped at will.

2. Retrieve and reason over your own content. Another popular emerging area is Retrieval-augmented generation (RAG). I'm going to assume you've used ChatGPT at this point. Imagine if it had direct access to all the information in your business? To all your databases and documents and could refer directly to this information while composing an answer for you. And answer based on your own policies, SOPs, and knowledge base, under human approval.

3. Agentic Uses Everyone is releasing an Agentic something-or-other today. This is generally talking about putting some tight guardrails around what these LLM models can do, and possibly chaining together multiple steps to manage simple operations, or even possible make some sort of decisions. A great rule of thumb is: if it's a decision that can be made OFTEN (hundreds, thousands, tens of thousands of times a day) then maybe you could think about automating parts of the process this way. If it's an infrequent or important step, by all means keep that human in the loop.

And a crucial caveat: LLMs are probabilistic. That’s a feature in brainstorming, summarization, triage, routing, and first-draft work. It’s a risk for tasks demanding verified ground truth, unless you attach retrieval, guardrails, and checks. They may not answer the same way twice, just like a human might not. They are also not reasoning based on experience like a human, and they're not reasoning in the same way we do. If you have a task that's a perfect fit, fantastic. If you don't - don't force it. It won't give you the benefit you're looking for. Sometimes it takes an expert to figure out the difference - don't be afraid to work with one up front.

III. The 10x vs. 10% test

Ask these four questions:

1. Where is your “long tail”?
   Do you have a huge, varied problem space, like support tickets, SKUs, case notes, or compliance checks, where patterns are invisible at human scale? That’s where ML finds leverage you’re currently leaving on the table.

2. What’s your most valuable unstructured data?
   If your best insights live in free-text (notes, transcripts, reviews), embeddings + extraction can turn them into metrics you can manage.

3. When is probabilistic “good enough”?
   Drafting copy, clustering feedback, summarizing policy changes, triaging queues: all benefit from “good-enough quickly.” Legal citations, financial postings, safety-critical steps do not, unless you add retrieval and human checks.

4. Is the upside truly 10x, or just 10%?
   Be honest: will this change acquisition cost, gross margin, throughput, or risk by an order of magnitude? You have to find this inflection point and apply the technology there. Simply adding AI to your business will not magically uncover the 10x advantage. If the advantage is only incremental, it may still be valuable, but deal with it like any other operational enhancement. It's not an existential pivot.

If you can’t articulate the 10x, treat AI as a tool in your kit. If you can, pick a thin slice and build it now.

IV. Practical starting points that won’t burn trust

A) Unstructured-data audit (2-3 weeks)

Inventory your text sources. Sample embeddings across a few months of data. Stand up a demo of semantic search, clustering, and entity extraction on your own content. Deliverables: a live demo, labeled examples, and a ranked backlog of use cases with measured signal.

B) One thin-slice pilot (6-10 weeks)

Pick a single workflow with clear metrics, for example case-note normalization to standard fields, policy Q&A with citations, or ticket triage with deflection. Define success up-front (latency, accuracy, deflection rate, hours saved). Deliver to a small user group with opt-out and logging.

C) Cost and risk guardrails from day one

Cache prompts/responses, limit model calls, cap context, and attach retrieval filters. Add human approvals for actions that move money, touch customers, or update records of truth. Measure quality against a simple internal eval set (gold answers) before rollout.

Split here