Turing, Prompting, and the AI Company Blueprint
In March 2016, something unprecedented happened. AlphaGo, an AI developed by DeepMind, defeated Lee Sedol—one of the greatest Go players in history—4-1 in a five-game match.
Go is not chess. It has more possible positions than atoms in the universe. For decades, experts believed mastering Go would require human-like intuition, creativity, and strategic depth.
They were wrong.
In Game 2, AlphaGo played "Move 37"—a move so unconventional that expert commentators initially dismissed it as a mistake.
It wasn't. Move 37 ultimately won the game. More importantly, it revealed that AI could do something experts thought impossible: discover strategies that humans had never considered.
This wasn't just pattern matching. This was emergence—a system generating novel solutions that surprised even its creators.
Alan Turing asked a deceptively simple question: "Can machines think?" He proposed the Imitation Game—later called the Turing Test—as a way to sidestep philosophical debates and focus on behavior.
For 70 years, this remained largely theoretical. Then came Large Language Models.
The "Attention Is All You Need" paper introduced the transformer architecture. GPT, Claude, and their successors followed. Suddenly, machines could:
Working with LLMs is not programming in the traditional sense. It's more like directing—guiding an intelligent system toward the outcome you want.
Key prompting principles:
Most tasks you give an LLM fall into one of four categories. Understanding these helps you craft better prompts:
Distilling large volumes of text into key points. The power lies in focused summarisation—extracting specific angles rather than just "less text."
Example: "Summarise this complaint with a focus on shipping issues only."
Reading between the lines. Sentiment analysis, entity extraction, topic modelling—tasks that used to require specialized models can now be handled via simple prompts.
Example: "Is this review positive or negative? Extract the product name."
Changing form without changing meaning. Translation, tone conversion, format conversion—the conveyor belt of information processing.
Example: "Convert this paragraph into a clean JSON object."
Generating longer text from shorter inputs. The model as creative partner. Be careful: this is where "hallucinations" can creep in if you don't constrain the model to provided facts.
Example: "Draft a polite email explaining the delay and offering 10% discount."
We're now entering the era of AI agents—systems that don't just respond to queries, but take actions, use tools, and pursue goals autonomously.
This creates new possibilities:
The Model Context Protocol (MCP) is one framework for building these agentic systems.
Andrew Ng's vision of the "Truly AI Company" involves four pillars:
This isn't just about adding AI tools. It's about restructuring your organization around the flow of data and intelligence.
The goal is not replacement. It's augmentation.
AI excels at: pattern recognition, information retrieval, consistency, scale, speed.
Humans excel at: judgment under uncertainty, ethical reasoning, creativity, empathy, wisdom.
The future belongs to those who can combine both—leveraging machine capabilities while cultivating distinctly human skills.
Practical framework for building AI assistants with MCP
When to automate vs. when to explore
On correlation, causality, and dogma