A few days ago, I wrote an article about the paradigm shift from SaaS to Agents, discussing how AI Agents are reshaping the value of the software stack.
After finishing that article, one question stayed with me:
That article was mainly about enterprises and industries. But for individuals, or for most small and medium-sized teams, there is a much more direct question:
Do you really own the AI capabilities you use every day?
Most of the AI You Use Today Is “Rented Intelligence”
The barrier to using AI today is extremely low.
Subscribe to ChatGPT, enable Copilot, or turn on the AI features inside Notion or Slack, and within minutes you can have a fairly complete AI experience.
That is a good thing. It means almost everyone can start from roughly the same place.
But if you look one layer deeper, there is a problem:
What you actually own is only the right to use the service.
Your prompts, your context, the documents you upload, and the workflows you gradually build all live inside someone else’s product.
As long as you keep paying, you can keep using them. But when the platform changes its policies, adjusts pricing, switches models, or removes a feature, you generally have little choice but to accept it.
More importantly, after using a platform for two years, the value you have accumulated may still not belong to you.
The day you switch tools, much of that accumulation can disappear.
This is similar to the Evernote example I mentioned in my previous article.
The problem is not that the product is bad. The problem is that when your data and workflows are locked inside a single application, switching costs quietly become a liability.
And this is not merely a hypothetical risk.
In my previous article, I mentioned Airtable. The market can completely rethink the answer to the question, “What is this kind of software really worth?” within just a few years.
When platforms themselves can be revalued, acquired, or redirected, placing all of your knowledge and workflows inside someone else’s product creates a form of concentration risk.
General Intelligence Is Powerful, but It Is Not Your Intelligence
Today’s frontier models are extremely capable.
Anyone who subscribes can access roughly the same level of general-purpose intelligence.
But the real differentiation has never been the model itself.
It is the context:
your data, your preferences, your business or creative processes, and the decisions you have made over time.
Rented intelligence is essentially a general-purpose model combined with very little context.
Each conversation largely starts over. Even with memory features, only a limited amount of information is retained.
The system does not truly understand who you are or what you are working on.
Intelligence that you build yourself is fundamentally different.
It is built on top of your own data and workflows. The more you use it, the more useful it becomes.
The value accumulates on your side, rather than on the platform’s side.
In one sentence:
General-purpose models are becoming a commodity. Anyone can rent them. Your context and your data are the real moat.
What Does “Own Your AI Stack” Mean?
In my previous article, I suggested that the next generation of the software stack may gradually look like this:
Compute / Edge / Private Cloud → Data → API / Tools → Agent → Human
Apply the same structure to individuals and small teams, and you get what I mean by having your own AI stack:
- Data layer: Keep your notes, documents, and content in open formats such as Markdown, Git, local files, or your own database, rather than locking them inside closed applications.
- Knowledge layer: Allow Agents to index and understand your data through a local knowledge base, RAG, and long-term memory.
- Tool layer: Use open protocols such as MCP and APIs to connect your data and workflows to Agents.
- Agent layer: Use coding Agents such as Codex and Claude Code, together with workflow orchestration tools such as n8n.
- Model layer: This is actually the layer you need to own the least. Frontier model APIs can be rented, while open-weight models can be self-hosted. The key is not which provider you use, but whether you can replace it whenever necessary.
The core principle is simple:
Own the data and orchestration layers. Rent the model layer.
Models improve extremely quickly.
The capability you rent today may be surpassed six months from now, so there is little reason to lock yourself permanently into one model provider.
Your data, context, and workflows are different.
They are assets that compound over time.
Those are the things you should keep under your own control.
Why Now?
In the past, building your own tools required a relatively high level of technical expertise.
Three things have changed.
First, protocols such as MCP are beginning to standardize how tools, data, and services connect to Agents.
You no longer need to build every integration from scratch, and integration costs are falling quickly.
Second, coding Agents have dramatically lowered the barrier to building software.
If you can clearly describe what you need, an Agent can help you write scripts, build workflows, and connect APIs.
Many things that previously required an engineering team can now be built by one person working with an Agent.
Third, open-weight models and local inference are becoming increasingly mature.
You now have real alternatives instead of being completely dependent on a single vendor.
Together, these three developments have reduced the cost of ownership to a level that individuals and small teams can realistically afford.
Your Own Intelligence Is a Compounding Asset
I like to think about this in terms of expenses versus assets.
Rented intelligence is fundamentally an expense.
You pay every month for access to the same general-purpose capability. Stop paying, and the capability stops.
The accumulation does not necessarily belong to you.
Intelligence that you build yourself is different.
It becomes an asset.
Every note you organize, every repetitive workflow you automate, and every skill or memory you add to your Agent makes the stack better suited to you.
That is the fundamental difference between using AI and building your own intelligence.
The same logic applies to enterprises.
In my previous article, I discussed Taiwan Mobile and Systex, as well as SAP’s investment in n8n.
What I really wanted to emphasize was this:
In the future, the important question for enterprises will not be how many more SaaS seats they can buy.
It will be whether their ERP, CRM, and databases can be safely orchestrated by Agents — and whether that orchestration layer is built on top of their own data and governance.
If a company does not own that layer, then over time, part of its margin will simply become someone else’s rent.
How to Start: Four Small Steps
When people hear “build your own stack,” their first reaction is often:
That sounds like a major engineering project.
It does not have to be.
And you certainly do not need to self-host everything from day one.
I would suggest starting with four small steps:
- Get your data out of closed applications.
Export it, back it up, and convert it into open formats such as Markdown, CSV, or Git. Data portability is the first step toward ownership. - Choose one high-frequency, repetitive workflow and automate it with an Agent and MCP.
Do not aim for perfection. Just get one workflow running end to end. - Build your own knowledge base and memory.
Allow your Agent to reference what you have accumulated in the past instead of starting from zero every time. - Preserve replaceability.
Models and tools should always be replaceable. Do not lock your data into any single vendor’s user interface.
My own approach is to keep notes and content in Markdown and Git, allow coding Agents to directly read and write files and execute workflows, and connect frequently used services through MCP.
Models can be replaced at any time.
The data and workflows remain mine.
Ownership is about interfaces and data, not about owning every component.
Conclusion: Keep the Accumulation on Your Side
In my previous article, I argued that the real paradigm shift is that value is moving away from UI and SaaS seats toward:
Data, Compute, Identity, Workflow, APIs, Agent Runtime, and Orchestration.
The same logic applies to individuals.
In the AI era, the most valuable thing is not how many services you subscribe to. It is whether you have a stack that you control.
Own your AI stack. Every piece of value you accumulate can compound into your own intelligence.
Otherwise, you are simply renting someone else’s intelligence while leaving your accumulation on someone else’s platform.
You do not need to do everything at once.
Start by exporting your first note into Markdown, or by handing one workflow to an Agent.
At that point, you have already started owning your stack.
I believe this is one of the most worthwhile things for everyone to start doing today.
If you are also interested in the idea of owning your AI stack, feel free to reach out. I would also be very interested to hear how others are building their own AI stacks today.