Article
v1.0
The operating state behind continuous AI decisions
Decision models can make high-volume choices cheaply, but they still need a live system that maintains context, governs action and learns from outcomes.
There’s a lot of excitement about JEV at the moment.
Our R&D team had a play. Here’s what we think it means for AI-native GTM 👇
Most AI workflows today look something like:
Context → LLM → Answer → Action
JEV points to something different.
Imagine you have 100,000 accounts in your market. Your GTM operation is constantly making small decisions:
→ Is this actually ICP?
→ Is there a buying window?
→ Does this signal matter?
→ Should we engage now?
→ What should happen next?
Today, we increasingly throw these questions at LLMs.
But many aren’t language problems.
They’re decision problems.
JEV takes State and a defined set of outcomes and returns decisions with probabilities.
State → Decision → Probability.
That matters because if models become incredibly cheap and fast at making millions of these decisions, the bottleneck moves.
The harder problem becomes maintaining the context they decide against:
What is true right now?
What changed?
What does our company know?
What action should follow?
When should a human intervene?
What did we learn from the outcome?
This is why we’ve become obsessed with State at Revenue Labs.
CRM history. Calls. Emails. People. Engagement. External signals. Product behaviour. Company knowledge.
Continuously assembled into a representation of commercial reality.
The architecture becomes less:
Data → Prompt → LLM → Output
And more:
Evidence → State → Decision → Action → Outcome → Learning
Different intelligence can sit inside that loop.
Frontier models for complex reasoning. Specialist models like JEV for high-volume decisions. Deterministic systems where appropriate. Humans where consequence or uncertainty requires judgement.
The model becomes a component of the operation, rather than the operation itself.
That’s the shift from AI-assisted to AI-native GTM.
JEV is early. But what it represents is more interesting than any individual model.
LLMs gave software the ability to generate.
Now we’re giving software the ability to continuously decide.
The missing layer is the system those decisions operate against.
That’s what we’re building at Revenue Labs.
📩 Building AI-native GTM? Get in touch.
🔔 Interested in the Operating Model for AI-native GTM? Follow along.
This essay is versioned. Where our thinking develops materially, we will update the version and explain why - the revision history is preserved, not polished away.