🤖 AI to Agentic Commerce

The Big Picture

A high-level journey from neural networks to autonomous shopping

📋 Overview 🔧 Technical 🏢 Business

The Journey We'll Take

🧠 Neural Networks — How machines learn
🔮 Transformers — The attention breakthrough
📚 LLMs — Language models at scale
🤖 Agents — From chat to action
🛒 Commerce — The business transformation

🧠

Neural Networks

How machines learn from examples

The Core Idea

📊 → 🧮 → 🎯

Data in → Math → Prediction out


A neural network is a function approximator

It learns patterns by adjusting millions of numbers (weights)

Learning = Adjusting Weights

Make prediction
→
Measure error
→
Adjust weights
→
Repeat

This loop runs billions of times during training

The math that makes it work: backpropagation + gradient descent

Key Insight

"Deep" networks learn hierarchical representations —
simple patterns combine into complex concepts

Layer 1: Edges → Layer 2: Shapes → Layer 3: Objects → Layer N: Concepts

🔮

Transformers

The architecture that changed everything

The Breakthrough: Attention

Old approach (RNNs): Process words one at a time, sequentially

↓

New approach (Transformers): Look at all words at once, in parallel


"Attention Is All You Need" — Google, 2017

Self-Attention in Plain English

For each word, ask: "Which other words matter for understanding me?"


"The cat sat on the mat because it was tired"


The word "it" attends strongly to "cat" — that's what "it" refers to

Why Transformers Won

⚡

Parallel

Train 100x faster on GPUs

🔗

Long-Range

Connect distant words easily

📈

Scalable

More data + compute = better

📚

Large Language Models

Transformers trained on the internet

The Simple Training Objective

Predict the next word


"The capital of France is ___"

→ "Paris"


Do this trillions of times across all of human text

The model learns grammar, facts, reasoning, and more — all emerge from this simple objective

The Three Training Stages

1. Pre-training — Learn language from massive text corpus
2. Fine-tuning — Learn to follow instructions from examples
3. Alignment — Learn human preferences (helpful, harmless)

Scale Matters

GPT-2
1.5B params
2019
GPT-3
175B params
2020
GPT-4
~1T+ params
2023
Now
Claude, Gemini
2024-25

Emergent abilities appear at scale: reasoning, code, analysis

🤖

AI Agents

From conversation to action

LLM vs Agent

LLM

💬

Text in → Text out

Can only talk

Agent

🦾

Goal in → Actions + Results

Can do things

What Makes an Agent

🧠 LLM
Reasoning
+
🔧 Tools
Actions
+
💾 Memory
Context
+
🔄 Loop
Orchestration

The Agent Loop

Think: "I need to find milk options"

↓

Act: Call search_products("milk")

↓

Observe: Get results back

↓

Think: "User prefers organic, let me filter..."

↓

Repeat until task complete

Tools = Superpowers

🔍

Search APIs

🗄️

Databases

🌐

Web Browsing

📧

Email/Messages

🛒

E-commerce

📅

Calendars

The LLM decides which tools to use and when

MCP: The Universal Connector

Model Context Protocol

A standard way for agents to connect to any tool


Any Agent
↔
MCP
↔
Any Tool

Like USB, but for AI tools — plug and play

🛒

Agentic Commerce

The transformation of digital shopping

Shopping Today vs Tomorrow

Today

Search → Browse → Filter → Compare → Add → Checkout

15-30 minutes, lots of clicks

Tomorrow

"Get what we need for dinner this week"

2 minutes, done

What Changes

Natural Language Interface — Talk like you would to a helpful person
Personalization at Scale — Knows your preferences, dietary needs, budget
Proactive Assistance — "You usually buy milk on Tuesdays — running low?"
Multi-Step Task Completion — Handle entire workflows, not single searches

Real Example: Claw + GroceryCo

Dallas: "Plan dinners for the week, two people, ~$100"


Claw:

  • Created 7-day meal plan
  • Found sale items (NY strips 2-for-$21)
  • Added 19 items to cart
  • Total: $96.72

Using browser automation on GroceryCo.com — this actually happened

The Opportunity

Retailers who build agent-friendly systems will own the customer relationship.

Those who don't become backends for those who do.

The Complete Picture

🛒 Agentic Commerce
🤖 AI Agents (Tools + Memory + Loop)
📚 Large Language Models
🔮 Transformer Architecture
🧠 Neural Networks

Key Takeaways

  • Neural networks learn patterns from data
  • Transformers enabled parallel processing of language
  • LLMs learn to predict text, reasoning emerges
  • Agents combine LLMs with tools to take action
  • Commerce shifts from search to conversation

The Future is Already Here

In 2-3 years, customers will expect:


"Handle my groceries"


...and it just works.

🦞

Questions?


Deep dive available in technical.html

Business strategy: business.html