Where we stand
- Founded
- 2025, London
- Company
- Independent
- Building
- Our first applications. In development, not yet in live use.
The work sets the starting point. Evidence sets the next step.
Noetic designs, builds and measures AI applications for work where mistakes matter.
Step 1: Establish the work
How it is really done, who does it, and what a mistake would cost.
Step 2: Define the first useful step
And its limits, agreed in writing before anything is built.
Step 3: Test before extending
Against the agreed criteria, with silent running and supervised use where the consequences require it.
Step 4: Revisit when things change
A new model, new rules, a changed task: keep, revise, extend or withdraw.
Alongside
We explore new capabilities within agreed limits, on synthetic or low-stakes tasks. What we learn there informs the next change. It never goes straight into use.
Every change has to earn its place.
The starting arrangement varies. The basis for accepting a change stays the same.
Some work starts with a person and better tools. Some starts with a model under review. Either way, nothing goes into use until it meets criteria agreed in advance, and every decision is kept in the record.
New model · accepted for two steps, same criteria
Models change. The understanding of the work stays.
Representative cases
The tasks that show what the work really involves, including the awkward ones.
Known failures
The cases each approach missed, kept with the examples.
Agreed criteria
What counts as a good result, reviewed with the people who do the work.
Evidence about changes
Which changes helped, on which tasks, with which model version.
When something changes, whether a new model, a new rule, a change to the task or a tighter budget, this record shows which parts of the application should change and which should be left alone. We keep it up to date as the work changes.
From the demo to daily use.
For organisations whose work is consequential and document-heavy, and who want an AI application they can measure and maintain.
- 01A short discovery to understand the work and agree what success means.
- 02A working application built with a capable model and measured against those criteria.
- 03The steps that make it cheaper and more dependable as the work becomes understood.