I added NEES Core Engine to Indie Hackers today

Over the last few weeks I’ve been discussing a pattern I kept seeing across AI systems:

Most production AI failures are not actually model failures.

They are:

• workflow failures

• governance failures

• escalation failures

• observability failures

• operational ambiguity failures

The AI often did exactly what the system implicitly allowed it to do.

The problem was that the workflow itself was never fully defined.

After a lot of discussions with builders working on:

• customer support agents

• DeFi tooling

• WordPress AI systems

• ETL/data pipelines

• voice AI

• multi-agent systems

• workflow automation

I decided to formally add NEES Core Engine as a product on Indie Hackers.

What is NEES Core Engine?

NEES Core Engine is a governed AI runtime layer for production AI applications.

Instead of:

User → App → LLM → Response

the flow becomes:

User → App → NEES Core Engine → Model Provider → Governed Response

The idea is not to replace the model.

The idea is to add operational structure around the model:

• traceability

• memory boundaries

• escalation logic

• runtime governance

• permission boundaries

• observability

• workflow control

• auditability

Because once AI systems move into real operational environments, the difficult problems become less about:
“can the model generate text?”

and more about:

• what was it allowed to do?

• why did it make this decision?

• what workflow state existed?

• when should it escalate?

• what assumptions influenced the response?

• how do we inspect behavior later?

One realization changed my thinking

A comment from a builder in the discussions summarized it perfectly:

“The AI did not create the ambiguity. It exposed it.”

That line stuck with me.

Because the more I looked at production AI failures, the more it felt like organizations were discovering undocumented operational assumptions for the first time.

Humans silently compensate for:

• exceptions

• tacit heuristics

• hidden business rules

• unclear ownership

• escalation behavior

• incomplete workflows

AI systems force those assumptions into the open.

Developer Preview

I also opened a public developer preview repo:

https://github.com/NEES-Anna/nees-core-developer-preview

And there’s a live sample app connected to the governed runtime:

https://naina.nees.cloud

Still early.
Still learning.
Still refining the architecture.

But the conversations around workflow governance, operational trust, and AI observability have been some of the most valuable discussions I’ve had so far.

Curious if others building production AI systems are seeing the same thing:

When your AI system fails…

does the problem usually start with the model?

Or with the workflow around it?