Give your agent a brain like ours.
Cloned from the Quarterback.
The brain
The brain has two sides: one for language and one for math.
LANGUAGE
A large language model.
It converses and reasons. The side your agent already has: it converses with your users wherever they work.
MATH
Reinforcement Learning.
It decides and acts. The side Dynamiks adds: trained in a simulator of your process, fine-tuned on your data, learning continuously.
THE ARCHITECTURE
Decisions as a Service
How a decision model learns: by doing, the way people do.
YOUR AGENT
The world
1 · State and reward
2 · Next action
DECISIONS AS A SERVICE
The policy
Your agent holds your process, applies the action and reports what changed. The service chooses the action.
It learns by trial and error.
It acts, sees the result and adjusts. Tens of thousands of times.
It trains in a simulator.
A world built to be as close as possible to the real process.
It keeps improving.
Every decision and its outcome can feed the next training.
“Nobody’s reading books to learn how to play baseball. We all have to try and fail many times.”
Nicolas Maquaire
THE METHOD
Eight steps from a simulator to decisions in production.
Five of them are already built and run as they are. Only three change for your process: the world, the process and the rewards.
1
BUILD THE WORLD
Our job
2
MODEL THE PROCESS
Our job
3
SHAPE THE REWARDS
Our job
4
TRAIN THE POLICY
Already built
5
TUNE AND EVALUATE
Already built
6
SERVE THE DECISIONS
Already built
7
ORCHESTRATE
Already built
8
CONVERSE
Already built
With these foundations, a new agent takes a few weeks.
Bring us your process.
Only the decisions differ. Tell us the process and the next action that matters, and we will build its agent.