The Coming AI Data Crisis - and Why Datom.world Exists

Published Nov 17, 2025

Andrew Ng recently highlighted a problem that every serious AI practitioner is running into: AI agents are getting smarter faster than our data infrastructure can keep up.

As agents become capable of correlating signals across email logs, purchase histories, sensors, CRM events, contracts, PDFs, spreadsheets, and unstructured knowledge bases, the value of cross-system interpretation grows. But the cost of accessing that data - because of SaaS vendors - is rising even faster.

Some vendors now charge tens of thousands of dollars just for an API key to your own data.

This Isn't an Accident

It's an economic strategy:

Create a silo
Make it painful to leave
Charge for interoperability
Sell "AI add-ons" to replace your own agents

The problem is structural, not accidental. The architecture of SaaS inherently encourages lock-in.

This is exactly the problem Datom.world was designed to remove from computing altogether.

The Architectural Mismatch: Agents vs. Silos

AI agents don't want "products."

They want streams of structured and unstructured data flowing through them so they can correlate patterns and take action.

But today's SaaS world is built on the opposite assumption: each service is a container for your data.

Agents thrive in open systems.

SaaS thrives on closed systems.

Every API paywall, export fee, rate limit, or opaque schema is friction that prevents agents from doing what they're good at - building connections you didn't know to look for.

Andrew Ng's core argument is simple:

The economic value of "connecting the dots" has exploded.
But your data is trapped in boxes you don't control.

This is not a technology problem. It's an architecture problem.

Why Datom.world Takes a Different Path

Datom.world starts from the opposite assumption:

1. You Own Your Data. Absolutely. Locally. Physically.

SaaS vendors should be hired to operate on your data - not own it.

In Datom.world, data is not stored on someone else's servers behind an API. It lives as local, append-only streams of datoms: [e a v t m].

No proprietary schemas
No centralized database
No API lock-in

2. Software is an Interpreter, Not a Silo

Datom.world treats every application as an interpreter of streams. There is no "database product," no "CRM product," no "analytics product."

There are only:

Streams (your data)
Interpretations (apps or agents)

Apps don't own data. Apps observe it.

This is the same design insight that allowed the web to scale: HTML pages were yours, and browsers merely interpreted them.

Datom.world generalizes this to all business data, not just hypertext.

3. Agents Plug Into the Same Universal Substrate

AI agents can subscribe to streams, interpret them, produce new datoms, and migrate between nodes.

No integrations
No glue code
No API costs

Agents become first-class peers in the computation ecosystem, not second-class citizens.

A Single, Universal Data Substrate

SaaS treats each data domain as a separate world:

A marketing world
A sales world
An operations world
A finance world

Each with its own schema, dashboard, locked UI, and monetization model.

Datom.world collapses all of this into a single substrate - a tuple-space where every event is just a datom in a global stream.

This removes the very concept of a silo:

Email click event       → datom
Purchase event         → datom
Calendar event         → datom
PDF extraction         → datoms
IoT sensor             → datoms
LLM output             → datoms

Once everything reduces to datoms, interoperability becomes a natural property of the system.

No API needed
No schema negotiation
No vendor permission

You simply read the stream.

The Economic Consequence: A Siloless Future

Vendors can no longer charge you to access your data if they never owned it.

Apps compete on interpretation quality, not data captivity.

This flips the incentives of the software industry:

Instead of SaaS selling "AI features,"
They sell interpretations that run on your substrate.

This is the same shift that happened with Linux, the web, and containerization: ownership moves downward, and ecosystems flourish upward.

In the era of AI agents, this shift is inevitable.

The Obsidian Example: The Future of Software

Andrew Ng uses Obsidian as a personal example:

Notes are stored locally as plain Markdown
Apps are hired, not obeyed
Agents can freely read/write the same files

This is exactly the architectural pattern Datom.world generalizes.

Obsidian is one app proving the point. Datom.world is the ecosystem built around that principle.

AI Agents Aren't the Future Without Data Freedom

AI agents will reshape nearly every business process. But the real constraint is not their intelligence - it is the accessibility of the data they need.

SaaS vendors will not voluntarily open their silos. The architecture itself must change.

Datom.world exists to make that change possible.

It replaces:

Centralized databases
Proprietary APIs
Domain-specific SaaS silos

with:

A universal, local-first datom substrate
Agent-ready tuple streams
Free interoperability by design

This is not a new product category.

It is a new substrate for computing - one designed for a world where agents, not vendors, orchestrate workflows.

The AI Future Belongs to Whoever Controls Their Own Data

Datom.world makes that possible.

When AI agents need to correlate your email, purchase history, contracts, and sensor data, they shouldn't have to navigate a maze of paywalls and proprietary APIs.

They should just read the stream.

Because it's your data. Always has been. Always should be.

Learn More

DaoDB: The Datalog database that stores everything as datoms
Datoms as Streams: Local-first data architecture
Yin.vm: Universal Semantic AST: How code itself becomes datoms
Structure vs Interpretation: Why ownership matters
Datom.world: The future of computing

Referenced: Andrew Ng's tweet on AI data accessibility (April 2025)