In 2019, Rich Sutton, one of the most influential researchers in artificial intelligence, published an essay that rattled the field. He called it "The Bitter Lesson."
His thesis was simple: "The biggest lesson that can be read from 70 years of AI research is that general methods that leverage computation are ultimately the most effective, and by a large margin."
The lesson is "bitter" because it means human ingenuity, all those clever tricks, domain-specific heuristics, and hand-crafted rules we're so proud of, gets crushed, eventually, by systems that just throw more data and compute at the problem. Deep Blue beat Kasparov through brute force search, not chess wisdom. AlphaGo demolished human champions by playing itself millions of times, not by studying centuries of human strategic insight.
This suggests a corollary. Call it the bitter lesson of personal AI:
No matter how advanced the models become, the assistant with better personal context will always win.
The ImageNet Moment
In 2006, while her colleagues obsessed over better algorithms, Fei-Fei Li noticed something everyone else had missed. The problem wasn't the models. The problem was the data.
"Pre-ImageNet, people did not believe in data," Li later recalled.
She proposed something that seemed absurd: a database of 15 million labelled images across 22,000 categories. "I think you've taken this idea way too far," a mentor warned her.
But Li was right. When AlexNet, trained on ImageNet's massive dataset, crushed the 2012 competition by an unprecedented margin, deep learning exploded overnight. The models hadn't changed that much. What changed was the data.
The algorithm people had been dismissing the data people for years. Then the data people won, decisively and permanently.
The $400 Million Parallel
The previous article in this series described Drake's personal assistant Courtne Smith as a "$400 million context engine." She grew up with him in Toronto. She's seen him interact with the real world for over a decade, his emails, schedules, health routines, dining preferences, relationships, financial decisions.
That's not a few hundred data points. That's millions of interactions, compressed into a single person who can anticipate what he needs before he asks.
Smith didn't attend a special school for personal assistants. What she has is context, deep, rich, longitudinal context accumulated over years of proximity to one person's life.
This is the personal AI equivalent of 15 million images in ImageNet. The breakthrough isn't in how smart the assistant is (although Courtne is also incredibly smart). The breakthrough is in how much they know about you.
The Context Chasm
Today's AI assistants are phenomenally capable. Claude, ChatGPT, Gemini, they can write code, analyze documents, plan trips, draft emails. They've been trained on trillions of tokens representing the sum of human knowledge.
But ask them to plan your vacation, and they'll suggest Barcelona. Not because Barcelona is right for you, they have no idea what's right for you.
Your AI assistant knows everything about everything except the one thing that matters most: YOU.
Courtne Smith doesn't just know Drake's Aeroplan number. She knows which airline (or private charter service) he prefers, which routes he likes, how he behaves when he's exhausted versus energized, which hotels treat him right, what he orders when he's stressed versus celebrating.
That's not a profile. That's a model, an internal simulation of another person built from years of observation.
What Comes Next
The bitter lesson of personal AI is this: context will win. The assistant with more context will always outperform the assistant with less context, regardless of which underlying model is more powerful.
So if you want an AI assistant that rivals what Courtne Smith provides Drake, you need to give it the same advantage she has: years of accumulated context about your life.
You need a tool to manage your digital memory. Your context is scattered across hundreds of platforms; Netflix, Spotify, Amazon, airlines, hotels, restaurants, banks. You need a way to aggregate it, own it, and control which AI assistants can access which parts of it.
That's why we built RaLHF: a personal context library that you own completely, that works with any AI assistant, and that you can grant or revoke permissions to at any time. No more starting from scratch. No more repeating yourself. Your memory, your rules.
The data people always win. Make sure you're one of them.
This is Part 2 of our series: "The Personal Assistant Revolution: How AI Will Make Everyone Successful." (Read Part 1 here)