Darron Dickinson

I build systems that turn fragmented organizational knowledge into intelligence, better decisions, and action. Co-Founder, NeuralNomad + Sovrain Systems.

[Darron Dickinson @DarronDickinson]

Everyone is racing to build bigger models.
We’re exploring a different question. What if intelligence is an emergent property of network structure rather than model size? Today we released the first public version of IONS.
CBB = Knowledge Atom
Relationship = Knowledge Bond
Reasoning Path = Knowledge Molecule
Intelligence emerges from composition.

GitHub is live. Now the real testing begins. https://t.co/f1NnHZgUBh

#AI #OpenSource #KnowledgeNetworks #Reasoning #SystemsThinking

[Darron Dickinson @DarronDickinson]

Frontier models compress knowledge into parameters.

IONS asks a different question:

What if knowledge remained explicit…

relationships remained inspectable…

and reasoning became traversable?

[Darron Dickinson @DarronDickinson]

A workflow graph answers: “What happens next?”

A knowledge graph answers: “What do we know?”

A reasoning graph answers: “Why did we reach this conclusion?”

They’re related, but they’re not the same problem.

[Darron Dickinson @DarronDickinson]

Graph engineering is taking off, but I think we’re conflating two very different problems.

Most agent graphs describe how work executes.

I’m more interested in graphs that describe how knowledge composes into reasoning.

[George Sivulka @gsivulka]

https://t.co/aLTltYpmcj

[Darron Dickinson @DarronDickinson]

@Joi2James Distributed intelligence: https://t.co/6oOSUJuuZq

[Darron Dickinson @DarronDickinson]

Satya Nadella just told the WSJ:

"If all the value is accrued by only a few models, the political economy will simply not tolerate it."

He called for every organization to own a 'learning loop' using their own data — not depend on frontier model APIs.

We built exactly that.

IONS is an open protocol where your knowledge lives in a traversable network of typed claims — not locked inside someone else's model weights.

Any lightweight model can traverse it. You own the network.

https://t.co/f1NnHZgUBh

#AI #OpenSource #IONS

[Darron Dickinson @DarronDickinson]

Posted on Hacker News today — if you find the IONS protocol interesting, an upvote helps with visibility: https://t.co/NsXCKuPiYn

[Darron Dickinson @DarronDickinson]

What if intelligence lived in a network, not a model? An 8B model connected to a traversable knowledge graph matched frontier AI on domain-specific queries.
The model was the interpreter. The network was the intelligence.
Open source. Run your own node. 👇
https://t.co/VSi9kFWeGl

[Darron Dickinson @DarronDickinson]

We spent decades building systems of record.

We are now entering the era of systems of understanding.

The companies that win will not just store information.

They will understand how decisions, people, processes, and outcomes connect.

[Darron Dickinson @DarronDickinson]

Personalization is segmentation with better marketing. Understanding is knowing enough about an individual to predict needs before they ask.

[Darron Dickinson @DarronDickinson]

Most organizations do not have an AI problem.

They have a memory problem.

Teams forget.
Knowledge walks out the door.
Decisions lose context.

AI without organizational memory just creates faster confusion.

[Darron Dickinson @DarronDickinson]

Multi-agent AI systems don't fail because individual agents are wrong. They fail because individually correct agents produce collectively incoherent outcomes. The problem is architectural. No agent sees the whole system.

[Darron Dickinson @DarronDickinson]

How multi-agent systems fail in financial services:

1. Pricing agent optimizes for margin
2. Retention agent discounts to hold accounts
3. Credit agent approves based on risk model
4. All three "correct" decisions produce a member with a discounted high-risk loan

No fraud. No model error. Just incoherence.

[Darron Dickinson @DarronDickinson]

Hot take: most "AI governance" solutions solve the wrong problem.

They audit what happened. They log decisions. They flag exceptions.

What they do not do is prevent multiple agents from making locally rational, globally incoherent decisions simultaneously. That requires intent governance at the architectural level.

[Darron Dickinson @DarronDickinson]

Your pricing agent raises rates. Your retention agent discounts. Your credit agent approves. No one is wrong. The system is incoherent.

The cost of local optimization in AI systems is not bad decisions. It is contradictory decisions that quietly compound.

#aicoherence

[Darron Dickinson @DarronDickinson]

Building AI for finance isn’t just about automation. It’s about creating systems that make collective sense. That’s where true value lies and it’s hard.

[Darron Dickinson @DarronDickinson]

Why do most AI projects falter at scale? Because they optimize locally and ignore the bigger picture. Coherence isn't a feature, it’s a system design. Are yours aligned?

[Darron Dickinson @DarronDickinson]

AI agents are great at local decisions, but the real challenge is making those decisions work together coherently. Without that, you risk systemic failure.

[Darron Dickinson @DarronDickinson]

We’ve released two open standards to address multi-agent gaps:
- Standard Intent Specification (SIS):  A structured way to define a decision before execution.  
- Agent Decision System Specification (ADSS): A standard for how decisions are processed, evaluated, and executed across systems. 
https://t.co/UmnV3xazTA