SUNYA / ENERGY AI OBSERVATORY / 8 OCTOBER 2026

Where AI changes the work

Eight selected upstream companies, examined within the wider public-source register. The examples are not a market ranking or an exhaustive company inventory.

Public-source research, October 2024–October 2026. Reported performance has not been independently measured. Reach, pilot results, targets and cash returns stay separate.

Three conclusions to take into the room

A good answer only helps if someone can use it.

Analyst synthesis: Devon describes software that adjusts lift settings. Occidental describes a system that helps decide where field crews should focus. Continental describes agents that check incoming well data. Each changes a different part of the job. Before comparing the tools, ask what happens next, who can authorize it and how the result is checked.

What to do with that: If you run an asset, start with a decision people make regularly and find out who acts on it. If you sell software, show how your answer reaches that person or system.

What would change this conclusion: We would rethink this if tools repeatedly created substantial benefits without changing a decision, task or operating process.

Evidence: AI-006, AI-215, AI-248

EOG and EQT show two ways to organize the work.

Analyst synthesis: EOG describes building software for its own operations. EQT describes a shared digital workspace for important processes and its main communication platform. That could make new tools easier to put into everyday use. It does not show that either company earns better AI returns.

What to do with that: Look at how people use the tools, how they connect to existing systems and who keeps them working. A company can have useful software and processes even if it makes few public AI announcements.

What would change this conclusion: Do these systems help teams launch tools faster, keep using them or make better decisions? Until we know, we can only say they might give the company a head start.

Evidence: AI-089, AI-095

We know where some tools run. We know less about what they earn.

Analyst synthesis: Devon reports how many wells use its system. Chord reports rod-lift coverage and an early improvement in runtime. Conoco’s supplier compares performance with a forecast. Those figures measure different things over different periods. They do not give us a common measure of cash return or show how rollout is changing across the industry.

What to do with that: Use these examples to decide what to investigate. Before trying the same approach, ask which assets were compared, how they performed before, what it costs to run and what still goes wrong.

What would change this conclusion: A fair comparison showing benefits that last after integration, service and support costs would tell us more than a well count or a comparison with a forecast.

Evidence: AI-006, AI-097, AI-202

Follow the work, company by company

CompanyTaskActionReported reachReported result
Devon EnergyClosed-loop lift optimization.Adjusts lift settings in real time, around the clock. The record describes a controller, rather than only an engineering recommendation.1,000 live wells in August 2026. More than 2,000 additional Permian wells are an opportunity, not current deployment.A separate 2025 gas-lift pilot reported 2–3% uplift in May 2026. That result is not established for the full August fleet.
Chord EnergyAI-driven rod-lift optimization.Optimizes rod-lift pumping. The reviewed record does not specify whether adjustments are autonomous or require approval.99% of rod-lift wells in the 2025 disclosure. This is not 99% of all company wells.Early approximately 25% improvement in rod-pump runtime. The issuer also associates optimization with about 1,200 fewer workover rig days in 2025; broader cost improvements are not isolated to AI.
ConocoPhillipsMonitor artificial lift, hydrate risk and compressor condition.Flags exceptions for engineers and operators. The Montney narrative describes interventions on chemicals and equipment; it does not establish fully autonomous control there.Kickoff to full deployment in roughly four months at the described Montney development. Asset-wide well count and exact deployment dates are not supplied.Vendor reports production 3–4% above forecast and approximately 5% lower LOE after four operating months. A 6% uplift is anticipated, not observed.
OccidentalPrioritize lease-operator attention in routeless operations.Directs field attention to exceptions rather than a routine route. Remote centers resolve some issues; field visits still remain part of the operating model.Approximately 40% of U.S. production under routeless/remote-center coverage in February 2026. Production share is not a well or personnel count.Management reports roughly 300 issues per day resolved remotely during a Rockies winter storm. The result belongs to the combined operating model, not AI alone.
Continental ResourcesInterpret incoming data, validate context and score accuracy.UiPath and Databricks agents check data. Robots return errors to suppliers and deliver validated data to engineers.Undated vendor case reports more than 200 emails per day and one reusable robot supporting more than 40 processes. Those are not 40 AI initiatives.Vendor reports about five minutes saved per previously manual email. Award material reports accuracy rising from roughly 70% to the high 90% range; dates and test method are unspecified.
EOG ResourcesProprietary machine-learning production optimization.Annual report says the optimizer improved base production and runtime. It does not specify the model, the number of wells or the human-control boundary.No AI-specific rollout count. More than 140 in-house applications is a separate software-estate measure; those applications are not all established as AI.No AI-specific numerical outcome. The report's 7% lower average well cost is a wider operating result.
EQTTarget service rebids and procurement effort.Points procurement attention toward potential rebids. The call does not describe automated tendering, contract awards or a confirmed live workflow.No rebid-tool deployment count. At year-end 2025, 56% of employees worked remotely; the filing describes a primary platform for communication and critical processes.Prospective low-single-digit procurement cost reduction, not a realized saving or isolated AI effect.
Diamondback EnergySelect and optimize lift type and cost using machine learning and AI.The company identifies downtime and late-life performance as intended improvements. It does not disclose how recommendations become equipment changes.No well count or eligible fleet denominator. The reviewed letters describe early work and expected future gains.No result specifically attributed to this workflow. Expected uptime and LOE improvements should remain separate from reported company-wide results.
The examples and denominators differ. This is not an ROI or company capability ranking.

The gap between a working task and a cash case

An evidence map of selected sources, not a company scorecard. Incomplete includes earlier pilots, combined results, forecast comparisons, undated metrics and prospective work. Not established means the selected sources leave the question open.

Selected workflowAction describedOperating reachOutcome comparisonFull cost and cash
Devon EnergyDescribed
Adjusts lift settings in real time, around the clock. The record describes a controller, rather than only an engineering recommendation.		Described
1,000 live wells in August 2026. More than 2,000 additional Permian wells are an opportunity, not current deployment.		Incomplete
A separate 2025 gas-lift pilot reported 2–3% uplift in May 2026. That result is not established for the full August fleet.		Not established
No full integration, service, support and net cash comparison is established for this selected workflow.
Chord EnergyIncomplete
Optimizes rod-lift pumping. The reviewed record does not specify whether adjustments are autonomous or require approval.		Described
99% of rod-lift wells in the 2025 disclosure. This is not 99% of all company wells.		Incomplete
Early approximately 25% improvement in rod-pump runtime. The issuer also associates optimization with about 1,200 fewer workover rig days in 2025; broader cost improvements are not isolated to AI.		Not established
No full integration, service, support and net cash comparison is established for this selected workflow.
ConocoPhillipsDescribed
Flags exceptions for engineers and operators. The Montney narrative describes interventions on chemicals and equipment; it does not establish fully autonomous control there.		Incomplete
Kickoff to full deployment in roughly four months at the described Montney development. Asset-wide well count and exact deployment dates are not supplied.		Incomplete
Vendor reports production 3–4% above forecast and approximately 5% lower LOE after four operating months. A 6% uplift is anticipated, not observed.		Not established
No full integration, service, support and net cash comparison is established for this selected workflow.
OccidentalDescribed
Directs field attention to exceptions rather than a routine route. Remote centers resolve some issues; field visits still remain part of the operating model.		Described
Approximately 40% of U.S. production under routeless/remote-center coverage in February 2026. Production share is not a well or personnel count.		Incomplete
Management reports roughly 300 issues per day resolved remotely during a Rockies winter storm. The result belongs to the combined operating model, not AI alone.		Not established
No full integration, service, support and net cash comparison is established for this selected workflow.
Continental ResourcesDescribed
UiPath and Databricks agents check data. Robots return errors to suppliers and deliver validated data to engineers.		Incomplete
Undated vendor case reports more than 200 emails per day and one reusable robot supporting more than 40 processes. Those are not 40 AI initiatives.		Incomplete
Vendor reports about five minutes saved per previously manual email. Award material reports accuracy rising from roughly 70% to the high 90% range; dates and test method are unspecified.		Not established
No full integration, service, support and net cash comparison is established for this selected workflow.
EOG ResourcesIncomplete
Annual report says the optimizer improved base production and runtime. It does not specify the model, the number of wells or the human-control boundary.		Not established
No AI-specific rollout count. More than 140 in-house applications is a separate software-estate measure; those applications are not all established as AI.		Not established
No AI-specific numerical outcome. The report's 7% lower average well cost is a wider operating result.		Not established
No full integration, service, support and net cash comparison is established for this selected workflow.
EQTIncomplete
Points procurement attention toward potential rebids. The call does not describe automated tendering, contract awards or a confirmed live workflow.		Not established
No rebid-tool deployment count. At year-end 2025, 56% of employees worked remotely; the filing describes a primary platform for communication and critical processes.		Not established
Prospective low-single-digit procurement cost reduction, not a realized saving or isolated AI effect.		Not established
No full integration, service, support and net cash comparison is established for this selected workflow.
Diamondback EnergyIncomplete
The company identifies downtime and late-life performance as intended improvements. It does not disclose how recommendations become equipment changes.		Not established
No well count or eligible fleet denominator. The reviewed letters describe early work and expected future gains.		Not established
No result specifically attributed to this workflow. Expected uptime and LOE improvements should remain separate from reported company-wide results.		Not established
No full integration, service, support and net cash comparison is established for this selected workflow.
AI-006 / Equipment settings / Operator reported

Devon Energy

Autonomous artificial-lift optimization at operating wells

Data going in
Operating-well data; the latest disclosure does not give a full signal list or lift-type breakdown.
Task
Closed-loop lift optimization.
Action coming out
Adjusts lift settings in real time, around the clock. The record describes a controller, rather than only an engineering recommendation.
Control and approval
Closed-loop adjustment disclosed. Limits, overrides and stop rules are not described in the reviewed passage.
Delivery model
Management describes a mix of internal work and provider partnerships. The August Q&A does not name the partners for this lift application.
Reported reach
1,000 live wells in August 2026. More than 2,000 additional Permian wells are an opportunity, not current deployment.
Reported result
A separate 2025 gas-lift pilot reported 2–3% uplift in May 2026. That result is not established for the full August fleet.
Why this is worth studying (analyst interpretation)
The first legacy Coterra wells were included by the August update. This is evidence of some transfer, not a measured result across the acquired portfolio.
Still unknown
How many wells could use it? Which lift types are included? How does the whole fleet perform against comparable wells? How often do people override it, and what does support cost?
Operator question (analyst proposal)
Which lift types and asset conditions reproduce the pilot result after sustained operation?
Supplier question (analyst proposal)
Which part of the operating loop remains an integration, reliability or monitoring problem the company wants external help to solve?
A useful test (not yet run)
Compare eligible treated wells with a matched group, including downtime, energy use, overrides and costs. Keep the earlier pilot and later rollout periods separate.
Claim boundary
Management claims; no independent fleet-wide AI attribution or net cash return.
Basis: canonical initiative record and account facts DVN-F02, DVN-F03.

Original sources and review limits

  1. Devon Energy — deployment evidenceSW8-001 / 2026-08-05 / Devon’s operator-hosted Q2 2026 transcript (2026-08-05) says its closed-loop AI system autonomously optimized 1,000 wells in real time, 24/7; management said 200 wells in March and a path to broad deployment across >2,000 additional Permian wells, including first legacy Coterra wells. This supersedes the Q1 count of >850.
    Research review: Relevant source passages reviewed Comparison recheck 2026-10-08: Original PDF downloaded; relevant prepared remarks and Q&A reviewed. PDF pp. 5, 16-17: 1,000 wells, more than 2,000 opportunity, first legacy Coterra wells and provider partnerships.
  2. Devon Energy Q2 2026 Earnings Call TranscriptRS034 / 2026-08 / CEO remarks on Slide 9 and Q&A; source transcript not issuer-hosted
    Research review: Audit access: inaccessible. Earlier: excerpt reviewed
  3. Q4 2025 commentary transcriptS02 / 2026-02 / PDF p. 5; Clay Gaspar, prepared remarks
    Research review: Audit access: excerpt reviewed. Earlier: inaccessible
  4. AI-enabled lift and maintenance drive $1B optimization planR040 / 2026-02 / Locator not independently checked
    Research review: Audit access: full text reviewed. Earlier: Imported reference; source verification pending
  5. Smart Gas Lift closed-loop AI scaled to 850+ wellsR041 / 2026-05 / Locator not independently checked
    Research review: Audit access: full text reviewed. Earlier: Imported reference; source verification pending
  6. AI speeds Coterra integration; in-house plus vendor modelR042 / 2026-08 / Locator not independently checked
    Research review: Audit access: excerpt reviewed. Earlier: Imported reference; source verification pending
  7. Devon Q1 2026 earnings call — Smart Gas LiftFU001 / 2026-05-06 / Prepared remarks p. 4; Q&A pp. 12–14
    Research review: Latest access review: original document available; pp. 3-4 (PDF pp. 3-4), CEO Clay Gaspar prepared remarks, lines 108-128; fully read 20-page transcript. Earlier: Document accessible on 1 October 2026. New review: pp. 3-4 (PDF pp. 3-4), CEO Clay Gaspar prepared remarks, lines 108-128; fully read 20-page transcript. Earlier claim-level findings remain scoped to their original review.
  8. Devon Q2 2026 earnings call and transcriptDVN-S01 / 2026-08-05 / Q&A webcast commentary transcript, prepared remarks and CTO Q&A
    Research review: Reviewed 2026-10-07. Management statements; well coverage has no eligible-well denominator or independently validated uplift.
AI-097 / Equipment settings / Operator reported

Chord Energy

AI-driven rod-lift optimization

Data going in
Rod-lift pumping operations; model inputs and architecture are not disclosed.
Task
AI-driven rod-lift optimization.
Action coming out
Optimizes rod-lift pumping. The reviewed record does not specify whether adjustments are autonomous or require approval.
Control and approval
Control and approval boundary unknown.
Delivery model
Operator describes machine learning; vendor and platform are not identified.
Reported reach
99% of rod-lift wells in the 2025 disclosure. This is not 99% of all company wells.
Reported result
Early approximately 25% improvement in rod-pump runtime. The issuer also associates optimization with about 1,200 fewer workover rig days in 2025; broader cost improvements are not isolated to AI.
Why this is worth studying (analyst interpretation)
A near-fleet rod-lift rollout is worth studying for reliability and support, but does not establish applicability to ESPs or gas lift.
Still unknown
What does runtime mean here? What was the comparison? What did interventions and service fees cost, and how often did the system fail or need an override?
Operator question (analyst proposal)
Did longer runtime reduce workover cost without trading off production or maintenance risk?
Supplier question (analyst proposal)
Who supports this optimization at scale, and which data or equipment interfaces constrain its next extension?
A useful test (not yet run)
Reconcile runtime and workover records by comparable rod-lift well cohort, including production, repair cost and the period before rollout.
Claim boundary
Company-reported early performance; no disclosed controlled comparison or isolated net AI savings.
Basis: canonical initiative record and account facts CHRD-F02, CHRD-F03, CHRD-F06.

Original sources and review limits

  1. Chord Energy FY2025 results releaseSW7-005 / 2026-02-25 / Operations Update, lines 210–216; specifically line 216 (AI deployment, run time metric, workover rig days).
    Research review: Full source opened and read.
  2. AI rod-pump optimization arrests base declineR059 / 2026-05-06 / Locator not independently checked
    Research review: Audit access: full text reviewed. Earlier: excerpt reviewed
  3. AI rod-pump program deployed field-wideR060 / 2026-08 / Locator not independently checked
    Research review: Audit access: full text reviewed. Earlier: Imported reference; source verification pending
  4. Chord Energy 1Q26 Earnings PresentationSW7-006 / 2026-05-05 / Slide 16; full extracted slide lines 457–505.
    Research review: Full 28-page PDF opened; slide 16 read.
  5. Chord Energy 2Q26 results releaseSW7-007 / 2026-08-05 / Operations Update, lines 156–161; continued AI artificial-lift optimization, with no fresh AI-specific result.
    Research review: Full source opened and read.
  6. Chord Energy 2Q25 Investor PresentationSW7-008 / 2025-08-06 / Slide 11 / parsed PDF page 10, lines 273–306.
    Research review: Full 24-page PDF opened; lists AI/ML scheduling and artificial-lift optimization, gas-lift algorithms, predictive maintenance, and remote monitoring as continuous-improvement initiatives; no distinct deployed AI result or attribution.
  7. Chord Energy Q4 2025 results and 2026 outlook presentationCHRD-S01 / 2026-02-25 / Slide 8; AI-driven ML on rod lift and 2025 run-rate results
    Research review: Reviewed 2026-10-07. AI is one of several continuous-improvement categories; group savings and workover days cannot be attributed solely to AI. Comparison recheck 2026-10-08: Original PDF downloaded; relevant initiative table reviewed. PDF p. 8: scaled AI to 99% of rod-lift wells; separate ESP measures and broader Production and LOE savings.
  8. Chord Energy Q3 2025 earnings-call transcriptCHRD-S03 / 2025-11-05 / COO Q&A on AI rod-pump parameter control
    Research review: Reviewed 2026-10-07. Secondary transcript; management detail, not independent technical validation.
  9. Chord FY2025 results: operations updateCHRD-A01 / 2026-02-25 / Issuer FY2025 release, Operations Update (web lines 216+): approximately 99% of rod-lift wells, early ~25% improvement in rod-pump run times; company says optimization reduced failures and resulted in ~1,200 fewer workover-rig days in 2025. Cross-check the issuer presentation, printed slide 8 / PDF page 8 (web lines 247–259): grouped ~$50MM continuous-improvement savings bucket includes >50% ESP cycle-time reduction, 25% failure-rate improvement, and scaled AI to 99% of rod-lift wells.
    Research review: Reviewed 2026-10-07. Company-reported optimization-associated result; broader dollar savings combine several initiatives.
AI-202 / Field work / Vendor reported

ConocoPhillips

Montney AI surveillance and flow assurance

Data going in
Standardized SCADA and historians at the Montney asset; the vendor describes reusing integration templates.
Task
Monitor artificial lift, hydrate risk and compressor condition.
Action coming out
Flags exceptions for engineers and operators. The Montney narrative describes interventions on chemicals and equipment; it does not establish fully autonomous control there.
Control and approval
Human-guided exception response in the named narrative; do not transfer Chevron's separate autonomy description to Montney.
Delivery model
OPX Ai's Integrated Operations Center as a Service, described by its authors in JPT.
Reported reach
Kickoff to full deployment in roughly four months at the described Montney development. Asset-wide well count and exact deployment dates are not supplied.
Reported result
Vendor reports production 3–4% above forecast and approximately 5% lower LOE after four operating months. A 6% uplift is anticipated, not observed.
Why this is worth studying (analyst interpretation)
The case attributes faster implementation partly to newer digital infrastructure and reused configurations. That makes data standardization a concrete condition to test elsewhere.
Still unknown
How was the forecast made? What else affected performance? What are the service fees, how much field work remains, and did the benefits last beyond four months?
Operator question (analyst proposal)
Would the same integration and performance hold on older assets with inconsistent tags and data?
Supplier question (analyst proposal)
Can the supplier package integration and operational response as reliably as the model itself?
A useful test (not yet run)
Compare actual performance against both the original forecast and a comparable untreated cohort; include ongoing service and labor cost.
Claim boundary
Vendor-authored case. The forecast comparison is not an independent controlled estimate of net AI value.
Basis: canonical initiative record and account facts COP-F06.

Original sources and review limits

  1. Field deployments of AI-based IOCaaSFU003 / 2026-04-01 / Kaybob and Montney deployment sections; integration and measurement-period paragraphs
    Research review: Latest access review: original document available; Montney case; architecture, implementation, first-winter results and authorship.. Earlier: Full vendor-authored JPT article rechecked 6 October 2026 Comparison recheck 2026-10-08: Named Montney section and vendor authorship reviewed. Montney implementation, first-winter interventions and four-month forecast comparison; JPT web lines 653-665, author biography 681.
  2. ConocoPhillips 2026 annual meeting transcriptCOP-S02 / 2026-05-14 / CEO Ryan Lance answer on enterprise AI strategy and governance
    Research review: Reviewed 2026-10-07. Strategy-only statement; no named implementation or measured result.
AI-215 / Field work / Operator reported

Occidental

AI-assisted prioritization of lease-operator visits in routeless operations

Data going in
Field sensors and operational data in a combined automation and AI approach.
Task
Prioritize lease-operator attention in routeless operations.
Action coming out
Directs field attention to exceptions rather than a routine route. Remote centers resolve some issues; field visits still remain part of the operating model.
Control and approval
Dispatch logic, human overrides and the share of actions driven specifically by AI are not disclosed.
Delivery model
Operator describes an integrated operating model, an AI Center of Excellence and an Operations SWAT Team; this does not identify a supplier contract for routeless work.
Reported reach
Approximately 40% of U.S. production under routeless/remote-center coverage in February 2026. Production share is not a well or personnel count.
Reported result
Management reports roughly 300 issues per day resolved remotely during a Rockies winter storm. The result belongs to the combined operating model, not AI alone.
Why this is worth studying (analyst interpretation)
A remote operating organization can provide somewhere for AI alerts to go. Whether the alert improves a specific response still needs measurement.
Still unknown
How often are alerts wrong or problems missed? How quickly do crews respond? How many trips are avoided, how serious are the issues, and how much of the change comes from AI?
Operator question (analyst proposal)
Which problems can be resolved remotely, and which still need a field response?
Supplier question (analyst proposal)
Which exceptions are unresolved by the existing remote centers, and who owns response quality?
A useful test (not yet run)
Track alerts through disposition, resolution time and field visits; separate weather, automation and AI contributions.
Claim boundary
Combined sensors, automation and AI disclosure; no isolated route-level cost or AI performance measure.
Basis: canonical initiative record and account facts OXY-F01, OXY-F07.

Original sources and review limits

  1. Occidental Q2 2025 Earnings Call TranscriptSW2-003 / 2025-08-07 / Page 3, Vicki Hollub prepared remarks
    Research review: Full 17-page company-hosted transcript reviewed.
  2. 2025 Annual ReportOXY-01 / 2025 reporting year; report publication day not established / CEO letter, printed pp. 2–3 / PDF pp. 2–3, lines 41–129
    Research review: Reviewed 2026-10-07. Management-reported priorities and results; AI impact is discussed as a portfolio opportunity, not separately measured.
  3. Occidental Q4 2025 earnings-call transcriptOXY-A1 / 2026-02-19 / COO Richard Jackson Q&A, printed pp. 13–14, lines 489–498
    Research review: Reviewed 2026-10-07. Issuer transcript: ~40% U.S. production routeless/remote-center coverage and ~300 issues/day resolved remotely during Rockies winter storm. AI is in a combined digital/field-sensors/automation approach; remote resolution is not an AI-only outcome. Comparison recheck 2026-10-08: Original PDF downloaded; relevant COO answer reviewed. PDF/printed p. 14: approximately 40% of U.S. production covered and roughly 300 storm issues per day resolved remotely.
AI-248 / Information & commercial work / Vendor reported

Continental Resources

Agent-assisted drilling and completion data validation

Data going in
Inconsistent drilling and completion data arriving by email; WellView is named in award material.
Task
Interpret incoming data, validate context and score accuracy.
Action coming out
UiPath and Databricks agents check data. Robots return errors to suppliers and deliver validated data to engineers.
Control and approval
Vendor says initial Action Center review was later removed. Remaining exception controls and current error monitoring are not described.
Delivery model
Named UiPath agents and robots, a Databricks agent and WellView integration. Separate owner-relations campaign material is not treated as the same deployed workflow.
Reported reach
Undated vendor case reports more than 200 emails per day and one reusable robot supporting more than 40 processes. Those are not 40 AI initiatives.
Reported result
Vendor reports about five minutes saved per previously manual email. Award material reports accuracy rising from roughly 70% to the high 90% range; dates and test method are unspecified.
Why this is worth studying (analyst interpretation)
This is a concrete handoff between suppliers and engineers. The reusable integration may matter as much as the language model; that is an interpretation to test.
Still unknown
When were the results measured, and on what test data? How often were serious errors accepted? Who handles exceptions after human review is removed? Did the time saved reduce costs?
Operator question (analyst proposal)
What is the cost of a wrong accepted record, and how is it caught after human review is removed?
Supplier question (analyst proposal)
Can the agent fit the existing well-data handoff and keep an auditable error trail?
A useful test (not yet run)
Replay a dated, representative set of emails with labeled ground truth; measure critical errors, corrections and full processing time.
Claim boundary
Undated case. Award announcement on September 29, 2025 does not date the operating metrics.
Basis: canonical initiative record and account facts CLR-F05.

Original sources and review limits

  1. Continental Resources Streamlines Operations with Automation: From fragmented emails to orchestrated operations at Continental ResourcesSW7-001 / No displayed publication/update date; accessed 2026-10-01 / Full page opened/read. Workflow and transition from human validation to automated handling at lines 241-255; quoted Sr. IT Manager Justin Cornell at 257-259; results at 260-265; 200+ emails/40+ processes/~5 minutes at 228-237.
    Research review: Full primary vendor case-study page read. Customer contact is named and quoted; performance remains vendor-published/customer-attributed.
  2. Honoring the 2025 UiPath AI25 WinnersSW7-002 / Undated page; references 2025 award cohort; accessed 2026-10-01 / Full page opened/read; Continental listing at lines 294-302, reports WellView and Databricks integration and high-90% accuracy vs ~70%, near-equal manual-intervention reduction.
    Research review: Full page read. No publication/update or metric-measurement date displayed.
  3. UiPath Announces 2025 AI25 Awards WinnersSW7-003 / 2025-09-29 / Full press-release content reviewed from issuer's newsroom/search result. Names Continental among 2025 AI25 award winners. Does not disclose workflow details or performance metrics.
    Research review: Dated primary UiPath press release reviewed. Used solely to date the award announcement, not the operational metrics.
  4. UiPath + Databricks: Data Intelligence Meets Agentic AutomationSW7-004 / Undated page; accessed 2026-10-01 / Continental customer panel: Databricks Genie Agents and UiPath Agents for email processing.
    Research review: Original page reviewed; no date or incremental result.
AI-089 / Equipment settings / Operator reported

EOG Resources

Proprietary machine-learning production optimizer

Data going in
EOG's wider operating systems include internally developed software and real-time data acquisition. Exact optimizer signals are not described.
Task
Proprietary machine-learning production optimization.
Action coming out
Annual report says the optimizer improved base production and runtime. It does not specify the model, the number of wells or the human-control boundary.
Control and approval
Recommendation versus automatic execution is not established.
Delivery model
Proprietary optimizer. The wider internal software estate provides context, not proof of which model or cloud runs this tool.
Reported reach
No AI-specific rollout count. More than 140 in-house applications is a separate software-estate measure; those applications are not all established as AI.
Reported result
No AI-specific numerical outcome. The report's 7% lower average well cost is a wider operating result.
Why this is worth studying (analyst interpretation)
Internal software can offer a familiar place to integrate new capabilities. That does not demonstrate faster deployment, superior models or a financial advantage.
Still unknown
How much is the tool used? What does internal support cost? Which wells are suitable, what can the system change, and how much value comes from the optimizer?
Operator question (analyst proposal)
What does EOG's internal delivery model let it change faster, and at what maintenance cost?
Supplier question (analyst proposal)
Which component complements internal engineering rather than duplicating a tool the company already builds?
A useful test (not yet run)
Measure sustained use, release cycle time and operating outcomes for this tool; evaluate internal and purchased alternatives on the same task.
Claim boundary
The software estate and broader operating gains cannot be counted as AI deployments or attributed AI returns.
Basis: canonical initiative record and account facts EOG-F01, EOG-F05.

Original sources and review limits

  1. Proprietary machine-learning production optimizerOCT5-S034 / 2026 annual report covering 2025; exact publication day not confirmed / PDF page 4, shareholder letter: Consistent Operational Excellence Compounds Over Time
    Research review: full text reviewed Comparison recheck 2026-10-08: Original annual-report PDF downloaded; relevant CEO letter reviewed. PDF p. 4: proprietary genAI, production optimizer, machine learning and separate broader 7% well-cost reduction.
  2. Proprietary generative AI for collaboration and operational dataOCT5-S033 / 2026 annual report covering 2025; exact publication day not confirmed / PDF page 4, shareholder letter: Consistent Operational Excellence Compounds Over Time
    Research review: Full original document reviewed on 6 October 2026 Comparison recheck 2026-10-08: Original annual-report PDF downloaded; relevant CEO letter reviewed. PDF p. 4: proprietary genAI, production optimizer, machine learning and separate broader 7% well-cost reduction.
  3. ML 'production optimizers' and 24-hour control roomR046 / 2026-02-25 / Locator not independently checked
    Research review: Audit access: full text reviewed. Earlier: excerpt reviewed
  4. EOG Resources 2025 Form 10-KEOG-01 / 2026-02-24 / Technology/cyber reliance pp. 19–20; reserve technology pp. F-48–F-50
    Research review: Reviewed 2026-10-07. Discloses broad technology reliance and cyber risks, not named cloud/model vendors; does not attribute reserve values to AI. Comparison recheck 2026-10-08: Relevant original passage available and reviewed. Risk factors, printed p. 26; internally developed software and real-time data acquisition; SEC web lines 748-749.
  5. 2025 Sustainability Management ReportEOG-05 / 2025 reporting period; publication day not established / Printed pp. 4, 6, 14, 32, 35
    Research review: Reviewed 2026-10-07. 140+ is total in-house application portfolio, not AI count. iSense measures continuous leak-detection coverage, not an AI deployment or savings metric.
AI-095 / Information & commercial work / Operator reported

EQT

AI-assisted service rebids and procurement targeting

Data going in
Management describes AI tools that could identify service-rebid opportunities; exact commercial inputs are undisclosed.
Task
Target service rebids and procurement effort.
Action coming out
Points procurement attention toward potential rebids. The call does not describe automated tendering, contract awards or a confirmed live workflow.
Control and approval
Commercial decision support is described as an opportunity; execution and approval boundary remain unknown.
Delivery model
Vendor and model for rebid targeting are not named. EQT's separate digital work environment is an operating foundation, not proof of this tool's deployment.
Reported reach
No rebid-tool deployment count. At year-end 2025, 56% of employees worked remotely; the filing describes a primary platform for communication and critical processes.
Reported result
Prospective low-single-digit procurement cost reduction, not a realized saving or isolated AI effect.
Why this is worth studying (analyst interpretation)
A shared workflow could help distribute decisions and track completion. Salesforce appears in other reviewed EQT sources, but not every communication is established as occurring there.
Still unknown
Is it running? Which spending can it influence? What were prices before? Are the suggested rebids useful and allowed under the contracts? How much was actually saved?
Operator question (analyst proposal)
Does better targeting produce better completed rebids, after market pricing and supplier-quality changes?
Supplier question (analyst proposal)
Which procurement decision is still difficult inside the existing work environment, and is outside software wanted?
A useful test (not yet run)
Track eligible contracts from recommendation through rebid and award; compare like-for-like price and service outcomes rather than gross quoted savings.
Claim boundary
Remote work and digital collaboration are not proof of AI returns. The procurement estimate is prospective.
Basis: canonical initiative record and account facts EQT-F01, EQT-F03, EQT-F05.

Original sources and review limits

  1. EQT Q4 2025 earnings call — AI-assisted procurement targetingEQTD01 / 2026-02-18 / Printed pp. 20–21; Toby Rice, service rebids and procurement
    Research review: full text reviewed
  2. AI tools to target service rebids and procurementR056 / 2026-02-18 / Locator not independently checked
    Research review: Audit access: full text reviewed. Earlier: excerpt reviewed
  3. EQT Corporation 2025 Form 10-KEQT-01 / 2026-02-18 / Strategy and DWE pp. 7–8; human capital/DWE p. 32; cybersecurity/CIO pp. 58–59; CIO bio in executive officers
    Research review: Reviewed 2026-10-07. Company-wide description; does not identify DWE provider in the 10-K or AI tools in the platform. Comparison recheck 2026-10-08: Relevant original passages available and reviewed. Human Capital Resources, printed p. 32; approximately 56% remote and digital work environment; SEC web lines 946-949.
  4. Land Data Analyst II/III/Sr. — current job postingEQT-02 / Current posting; posting date not established / Responsibilities involving Land Data Scientist, cross-functional land teams, Salesforce, Databricks and AI pipelines
    Research review: Reviewed 2026-10-07. Shows current hiring/task requirements, not proof every pipeline is deployed, production adoption, or buying authority.
AI-045 / Equipment settings / Operator reported

Diamondback Energy

Machine learning / AI for artificial-lift choice and late-life well optimization

Data going in
Production operations and late-life wells; exact training data and model inputs are not described.
Task
Select and optimize lift type and cost using machine learning and AI.
Action coming out
The company identifies downtime and late-life performance as intended improvements. It does not disclose how recommendations become equipment changes.
Control and approval
Automation, approval mode and deployed boundaries are unspecified. The code of ethics separately requires vetted AI tools and independent human review.
Delivery model
Qualitative issuer disclosure; no named lift-optimization supplier or platform.
Reported reach
No well count or eligible fleet denominator. The reviewed letters describe early work and expected future gains.
Reported result
No result specifically attributed to this workflow. Expected uptime and LOE improvements should remain separate from reported company-wide results.
Why this is worth studying (analyst interpretation)
A defined late-life lift problem is a useful starting point for investigation. Formal AI rules describe a review constraint, not proof of field adoption.
Still unknown
Is it running today? Which model and lift types does it use? Who owns it and reviews the recommendations? What does it cost, and what has improved because of it?
Operator question (analyst proposal)
Which late-life decision is being changed today, and what has improved so far?
Supplier question (analyst proposal)
What does an approved evaluation require, and who owns the lift workflow as well as the tool review?
A useful test (not yet run)
Establish current scope first, then measure matched late-life wells for downtime, interventions, production and cost.
Claim boundary
Qualitative issuer disclosure and future expectations; no isolated AI result or verified purchasing signal.
Basis: canonical initiative record and account facts FANG-F01, FANG-F02.

Original sources and review limits

  1. Diamondback Energy Letter to Stockholders (Q3 2025)SW7-011 / 2025-11-03 / PDF p. 1, lines 61-64; full 3-page PDF read
    Research review: Original source opened; relevant passage read.
  2. Diamondback Q2 2026 Earnings Call TranscriptRS040 / 2026-08-10 / Existing register citation; not fully re-opened in primary materials
    Research review: Audit access: inaccessible. Earlier: excerpt reviewed
  3. Q2 2026 earnings call transcriptS26 / 2026-08-10 / Chad McAllaster / Kaes Van’t Hof response to Geoff Jay
    Research review: Audit access: full text reviewed. Earlier: Third-party-hosted transcript reviewed
  4. ML and AI in field ops cut downtime post-EndeavorR047 / 2026-05 / Locator not independently checked
    Research review: Audit access: full text reviewed. Earlier: Imported reference; source verification pending
  5. Diamondback Energy Letter to Stockholders (Q1 2026)SW7-012 / 2026-05-04 / lines 72-77, especially line 74; full release read
    Research review: Original source opened; relevant passage read.
  6. 2026 Q2 Letter to Stockholders (filed May 4, 2026)FANG-05 / 2026-05-04 / Q1 operating results and cost discussion
    Research review: Reviewed 2026-10-07. Same management outlook as issuer letter; no AI-specific result or quantified planned savings.
  7. Code of Ethics (2026-v2)FANG-02 / 2026-v2; exact revision day not established / PDF p. 17, Use of Artificial Intelligence Tools
    Research review: Reviewed 2026-10-07. Policy describes required controls, not which tools are currently deployed or which model suppliers are approved.

The unanswered questions that matter

What has stopped, been rolled back or failed its test?

Success stories cannot tell us how often projects fail. If a company publishes no result, that does not mean the project failed. Ask what the test required, what went wrong and whether anything was rolled back or abandoned. Keep a failed technical test separate from a cancelled commercial project.

Who keeps the system working after the pilot?

An internal team, a managed service and a purchased application each leave the company with different work to do. Ask who supports the system, owns the data, maintains the connections and responds when something breaks. Find out what happens if the company wants to stop using it.

Would the result hold on your assets?

Lift type, asset age, sensor quality and the way field teams work can all affect whether a tool works elsewhere. Keep those conditions in view when comparing companies or basins. A reported uplift on one asset does not tell us what another will earn.

Coverage: 283 initiative profiles, 196 company-use records and 460 source records. Profiles overlap; the collection is not a census of every global issuer or undisclosed deployment. Budget owners, current contracts, support burden and isolated net AI returns are frequently unknown.