Every company is asking the wrong AI question.
What can AI do? The better question is what remains ours when AI can do almost everything.
Charlie Munger would recognise the move. In 1986 he stood in front of the graduating class at Harvard School and refused to tell them how to succeed. He gave them prescriptions for guaranteed misery instead. He took the method from the mathematician Carl Jacobi, who told his students to invert, always invert. Munger said it plainly years later. If the World Bank hires you to help India, first find the three best ways to increase misery in India. Then avoid them.
The AI industry has spent three years thinking forward. Better models. Cheaper models. Faster models. Nobody has inverted the question. So I will. I call the answer unclaudability. Claude here is shorthand for a generation of frontier models that can write, code, reason and increasingly act. It has nothing to do with one company. To be unclaudable is to hold an advantage that does not vanish the morning the model gets better.
Every business, and every person inside it, has to become unclaudable.
Intelligence is becoming infrastructure
Capabilities that looked extraordinary eighteen months ago are standard features today. The coding assistant that saved an afternoon now saves a sprint. Reasoning that needed a research lab runs behind an API with a credit card on file. This will continue. The models will get much better.
Electricity was once an industrial advantage. Cloud computing was once a boardroom strategy. Both became a monthly bill. AI is on the same path, only faster.
The better the model becomes, the weaker access to it becomes as an advantage. Your competitor has the same model, through the same API, at the same price, on the same day. If everyone can buy the same intelligence, intelligence cannot be the moat.
This is uncomfortable for a lot of AI companies. If the product is public information arranged more conveniently, a prompt wrapped in an interface, or a workflow a competent team could rebuild in a few weeks, the company does not own a moat. It rents a feature from someone else’s roadmap.
Context is not data
The standard answer to this is data. That answer is half right, and half right loses.
A million unlabelled records in a warehouse are storage. Data becomes valuable when it sits inside a learning loop. The real advantage is the context created by doing the work. Why did the customer reject the output? Why did the expert disagree? Which edge case broke the workflow? What got corrected? What got escalated? What happened after the decision?
None of that can be downloaded. It accumulates through thousands of decisions made under real constraints with real consequences. A model can reason about what was captured. It cannot recover what was never captured. A competitor cannot recreate years of operating history overnight. The failed experiments, the exceptions, the corrections and the trust are not by-products of the business. In many businesses they are the business.
Data without a learning loop is an archive. Data inside a learning loop compounds.
There is a difference between reasoning about an event and having been there. A model can explain a balance sheet better than most CFOs. It cannot tell you why row 14 looks the way it does, because it was not in the room at 11pm when that call got made. It can write a five-year strategy in ninety seconds. It never sat through the version that failed in 2021, the one nobody mentions and every decision still routes around.
Then there is judgement. Two people can read the same dashboard with the same model and decide differently. A model can lay out every option. It cannot want one more than the others. It has no skin in the outcome.
Healthcare makes this obvious
The gap between a model and a working system is enormous.
A diagnostic model can be trained by anyone with enough compute, data and talent. What is hard to recreate is years of deployment across real hospitals. Clinical disagreements. Rare edge cases. Failure modes. Escalations. Quality checks. The trust a doctor needs before using the system when the answer matters.
At 5C Network, the visible layer is AI-assisted radiology. We process around 20,000 scans a day across 2,000+ facilities with 400+ radiologists. Every time a radiologist agrees, disagrees, corrects or escalates, the system learns something it did not know yesterday. We have done this across more than 20 million studies.
The model is not the moat. The learning system around it is. A model deployed without a learning loop gets older every day. A learning system used every day gets stronger.
This is also why accountability matters. AI can recommend. Increasingly it can act. A hospital, a radiologist and a healthcare company still have to own the outcome. Somebody’s name is on the report. Somebody explains it to the family. In high-stakes work, accountability is part of the product. No future model removes it.
People need to become unclaudable too
A person gets clauded when they stop using AI to improve their judgement and start using it instead of judgement. The model produces an answer. The answer sounds good. It arrives fast. After enough of them the person can no longer tell whether it is right.
An unclaudable organisation needs unclaudable people. People who use AI aggressively and still question, decide and own the consequence. AI should raise the rate at which judgement improves. It should not take judgement out of the loop.
You have to keep becoming unclaudable
None of this is permanent. Every advantage gets copied. Every workflow gets understood. Every moat gets a bridge built across it.
Google’s advantage was never search alone. It was context accumulated across Search, Maps, Android, YouTube and Photos, each product sharpening the others. Google became a verb. The moat is still not permanent. Every year a new contender attacks it. One year the contender will be right.
Nobody becomes unclaudable once. It is a discipline. Build an advantage. Spend the runway it buys. Accumulate context while you lead. Start on the next advantage before the model catches up on this one.
Asking what our moat is today gets you nowhere. Ask this instead. If the best model becomes ten times better tomorrow, what will still be ours? What gets commoditised? What becomes more valuable every time we operate? What are we learning that a competitor cannot download?
A moat is not a noun. It is a verb.
Use AI everywhere. Let it make the organisation faster, sharper and more capable. Do not confuse borrowed intelligence with owned advantage.
Intelligence will become abundant. Experience will not. The winners will be the businesses that convert execution into context, corrections into learning and outcomes into trust.
That is how you become unclaudable.
This article is authored by Kalyan Sivasailam, Founder & CEO, 5C Network.
Disclaimer: The above content is non-editorial, and TIL hereby disclaims any and all warranties, expressed or implied, relating to it, and does not guarantee, vouch for or necessarily endorse any of the content.
