Not a prompt engineer.
The unit of work is the composition, not the model.
Organizations are buying agents before they can describe their own work. That is the order of operations behind most stalled AI programs, and it is a measurement problem before it is a technology problem.
Most organizations have no shared language for the work they actually do. They have org charts, which describe reporting. They have job descriptions, which describe hiring. They have process documentation, which describes what someone hoped would happen. None of these describe how a task is composed: what judgment it requires, what evidence it produces, what has to be decided by a person and what can be delegated.
Without that description, mechanization is guesswork. Agents get pointed at whatever is legible, which is usually whatever was already documented, which is usually not where the work is. The program produces demonstrations rather than capability, and the failure gets blamed on the models.
Organizing is the prerequisite. It is also the part nobody has a method for.
Every credential in circulation assumes the individual is the thing being measured, assessed alone, under conditions that increasingly do not resemble the work. When a task is carried out by a person composing several agents, the thing that performed is the composition. The individual is one component of it.
This is a measurement failure rather than a training failure, and it will not be fixed by adding AI content to existing certifications. It requires asking what the unit of assessment should be, and building instruments that can measure it. That question is what this lab exists to work on.
A description language for work: what a task is made of, who holds the judgment, and what evidence it leaves behind. Published openly.
Read the methodThe role responsible for organizing work before it is mechanized, and for the judgment that cannot be delegated.
Read the role definitionThe measurement apparatus the lab builds and runs, including the ASSAY scorecard.
See what we measure withA five-axis diagnostic for knowledge assets. Developed by the lab and implemented as a working instrument on the cubelet.ai platform, which is the pattern we intend to repeat: method published here, implementation built by others.
The description language the lab's current research programme is built on.
Aggregate findings from completed canvases collected in teaching cohorts and applied engagements. Individual descriptions remain the property of the organizations that made them.
Not a prompt engineer.
The unit of work is the composition, not the model.
The two share ownership, and we would rather state that plainly than have a reader find it.
Euler Center holds the methods and the research. GRID42 builds and distributes platform implementations of them, including cubelet.ai. The two share ownership, and we would rather state that plainly than have a reader find it.
The arrangement is workable because of what the lab does not do. Euler Center holds no partnership with any certification body, and does not sell preparation for credentials it might later assess. Anything of that kind sits with GRID42. Our methods are licensed CC BY so that anyone can implement them, and a second independent implementation of any Euler method is an outcome we would welcome rather than resist.
Founded and led by Sravan Ankaraju, who has spent twelve years in workforce development, training measurement, and evaluation, and has run testing and training operations through several generations of technology change.
Three ways in. All of them assume the same starting point: that the work has to be described before it can be mechanized, measured, or improved.
We organize a body of your work before you mechanize it. The output is a set of Work Composition descriptions covering real tasks, an account of where judgment sits, and a measurement scheme for the parts you intend to delegate.
Read moreFor organizations, funders, and bodies with a measurement question that does not yet have an answer. Scope is agreed in advance and results are published. We do not undertake work whose findings are contingent on the sponsor's preference.
Read moreThe agentic AI supervisor course runs for individuals through Maven and for corporate cohorts directly. Participants work on their own tasks and leave with completed canvases.
Read moreWrite with the question you are trying to answer. A paragraph is enough to tell whether this is a fit.