A thoroughly enjoyable and thought-provoking evening at yesterday’s SPE London Distinguished Lecture, “Beyond the Hype: Transforming Subsurface & Production with AI Innovation.”
Shripad Biniwale of SLB, recently relocated to London after several years in Abu Dhabi, brought a particularly valuable perspective: practical experience of applying AI to real subsurface and production challenges.
Across an excellent hour-long lecture, Shripad explored applications ranging from waterflood management and history matching to fast-track field development planning and the emerging potential of multi-agent engineering.
What particularly interested me was not AI in isolation, but the engineering architecture around it.
One of Shripad’s slides captured the principle particularly well:
“The strongest solutions combine data, physics, optimisation and engineering judgement, rather than forcing every problem into one AI model.”
His field development example demonstrated exactly that. Full-physics simulation preserved fidelity. Machine learning proxies accelerated the exploration of uncertainty. Risk analytics helped rank development options. Engineering judgement ultimately determined which opportunities were robust.
That distinction matters.
AI does not remove the need for engineering judgement. It potentially allows engineering judgement to operate across a much richer body of evidence, much faster.
It also prompted me to think about another dimension of that evidential environment: comparative subsurface benchmarking.
AI can become increasingly sophisticated at understanding the reservoir in front of us: its data, physics, uncertainty and possible development responses.
Comparative benchmarking introduces a complementary question:
What can experience elsewhere, under genuinely comparable subsurface and development conditions, tell us about what may be achievable here?
As subsurface workflows become increasingly AI-enabled, and potentially increasingly agentic, establishing the right engineering context, appropriate comparators and defensible reference points may become more important, rather than less.
There is a fascinating intersection here between AI, physics, comparative evidence and engineering judgement, and one that I think deserves further exploration across the profession.
Many thanks to Shripad for an excellent lecture and for the thoroughly enjoyable discussions during the evening, and to SPE London Section for bringing together such a good group of people.
A particular thank you also to Farid Hadiaman, Wells Perfomance Advisor at bp and SPE London Section Treasurer, for the invitation and for encouraging me to come along. I’m very glad I did.
'
