Your Reservoir Model Just Changed the Subject. CLARISSA Kept Up.
The meeting starts with a familiar reservoir model. Then someone asks, “What about CO₂ injection?” Suddenly the model has a new assignment. It would like a moment to update its résumé.
CLARISSA has a useful answer to that kind of change. Our final ATCE manuscript reports a worked case in which the system received a black-oil model, a component table and an instruction to evaluate CO₂ injection. It converted the model to a compositional equation-of-state formulation without further user intervention.
+ component table↓02 / REFORMULATECompositional EOS
+ updated schedule↓03 / CHECK & RUNFour validation gates
+ OPM Flow completion
A new question deserves a matching model
The interesting part is the reformulation. The paper describes a change in how the fluids are represented, followed through into the input sections and operating schedule. That is more substantial than producing a new paragraph about the same old deck.
For engineers, the attraction is practical: an existing model can become the starting point for a different question. The conversation carries the objective; the system does the translation work needed to construct the new case. Engineering judgment still determines whether that case is appropriate.
What CLARISSA actually did
The architect and keywordists authored the equation-of-state and PVT sections from the supplied table.
The Ballparker bracketed the resulting properties against its correlation bands.
The scheduler regenerated schedule content from the updated parameter table.
The manuscript reports that the converted deck passed all four validation gates and ran to completion in OPM Flow. These are the paper’s reported results; this article is not an independent replication.
The visual technology walkthrough shows the roles of the agent, authoring layer, simulator and evaluation. In this case, their value is in keeping a changed engineering objective connected to a concrete, checked simulation input.
A completed run is a starting point for review
The demonstration does not, by itself, establish a field recovery uplift, an economic case for injection or a decision-ready forecast. The paper is explicit: passing its checks establishes a well-formed, self-consistent model, not fitness for every decision.
Capable frontier models can already tackle simulation workflows. CLARISSA’s strength in this reported case is the coordinated reformulation and the checks around it. The paper does not compare this demonstration with Opus 5.5 or establish superiority over it. RIGOR provides a framework for evaluating defined engineering tasks rather than relying on an impressive conversation alone.
On the Road to ATCE: follow the change
The October 3 roadmap also plans compositional work with JutulDarcy. That is a separate planned development track, not evidence of a completed integration. The worked demonstration described here ran on OPM Flow. Keeping those two facts separate makes progress easier to assess.
That distinction matters for our subsurface AI direction: show what has been demonstrated, identify what is planned and let engineers inspect the evidence.
Explore the ATCE presentation or request a CLARISSA demonstration. A useful question to bring: “When the engineering question changes, can you show me every part of the model that had to change with it?”
The reservoir model can keep its résumé. We would rather see its working.
Sources: final manuscript SPE-234136-MS, A Conversational User Interface For Autonomous Reservoir Simulation Deck Generation And Execution, worked compositional reformulation case in Section 6 and limitations in Section 7; Road to ATCE plan dated October 3, 2026. Source documents are not reproduced here.