Baker Hughes takes agentic AI from the back office to the wellhead: real-time automation of drilling and extraction

By Zendoric · September 11, 2026.

Baker Hughes, the oilfield services giant valued at around $63.5 billion, is bringing agentic artificial intelligence (AI) —systems that don't just answer questions, but carry out tasks and chain decisions autonomously within a workflow— to thousands of oil and gas wells, according to Simply Wall St. Its Leucipa platform is moving beyond automating administrative tasks to operating directly on production: adjusting drilling, artificial lift systems and reservoir management in real time.

The move is not an isolated one. According to the same source, other major contractors in the sector —SLB and Halliburton are among the peer companies cited in the article— are pursuing similar digital tools, in what is framed as an industry-wide effort to cut operating costs and redesign service models. The example Simply Wall St offers is the stimulation contract Baker Hughes won from bp in the North Sea, where uptime and reduced non-productive time weigh as heavily as the physical equipment deployed.

The financial backdrop helps explain why this is happening now: the investment thesis circulating on Baker Hughes —as summarized by Simply Wall St, a stock analysis platform that combines data with "narratives" written by its community of investors— bets on a shift toward higher-margin services, with a greater role for software and automation, as opposed to the traditional business of selling hardware by the unit. Embedding agentic AI in actual well operations, and not just in corporate presentations, is how that thesis becomes tangible: more recurring, outcome-linked contracts instead of one-off equipment sales.

The article itself flags the obvious risk: reliability and cybersecurity. If an agentic system fails while managing a producing well, the cost is not an unfortunate chatbot reply, it is a plant shutdown, an environmental incident or a real physical loss. SLB competes with equivalent offerings, so any stumble by Baker Hughes on reliability or delivery would compound the cost inflation and supply chain pressures the sector is already dealing with, according to the same text.

That, in our reading, is what matters most about this story beyond the stock ticker: it is a qualitative step in the deployment of agentic AI, moving out of office software —where the worst-case scenario is a badly written document— and into critical physical infrastructure, where the worst-case scenario is an accident. It is the same logic we have been pointing to in other sectors: when the agent decides and acts on real systems, what has to be governed is not just model quality, but its permissions, its action logs —what decision it made and why— and the limits of what it can touch without direct human supervision.

On the employment front, the pattern matches what we have been observing sector by sector: agentic automation does not replace the reservoir engineer who interprets a complex anomaly, nor the industrial cybersecurity specialist who audits the system, but it does reduce the need for staff dedicated to routine monitoring and manual parameter adjustment in the field. It is the same frontier we have seen in banking, insurance or business administration: fewer hands for the repetitive work, more weight on whoever governs the system and answers when something goes off script.

In the long run, if these platforms deliver what they promise —less non-productive time, less energy wasted per barrel extracted, fewer accidents caused by human error on repetitive tasks— they fit Zendoric's underlying argument: well-governed AI does not just cut costs, it frees up capital and people for higher-value work, including the energy transition Baker Hughes itself says it is pursuing. But that "if" is no small thing. The difference between a technological proof of concept and a new source of operational risk, in a sector where mistakes are measured in spills rather than billing errors, will be decided in execution, not in the sales pitch.