You Could Have Steve Jobs and Taiichi Ohno on Your Board and Instead You've Got Them Answering WhatsApps: AI and RAG for Leadership Decisions
Most people use AI for tasks a human does better. Meanwhile, they leave the one seat where it would truly matter empty: the leadership seat.
You're Using a Brain to Make Photocopies
Most companies have brought AI in through the lowest possible door.
Answering emails. Summarizing documents. Reviewing contracts. Replying to WhatsApps.
Low-value tasks. And I'll say it plainly: answering a WhatsApp has no merit whatsoever. On top of that, for now, people do it better.
I'm Cinto Casals, AI engineer and architect of agentic systems. And I've been watching for a while how the interesting part of this technology gets wasted on the most boring stuff.
What AI Lacks Isn't Intelligence
Today's AI can already reason about leadership decisions. Truly.
What it lacks isn't capability. It's knowing your company.
A generic AI reasons about a textbook company. Not yours. It doesn't know your numbers, your processes, or why you made that decision two years ago.
You just have to explain it properly. That's the whole game.
My Case: Two Advisors Who Argue With Each Other
I have two advisors. And they argue with each other to help me decide.
One is a Twin of Steve Jobs. The other, a Twin of Taiichi Ohno, the father of the Toyota Production System.
When I have to make a business decision, they give me a consensus answer. And they explain how they got there.
- Jobs represents product, vision, and simplicity.
- Ohno represents the elimination of waste and continuous improvement.
The magic isn't in either of them separately. It's in the friction between the two.
Jobs wants to add. Ohno wants to remove. Jobs talks about desire. Ohno talks about flow. And out of that clash comes something no single advisor would ever give me.
This delivers far more value than automating replies. And it's something that, without AI, I couldn't afford.
The Bridge Is Called RAG
Between a generic AI and an advisor that knows your company there's a bridge. It's called RAG (Retrieval-Augmented Generation).
It works like this: before answering, the model consults your company's real knowledge.
Numbers. Processes. Contracts. Meeting minutes. Decision history.
And it reasons about YOUR company. Not a textbook company.
RAG, Not Fine-Tuning
A lot of people wonder why not just retrain the model with the company's data. Short answer: it's a financial trap.
A company's knowledge changes every week. Retraining a model every time something changes is expensive and slow.
With RAG, you update the documents and the advisor already knows. And on top of that, you can see which document each argument comes from.
A Poorly Built RAG Is Worse Than None at All
Here comes the part almost nobody tells you.
A poorly built RAG is an advisor that hallucinates using your own data. And it does so with tremendous confidence. That's more dangerous than having nothing.
These are the keys to a serious implementation:
- Curate the sources. Don't dump the entire Drive. Only what Leadership needs: strategy, accounts, KPIs, processes, key contracts, minutes.
- Clean data, dated and free of duplicate or obsolete versions.
- Chunk the documents sensibly and tag them with metadata: area, date, reliability.
- Permissions. What the Leadership advisor sees is not what the customer service advisor sees.
- Sovereignty. The business knowledge on your own or controlled infrastructure. Not scattered across personal accounts (that's Shadow AI).
- Traceability. Every answer cites its sources. And if there's no source, it says so.
- Continuous evaluation. Control questions with known answers to detect when the system degrades.
Debate as a Method
Two Twins with the same RAG, but with different criteria, contradict each other. And in contradicting each other, they're forced to justify themselves.
What reaches Leadership isn't an opinion. It's a line of reasoning that can be audited.
And let's be clear: the decision still belongs to whoever leads. The Twins don't decide. They set the stage to decide better.
If you want to see it working, I have published it in the open. Focus & Sian is a free plugin for Claude: two AI advisors, one using Steve Jobs's method and the other Taiichi Ohno's, debate your project until they agree on a joint strategy. The minutes record what each one conceded and why. Code and instructions on GitHub: github.com/cintocasals/focus-sian
The Metaphor That Sums It All Up
Using AI to answer WhatsApps is like having Steve Jobs and Taiichi Ohno on your payroll and setting them to make photocopies.
And giving Leadership the best AI without RAG is seating them on the board without showing them the accounts.
In both cases, you have a brutal brain and you leave it blind or with its arms crossed.
So, What's Your Move?
The question isn't which tasks AI can automate.
The question is who you seat at your leadership table. And what you've told them about your company.
Because you can have the best advisors in the world. If you don't show them the accounts, they're useless.
In the AI-Native Company Manual I dedicate steps 2 through 5 to building exactly that brain with RAG: https://www.cintocasals.com/#manual-ia-nativa
Why use AI for leadership decisions and not just for tasks like answering WhatsApps?
Because answering a WhatsApp is a low-value task and, for now, people do it better. Today's AI can already reason about leadership decisions. What it lacks isn't intelligence, but knowing your company. Applying it at the top, at the leadership table, delivers far more value than automating low-level replies.
What is RAG and why is it key for AI to advise on my company?
RAG (Retrieval-Augmented Generation) is the bridge between a generic AI and an advisor that knows your company. Before answering, the model consults the real knowledge of the business (numbers, processes, contracts, minutes, decision history) and reasons about YOUR company, not a textbook company. It also lets you see which document each argument comes from.
Why RAG and not retraining the model (fine-tuning)?
Because a company's knowledge changes every week and retraining a model each time is a financial trap. With RAG you just update the documents and the advisor already knows, with the added advantage of traceability: every answer cites its sources.
What's the risk of a poorly implemented RAG?
A poorly built RAG is worse than having none: it's an advisor that hallucinates using your own data, and does so with great confidence. To avoid this you must curate the sources, keep data clean and dated, chunk and tag with metadata, manage permissions, ensure knowledge sovereignty, guarantee traceability, and run continuous evaluation with control questions.
Comentarios