{ "as_of": "2026-10-05", "timelines": [ { "company": "Devon Energy", "id": "AI-006", "date": "2026-02-18", "stage": "Rollout discussed", "title": "Moving beyond pilots", "detail": "Management connects second-half 2025 lift trials with broader 2026 rollout. No fleet result quantified.", "source_id": "S02", "source_url": "https://s2.q4cdn.com/462548525/files/doc_financials/quarterly/2025/q4/Q4-2025-DVN-Q-A-Webcast-Commentary-Transcript.pdf" }, { "company": "Devon Energy", "id": "AI-006", "date": "2026-05-06", "stage": "Scaled", "title": "More than 850 wells live", "detail": "Management reports a 2–3% production uplift in the earlier 2025 gas-lift pilot. That range does not measure the live fleet.", "source_id": "FU001", "source_url": "https://s2.q4cdn.com/462548525/files/doc_financials/quarterly/2026/q1/Q1-2026-DVN-Q-A-Webcast-Commentary-Transcript.pdf" }, { "company": "Devon Energy", "id": "AI-006", "date": "2026-08-05", "stage": "Scaled", "title": "1,000 wells live", "detail": "Real-time autonomous lift optimization includes first legacy Coterra wells. More than 2,000 additional wells are prospective.", "source_id": "SW8-001", "source_url": "https://s2.q4cdn.com/462548525/files/doc_financials/quarterly/2026/q2/Q2-2026-DVN-Q-A-Webcast-Commentary-Transcript.pdf" }, { "company": "Chord Energy", "id": "AI-097", "date": "2026-02-25", "stage": "Scaled", "title": "About 99% of rod-lift wells", "detail": "FY2025 release reports rollout during 2025 and an early ~25% improvement in rod-pump runtime. Exact runtime cohort is undisclosed.", "source_id": "SW7-005", "source_url": "https://ir.chordenergy.com/2026-02-25-Chord-Energy-Reports-Fourth-Quarter-and-Full-Year-2025-Financial-and-Operating-Results%2C-Issues-2026-Outlook-and-Declares-Base-Dividend" }, { "company": "Chord Energy", "id": "AI-097", "date": "2026-08-05", "stage": "Continuing use", "title": "Artificial-lift optimization continues", "detail": "Q2 release confirms ongoing use. Broader program results cannot be assigned entirely to AI.", "source_id": "SW7-007", "source_url": "https://ir.chordenergy.com/2026-08-05-Chord-Energy-Reports-Second-Quarter-2026-Financial-and-Operating-Results%2C-Declares-Base-Dividend-and-Updates-2026-Outlook" }, { "company": "Diamondback Energy", "id": "AI-045", "date": "2025-11-03", "stage": "Development", "title": "AI for lift choice and late-life wells", "detail": "Stockholder letter describes work under development; no measured AI-only outcome.", "source_id": "SW7-011", "source_url": "https://ir.diamondbackenergy.com/node/17156/pdf" }, { "company": "Diamondback Energy", "id": "AI-045", "date": "2026-05-04", "stage": "Scaling discussed", "title": "Automation and predictive maintenance", "detail": "Management describes scaling field automation and data-driven maintenance, including AI initiatives. No rollout denominator.", "source_id": "SW7-012", "source_url": "https://ir.diamondbackenergy.com/news-releases/news-release-details/letter-stockholders-issued-diamondback-energy-inc-11" }, { "company": "EQT", "id": "AI-095", "date": "2026-02-18", "stage": "Unclear", "title": "AI-assisted service rebids", "detail": "CEO describes tools that would help target rebids. The disclosure does not establish production usage or AI-specific savings.", "source_id": "EQTD01", "source_url": "https://s205.q4cdn.com/630272887/files/doc_financials/2025/q4/EQT-Q4-2025-Earnings-Call-Transcript.pdf" }, { "company": "Continental Resources", "id": "AI-248", "date": "2025-09-29", "stage": "Award announcement", "title": "UiPath AI25 recognition", "detail": "The detailed vendor case describes operating email validation and engineer data delivery. The award date is not a go-live or outcome date.", "source_id": "SW7-003", "source_url": "https://www.uipath.com/newsroom/uipath-announces-2025-ai25-awards-winners" }, { "company": "ConocoPhillips", "id": "AI-202", "date": "2026-04-01", "stage": "Pilot", "title": "Four-month Montney evaluation", "detail": "Vendor-authored case reports 3–4% production above forecast and ~5% lower LOE. Evaluation dates and well denominator are undisclosed.", "source_id": "FU003", "source_url": "https://jpt.spe.org/case-study-field-deployments-of-ai-based-iocaas-advancing-artificial-lift-and-flow-assurance" }, { "company": "ExxonMobil", "id": "AI-235", "date": "2025-10-01", "stage": "Scaled", "title": "More than 1,300 Permian wells", "detail": "JPT deployment highlight describes closed-loop ML gas-lift optimization. Earlier trial results are not newly measured 2025 outcomes.", "source_id": "SW4-005", "source_url": "https://jpt.spe.org/gas-lift-optimization-in-the-permian-uses-machine-learning-artificial-intelligence-restricted" }, { "company": "ExxonMobil", "id": "AI-178", "date": "2025-10-23", "stage": "Operating described", "title": "AI monitors production bottlenecks", "detail": "Company article describes use at operating sites. No AI-only cash result is established by this disclosure.", "source_id": "RS112", "source_url": "https://corporate.exxonmobil.com/locations/india/newsroom/2025/turning-artificial-intelligence-into-business-intelligence" } ], "timeline_note": "The dates show when companies talked about the work, which may be later than when it started. We keep earlier pilots and future plans separate.", "search_summary": { "queries": 34, "vectors": 13, "stopped": 0, "regulator_orders": 2, "vendor_ended": 0, "vendor_note": "Vendor contracts we checked were renewed, not ended: C3 AI–Baker Hughes runs through June 2028, and Shell extended its C3 AI reliability agreement in June 2026." }, "failures_new": [ { "key": "havtil-equinor", "tone": "regulator", "window": "in", "company": "Equinor · Havtil (Norwegian Ocean Industry Authority)", "date": "19 December 2025", "kind": "Automation qualification · regulator order", "title": "Regulator blocked wider use of automated drilling until it is qualified", "summary": "Havtil found Equinor had taken Novos automated drilling control into use on further facilities while the technology’s qualification stayed at TRL-4 (“ready for first use”). It called that “a serious breach” and ordered Equinor to show, through qualification, that requirements can be met before more facilities use it — deadline 1 June 2026.", "scope": "Novos automated drilling control (NOV). In use on five named facilities at the original 2022 audit. We did not check compliance status after the June 2026 deadline.", "lesson": "Rolling drilling automation out across more rigs can get ahead of formal technology qualification, and a regulator can then stop further rollout until the evidence catches up.", "limit": "Havtil does not describe Novos as AI; it is included because drilling automation is within this site’s scope. The order is a qualification and compliance finding. It does not say the system malfunctioned or caused an incident.", "url": "https://www.havtil.no/en/supervision/important-messages/2025/order-to-equinor--technology-development-and-use-of-digital-well-planning/" }, { "key": "havtil-var", "tone": "regulator", "window": "in", "company": "Vår Energi · Havtil", "date": "5 February 2026", "kind": "Automation qualification · regulator order", "title": "Three non-conformities in how automated drilling is qualified", "summary": "A December 2025 Havtil audit of how Vår Energi assesses risk and qualifies automated drilling operations found three non-conformities: technology qualification, analyses and non-conformity handling. Havtil ordered criteria for developing, testing and using automated drilling technologies to be in place before such systems are taken into use.", "scope": "The notice names no rig, field or system and does not say whether automated drilling was already in use.", "lesson": "Offshore Norway, documented qualification criteria and risk analyses are a precondition for automated drilling. Without them, deployment waits.", "limit": "The notice does not mention AI. It is a compliance finding, not evidence of a technical failure.", "url": "https://www.havtil.no/en/supervision/audit-reports/2026/var-energi--risk-assessment-and-qualification-of-automated-drilling-operations/" }, { "key": "ey-survey", "tone": "survey", "window": "in", "company": "EY Future of Energy Survey · oil & gas and chemicals, n=508", "date": "March 2025 · fielded Nov–Dec 2024", "kind": "Survey · perceptions, not outcomes", "title": "27% say emerging technology is delivering on its promise", "summary": "EY’s survey found that just 27% of executives say emerging technologies such as AI are delivering on their promise, and 85% cite difficulty scaling successful pilots into organisation-wide implementation.", "metrics": [ [ "Say AI is delivering", 27 ], [ "Find pilots hard to scale", 85 ] ], "metrics_note": "Share of survey respondents. Oil & gas is pooled with chemicals.", "scope": "508 respondents (Americas 285, EMEA 115, APAC 108); technology questions went to director level and above.", "lesson": "Investment intent is running ahead of perceived results, and scaling beyond the pilot is the barrier executives name most.", "limit": "These are perceptions, not measured project outcomes or counts of abandoned pilots. EY sells digital services, so read the framing with care.", "url": "https://www.ey.com/content/dam/ey-unified-site/ey-com/en-us/campaigns/energy-resources/documents/ey-ogc-future-of-energy-external-report.pdf" } ], "failures": [ { "key": "adnoc-gaslift", "tone": "limit", "window": "in", "company": "ADNOC · AIQ", "date": "3 November 2025", "kind": "Deployment constraints", "title": "Gas-lift control still needs a safe response to failed sensors", "summary": "An operator- and AIQ-authored paper describes autonomous gas-lift optimization across 200 onshore wells. Its control logic includes validation, fallback and override for sensor or valve failures.", "scope": "The paper reports up to 2% incremental oil and 5–30% gas-injection reductions at well level. These measures have different denominators.", "lesson": "A deployed model needs a safe fallback and a way to update its training data. The authors describe poorer accuracy where operating data are missing, alongside inferential models and transfer learning.", "limit": "This is a deployment constraint, not a cancelled project. Publisher-indexed abstract reviewed; full paper, independent validation and full costs were not available.", "url": "https://onepetro.org/SPEADIP/proceedings-abstract/25ADIP/25ADIP/D021S043R003/793441" }, { "key": "esp", "tone": "test", "window": "in", "company": "Baker Hughes · operator unnamed", "date": "29–30 October 2024", "kind": "Pre-deployment test result", "title": "An ESP failure model missed 9 of 18 held-out failures in testing", "summary": "In a vendor-authored SPE paper, the model caught half of the 18 pump failures held out for pre-deployment testing (50% sensitivity); the nine it missed included five short circuits. Specificity was 78% and balanced accuracy 64%. The model was then deployed after client approval.", "scope": "Three Americas assets; 137 ESPs in the dataset; 83 of 101 recorded failures used for training and 18 held out for testing. Definitions are specific to this study.", "lesson": "Ten-minute telemetry and missing sensor data limited what the model could detect. A model can be useful enough to deploy and still miss half of the failures it is meant to catch — buyers should ask for held-out sensitivity, not just success stories.", "limit": "Operator unnamed; no independent replication or program economics. This is a test result, not evidence that the deployment was abandoned.", "url": "https://dam.bakerhughes.com/asset/2145a967-01f1-4ffa-8ee1-37b085e6fff1/spe-221571-ms-bkr-licensed.pdf", "locator": "SPE-221571-MS, pp. 4–7", "metrics": [ [ "Sensitivity", 50 ], [ "Specificity", 78 ], [ "Balanced accuracy", 64 ] ] }, { "key": "slb", "tone": "limit", "window": "in", "company": "SLB · Microsoft customer case", "date": "Source updated August 2025", "kind": "Limitation observed · exception handling", "title": "Some shipping documents still need a person", "summary": "Inside a success story, Microsoft’s case describes a step that routes documents the model cannot process to process owners for manual handling, new templates or retraining.", "scope": "Document-processing workflow; no exception rate published.", "lesson": "Exception handling and retraining are part of running the system, not signs that it failed.", "limit": "Vendor customer case; update date does not establish go-live. No project failure or withdrawal documented.", "url": "https://learn.microsoft.com/en-us/power-platform/guidance/case-studies/slb-enhances-productivity", "record": "AI-234" }, { "key": "oxy", "tone": "limit", "window": "in", "company": "Oxy", "date": "23 October 2025", "kind": "Limitation observed · connectivity", "title": "Remote-site connectivity became the second-phase question", "summary": "An SPE webinar description covers a two-phase construction-safety vision pilot in the Permian. Phase one is described as successful; phase two examined deployment where communications infrastructure is limited.", "scope": "Oil-processing-facility construction site. Phase-two outcomes are not detailed.", "lesson": "A field pilot needs a workable connection and operating setup before it can scale.", "limit": "Event description, not a technical evaluation; no error rate or cancellation disclosed.", "url": "https://connect.spe.org/rdts/events/eventdescription?CalendarEventKey=02eb6208-5245-4fbf-87c3-0199eb34b84c&CommunityKey=a0c6b65c-a27b-496c-8e07-9eb5cbfcea2c" }, { "key": "norce", "tone": "limit", "window": "out", "company": "NORCE / UiS / Stanford", "date": "Published 22 May 2025 · 2021 virtual benchmark", "kind": "Limitation observed · benchmark", "title": "Repeated geosteering runs produced different well paths", "summary": "NORCE reports meaningful trajectory differences between repeated runs of the same AI agent under geological uncertainty in a 2021 virtual competition. Average performance remained competitive with human participants.", "scope": "Virtual geosteering benchmark, not an operating well.", "lesson": "Decision variability under geological uncertainty needs explicit evaluation.", "limit": "Research-institution summary of earlier competition work; no field deployment involved.", "url": "https://www.norceresearch.no/en/news/ai-vs-human-decision-making" }, { "key": "novi", "tone": "limit", "window": "out", "company": "Novi Labs · customer unnamed", "date": "5 July 2023", "kind": "Limitation observed · data quality", "title": "A unit mismatch distorted a forecasting pilot", "summary": "Novi describes unexpectedly low forecasts traced to completion-fluid records that mixed barrels and gallons.", "scope": "Anonymous customer pilot; published before the October 2024 – October 2026 window.", "lesson": "Input units need validation before a model’s forecast is interpreted.", "limit": "First-hand vendor account; correction, cost and eventual deployment outcome are not quantified.", "url": "https://novilabs.com/blog/11-costly-mistakes-ep-operators-make-in-building-machine-learning-capabilities/" }, { "key": "watch16am-total-drum", "tone": "limit", "window": "in", "company": "TotalEnergies", "date": "2026-06-11", "kind": "Reported decision to drop one use case; issuer corroboration pending", "title": "TotalEnergies reportedly dropped one drum-cycling use case", "summary": "A later conference account raises a different test: does the prediction help the operator’s actual decision?", "scope": "One reported use-case decision inside AI-207. Other work in the broader pilot continued.", "lesson": "Test the decision the operator needs to make, including conditions outside the predicted variable.", "limit": "Original conference reporting, labeled Sponsored. We could not retrieve the presentation or a later company statement. The earlier company release confirms the broader pilot, not the decision to drop this use case. Keep these figures separate from its five-event result.", "url": "https://www.controlglobal.com/show-coverage/honeywell-users-group/article/55383668/honeywell-ai-pilot-aids-coker-unit-operations-at-totalenergies-refinery", "record": "AI-207" } ], "failure_finding": "A 16 October review adds a reported decision to drop one use case at TotalEnergies, still awaiting confirmation from the company or original presentation. This is distinct from abandoning the whole pilot. The figures below remain the dated 5 October search results; other examples cover pauses, model errors and operating limits.", "survivorship_note": "Companies and vendors publish successes more often than stopped pilots. Some projects may have ended quietly. We can’t calculate an industry success rate from these accounts, and we still need better evidence on cancellations.", "search_scope": "The figures above describe the 5 October failure search, not a cumulative failure rate. Later dated source reviews include a Devon development pause and the reported TotalEnergies use-case decision. Newly found evidence is not necessarily recent news.", "latest_targeted_review_on": "2026-10-16" }