Chevron, SLB and TotalEnergies Are Advancing Digital Subsurface. But How Do We Validate Autonomous Reservoir Simulation?
CLARISSA · · Industry & Engineering
On October 6, Chevron joined the long-term Arena digital subsurface collaboration established by SLB and TotalEnergies. The effort is focused on next-generation technologies for reservoir engineering, geoscience, uncertainty analysis, optimization and field-development planning. SLB’s announcement ↗
The announcement is another sign that digital subsurface workflows are moving rapidly from experimentation toward core engineering infrastructure.
But as AI becomes capable of doing more of the work, an important question becomes harder to ignore:
How do we know the engineering is right?
Running a reservoir simulator is no longer the only challenge. Frontier AI models are becoming increasingly capable of reading technical documentation, generating simulator inputs, using tools and repairing failures.
The next challenge is verification.
Five Tests for an Autonomous Reservoir-Simulation Agent
1. Does the model actually run?
The first test is still basic: can the system create a valid simulator deck and execute it successfully?
Syntax errors, missing keywords and inconsistent inputs remain useful first-line checks.
But a successful run is only the beginning.
2. Are the physics and engineering assumptions reasonable?
A simulator can converge while the underlying engineering is questionable.
An autonomous system should be able to identify suspicious assumptions involving PVT, saturation functions, relative permeability, reservoir pressure, fluid properties and other physical relationships.
CLARISSA is being designed around this idea: use the intelligence of frontier models while supplying additional physics-grounded checks and reservoir-engineering context.
3. Are wells and operating constraints handled correctly?
Reservoir forecasts depend heavily on how wells are represented.
Rate constraints, pressure limits, injectivity, productivity, completions and operating changes all need to remain internally consistent.
An agent should not simply create a model that runs. It should understand how engineering decisions change reservoir behavior.
4. What happens when something fails?
Failure recovery is one of the most important tests of an autonomous engineering system.
Can the agent distinguish between:
- a syntax problem,
- a numerical problem,
- a bad assumption,
- and a physically unreasonable result?
And when it makes a repair, does it solve the underlying problem—or merely force the simulator to finish?
5. Can another engineer reproduce and audit the result?
This may ultimately be the most important test.
Engineering decisions need to be explainable.
A useful autonomous reservoir system should preserve what changed, why it changed, what assumptions were made, which simulator was used, and how the final result was produced.
That is one reason Blauweiss has developed RIGOR alongside CLARISSA: evaluation becomes more important as the underlying AI becomes more capable. The team’s architecture is intended to add checks, benchmarks and reproducibility around the generated work rather than accepting a result simply because it looks convincing.
The Frontier Is Moving
The Chevron–SLB–TotalEnergies collaboration is important because it reflects a wider shift toward scalable, open and extensible digital subsurface technology. Chevron is contributing technical expertise in uncertainty analysis, optimization and field-development planning directly into that development process. Read the primary source ↗
That is good news for the industry.
It also raises the bar for everyone building reservoir AI.
The question is no longer simply:
Can AI create and run a reservoir model?
The better question is:
Can we measure whether the resulting engineering deserves to be trusted?
That is the problem CLARISSA and RIGOR are being built to address.
Better AI makes verification more important—not less.
Bring your engineering questions to ATCE
Explore the CLARISSA ATCE presentation and session details, or request a CLARISSA demonstration to discuss verification in your own reservoir workflow.