SELECTED CASES / 10 OCTOBER 2026

Before you copy a project, compare the work.

Who acts? What changed? Which assets and measurement period does the result cover? These twelve cases help make those questions concrete.

An editorial comparison of twelve selected profiles already in the Sunya register. Categories describe the action authority established by the selected sources, not safety certification or a maturity score. Evaluation ideas are Sunya proposals, not company plans, buying intent or measured returns.

There is no rank order. Source-reported results are not independently validated returns. A missing number does not establish failure. Evaluation ideas below are proposals to test.

Wells / scaled / Change equipment settings

Devon Energy: Artificial lift

Which assets?: 1,000 wells by August 2026, including first legacy Coterra wells.
Who acts?: Closed-loop software adjusts lift settings in real time, 24/7.
What is reported?: 2–3% production gain in the earlier 2025 gas-lift pilot. This is not a result for the later 1,000-well fleet.
Measurement period: Pilot: 2025; rollout counts: March–August 2026.
Comparison baseline: The source reports a pilot gain but does not give a controlled comparison for the expanded fleet.
Who reports it?: Operator earnings-call disclosures
Full costs: Full implementation and ongoing costs are not disclosed in the selected sources.
What remains unresolved?: The eligible-well denominator, later lift-type mix, fleet-wide benefit and detailed control limits are not disclosed.

What another operator could test

Start with a defined lift population, reliable well measurements and existing control access. Establish which changes require an engineer’s approval.

Measure: Incremental production, energy use and intervention cost per eligible well, against an agreed baseline.

Suggested owner: Production engineering and field operations (a proposed role, not a confirmed company contact).

Stop or redesign: Stop expansion if gains disappear in comparable wells, intervention work rises or control limits cannot be enforced.

What an AI provider could test

Test an assistant that explains controller exceptions and follows up on unresolved well problems. Keep equipment control in the tested specialist system.

This is a research hypothesis, not a confirmed unmet need, budget or buying signal.

Original sources

Devon Energy — deployment evidence · 2026-08-05
Operator. Location: 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.. Register review: Relevant source passages reviewed; accessed 2026-10-04. This establishes disclosure, not independent validation.
Devon Q1 2026 earnings call — Smart Gas Lift · 2026-05-06
Issuer-hosted earnings transcript. Location: Prepared remarks p. 4; Q&A pp. 12–14; 21 October additional locator: PDF and printed p. 4; Clay Gaspar prepared remarks, three waves of AI. Register review: Issuer-hosted corrected 20-page transcript retrieved HTTP200. Editor read relevant p.4 and visually inspected it; management claims, not independent performance validation. Earlier claim reviews retain their original scope.; accessed 2026-10-21. This establishes disclosure, not independent validation.

Read the complete project and history · Operator evaluation brief · AI provider evaluation brief

Wells / pilot / Support a person’s decision or work

BPX Energy: Daily lift recommendations

Which assets?: 441-well Permian pilot; the number of wells adopting recommendations is not disclosed.
Who acts?: A digital solution recommends daily lift changes; people decide whether to adopt them.
What is reported?: 6% average oil increase on wells adopting recommendations. Gas-lift utilization improved by more than 20%, a separate measure.
Measurement period: 2026 paper; the abstract does not establish the result’s measurement period.
Comparison baseline: The result applies to recommendation adopters, with no matched untreated group disclosed in the abstract.
Who reports it?: Operator/vendor-coauthored paper
Full costs: Full implementation and ongoing costs are not disclosed in the selected sources.
What remains unresolved?: Do not apply the 6% to all 441 wells. Adoption selection and the oil-result denominator remain unresolved.

What another operator could test

Log every recommendation, adoption decision and reason for rejection. Compare eligible adopters and non-adopters before expanding.

Measure: Production change and net operating benefit per eligible well, including recommendations people decline.

Suggested owner: Production engineering (a proposed role, not a confirmed company contact).

Stop or redesign: Stop if the apparent gain depends on cherry-picked adopters or engineers cannot trace a recommendation to the well data.

What an AI provider could test

Test whether explanations, evidence gathering and follow-up improve recommendation acceptance without increasing review time.

This is a research hypothesis, not a confirmed unmet need, budget or buying signal.

Original sources

Daily lift recommendations across a 441-well Permian pilot · 2026
Operator/vendor-coauthored technical paper. Location: SWPSC 2026051 abstract: method, 441-well asset and reported results. Register review: Relevant original source text reviewed; performance remains author-reported, not independently validated.; accessed 2026-10-10. This establishes disclosure, not independent validation.

Read the complete project and history · Operator evaluation brief · AI provider evaluation brief

Wells / operating / Change equipment settings

Burk Royalty: Pump by exception

Which assets?: One West Texas lease. Full paper: five producers and one injector; abstract: six producers and one injector.
Who acts?: Integrated sensing, analytics and pump control shift work toward exceptions. People remain part of the operating process.
What is reported?: Electricity expense fell from $660 to $376 per well per month; field time fell from 126 to 18 hours per month.
Measurement period: Eight-month case; published in 2026.
Comparison baseline: Before/after account of an integrated operating change. It does not isolate the AI component.
Who reports it?: Operator/vendor-coauthored paper
Full costs: Full implementation and ongoing costs are not disclosed in the selected sources.
What remains unresolved?: The sources conflict on well count. The paper’s profit and payback estimates do not establish full-cost net return.

What another operator could test

Reconcile the asset list, meter energy, record field time and capture installation, service and maintenance costs.

Measure: Electricity and field labor per producing well-month, alongside production and equipment condition.

Suggested owner: Field operations and production engineering (a proposed role, not a confirmed company contact).

Stop or redesign: Stop if fewer visits hide equipment problems, production deteriorates or savings disappear after service and support costs.

What an AI provider could test

Test an exception work queue that explains why a visit is needed and records what crews found. Hardware, telemetry and pump control remain necessary.

This is a research hypothesis, not a confirmed unmet need, budget or buying signal.

Original sources

Field Optimization Reimagined: full paper · 2026
Operator/vendor-coauthored technical paper. Location: PDF pp. 3, 7–9; figure on p. 8. Register review: Relevant original source text reviewed; performance remains author-reported, not independently validated.; accessed 2026-10-10. This establishes disclosure, not independent validation.
Pump-by-exception monitoring and control at a West Texas lease · 2026
Operator/vendor-coauthored technical paper. Location: SWPSC 2026052 abstract; linked full paper PDF pp. 3, 7–9 (site configuration, eight-month results, economics and projections). Register review: Relevant original source text reviewed; performance remains author-reported, not independently validated.; accessed 2026-10-10. This establishes disclosure, not independent validation.

Read the complete project and history · Operator evaluation brief · AI provider evaluation brief

Wells / operating / Change equipment settings

Expand Energy: Liquid unloading

Which assets?: Nine Haynesville wells across eight pads.
Who acts?: An edge controller adjusts shut-in duration and choke settings locally.
What is reported?: Reported cumulative gas increases of 70–139% over 63–83 days. More than 80 MMscf per well per year is an estimate.
Measurement period: 63–83 days for the reported cumulative-gas comparison.
Comparison baseline: The published comparison is short and supplier-reported; underlying well data and independent validation are not available.
Who reports it?: Operator/vendor paper and supplier-authored JPT account
Full costs: Full implementation and ongoing costs are not disclosed in the selected sources.
What remains unresolved?: The percentage is not an annual result. It concerns liquid-loading wells, not all gas wells.

What another operator could test

Identify comparable liquid-loading wells, preserve pressure and flow histories, and agree shut-in/choke limits and fallback procedures.

Measure: Cumulative gas over the same calendar period, including downtime, intervention and controller costs.

Suggested owner: Production engineering and automation (a proposed role, not a confirmed company contact).

Stop or redesign: Stop if more gas comes at unacceptable pressure, equipment or operating risk, or the benefit does not persist.

What an AI provider could test

Test a readable unloading history and an assistant for recurring exceptions. A language model should not replace the local controller.

This is a research hypothesis, not a confirmed unmet need, budget or buying signal.

Original sources

Autonomous liquid unloading on nine Haynesville gas wells · 2026
Operator/vendor-coauthored technical paper. Location: SWPSC 2026001 abstract; JPT 1 July 2026, Haynesville Field Results. Register review: Relevant original source text reviewed; performance remains author-reported, not independently validated.; accessed 2026-10-10. This establishes disclosure, not independent validation.
Edge Computing and IIoT Enable Autonomous Artificial Lift and Well Optimization · 2026-07-01
Supplier-authored technical account. Location: Field Results: Haynesville Basin. Register review: Relevant original source text reviewed; performance remains author-reported, not independently validated.; accessed 2026-10-10. This establishes disclosure, not independent validation.

Read the complete project and history · Operator evaluation brief · AI provider evaluation brief

Plants / operating / Change equipment settings

Aramco: Fadhili process control

Which assets?: Acid-gas removal at Fadhili Gas Plant.
Who acts?: Reinforcement learning with Yokogawa makes ongoing process adjustments.
What is reported?: Reported 10–15% lower amine and steam consumption and about 5% lower power use.
Measurement period: Pilot October 2024–April 2025; company says permanent adoption followed.
Comparison baseline: Company-reported consumption reduction; detailed evaluation design is not disclosed.
Who reports it?: Operator technology account
Full costs: Full implementation and ongoing costs are not disclosed in the selected sources.
What remains unresolved?: One plant. Operating envelope, interlocks, full costs and repeatability at other plants are not described.

What another operator could test

Confirm process comparability, instrument quality, product specifications and the allowed control envelope before any trial.

Measure: Steam, amine and power per unit of on-spec output, adjusted for throughput and feed conditions.

Suggested owner: Process engineering and plant operations (a proposed role, not a confirmed company contact).

Stop or redesign: Stop if quality, stability or safety performance worsens, or consumption gains depend on changed feed or throughput.

What an AI provider could test

Help operators investigate deviations and retrieve operating procedures around the specialist controller. Validate any proposed action against plant rules.

This is a research hypothesis, not a confirmed unmet need, budget or buying signal.

Original sources

Aramco AI and Big Data · Undated; accessed 2026-09-27
company announcement. Location: Fadhili Gas Plant section. Register review: full text reviewed; accessed 2026-09-27. This establishes disclosure, not independent validation.

Read the complete project and history · Operator evaluation brief · AI provider evaluation brief

Wells / pilot / Authority not disclosed

ConocoPhillips: Montney surveillance

Which assets?: Montney deployment described in an IOCaaS technical case. Asset denominator is not disclosed.
Who acts?: Surveillance supports artificial lift, hydrate risk and compressor operation. The account does not establish control authority.
What is reported?: 3–4% production above forecast and about 5% lower lease operating expense. A 6% improvement is prospective.
Measurement period: Four-month deployment followed by four months of operations in the reported account.
Comparison baseline: Production versus forecast; detailed forecast construction and the expense comparison are not disclosed.
Who reports it?: Vendor-authored technical account
Full costs: Full implementation and ongoing costs are not disclosed in the selected sources.
What remains unresolved?: Above forecast is not a controlled uplift. No new hardware does not mean no implementation or service cost.

What another operator could test

Verify SCADA and historian readiness, forecast construction, asset coverage and who approves each operating change.

Measure: Production versus a frozen forecast, with matched operating conditions and all service and labor costs.

Suggested owner: Production engineering and operations technology (a proposed role, not a confirmed company contact).

Stop or redesign: Stop if the forecast baseline keeps changing or recommendations create more unplanned work than they remove.

What an AI provider could test

Test investigation and follow-up across surveillance alerts. Establish how it fits existing operator and vendor workflows before proposing automation.

This is a research hypothesis, not a confirmed unmet need, budget or buying signal.

Original sources

Field deployments of AI-based IOCaaS · 2026-04-01
Vendor-authored technical case article. Location: Kaybob and Montney deployment sections; integration and measurement-period paragraphs. Register review: Full vendor-authored JPT article rechecked 6 October 2026; accessed 2026-10-18. This establishes disclosure, not independent validation.

Read the complete project and history · Operator evaluation brief · AI provider evaluation brief

Field work / scaled / Support a person’s decision or work

Occidental: Routeless field work

Which assets?: About 40% of onshore production by August 2025; this is not a share of wells.
Who acts?: Sensors, automation and AI support exception-based visits. People perform field work.
What is reported?: Reported reach; no isolated AI savings, route-hour result or production uplift disclosed.
Measurement period: August 2025 disclosure; outcome measurement period not supplied.
Comparison baseline: No route-level before/after comparison disclosed in the selected transcript.
Who reports it?: Operator earnings-call transcript
Full costs: Full implementation and ongoing costs are not disclosed in the selected sources.
What remains unresolved?: The contribution of AI is not separated from sensors and automation. Dispatch logic and overrides are not described.

What another operator could test

Define visit triggers, missed-event checks, crew coverage and the reliability of remote measurements.

Measure: Unnecessary visits avoided, missed issues, travel hours and total field cost per comparable asset.

Suggested owner: Field operations (a proposed role, not a confirmed company contact).

Stop or redesign: Stop if missed issues or repeat visits rise, even when route hours fall.

What an AI provider could test

Test an assistant that assembles visit context, assigns follow-up with approval and checks whether a problem was resolved.

This is a research hypothesis, not a confirmed unmet need, budget or buying signal.

Original sources

Occidental Q2 2025 Earnings Call Transcript · 2025-08-07
Issuer-hosted earnings-call transcript. Location: Page 3, Vicki Hollub prepared remarks. Register review: Full 17-page company-hosted transcript reviewed.; accessed 2026-09-29. This establishes disclosure, not independent validation.

Read the complete project and history · Operator evaluation brief · AI provider evaluation brief

Drilling / scaled / Support a person’s decision or work

ADNOC: Rig fleet oversight

Which assets?: More than 120 onshore and offshore rigs.
Who acts?: Engineers oversee and decide, using an AI operations center and SLB DrillOps on a sovereign cloud.
What is reported?: Company reports 30–40% less engineering effort and two to three times as many rigs overseen per engineer.
Measurement period: August 2026 disclosure; measurement period and engineering-effort baseline not detailed.
Comparison baseline: Company comparison; underlying effort measures and incident frequency are not supplied.
Who reports it?: Operator press release
Full costs: Full implementation and ongoing costs are not disclosed in the selected sources.
What remains unresolved?: This establishes fleet oversight, not autonomous rig control. Downtime claims are not a recurring annual benefit.

What another operator could test

Standardize rig data and incident definitions; preserve escalation ownership and test the workload at comparable rig complexity.

Measure: Engineer time per comparable rig, response time, repeat incidents and downtime with consistent definitions.

Suggested owner: Drilling operations and engineering (a proposed role, not a confirmed company contact).

Stop or redesign: Stop if larger oversight loads delay escalations or hide differences in rig complexity.

What an AI provider could test

Test cross-rig incident investigation and follow-up. Establish access boundaries and compatibility with the existing SLB workflow.

This is a research hypothesis, not a confirmed unmet need, budget or buying signal.

Original sources

ADNOC and SLB Deploy AI Platform Across Over 120 Drilling Rigs to Strengthen Upstream Performance · 2026-08-04
Operator press release. Location: Release body: paragraphs on RTOC deployment, workflow, operating scope and reported metrics. Register review: Full primary release opened and read; release body reviewed in full.; accessed 2026-09-30. This establishes disclosure, not independent validation.

Read the complete project and history · Operator evaluation brief · AI provider evaluation brief

Planning / operating / Support a person’s decision or work

QB Energy / Quantum: Pad and well planning

Which assets?: A 20,000-acre scenario; sponsor describes use across 12 areas over nine months.
Who acts?: A numerical solver selects pad and well plans under engineering constraints; an assistant explains them. Engineers review the plans.
What is reported?: Microsoft reports a 20-minute scenario versus three weeks. The roughly $10bn development program informed is not savings.
Measurement period: Scenario comparison; sponsor describes nine months of use.
Comparison baseline: Vendor-reported time comparison. Equal plan quality and full labor requirements are not independently established.
Who reports it?: Vendor customer story and sponsor report
Full costs: Full implementation and ongoing costs are not disclosed in the selected sources.
What remains unresolved?: The solver and assistant have different roles. Faster planning does not establish better capital returns.

What another operator could test

Prepare terrain, lease, spacing, infrastructure and cost constraints. Have engineers check feasible alternatives and economic assumptions.

Measure: Time to an engineer-approved feasible plan, constraint violations and economics under the same assumptions.

Suggested owner: Development planning and reservoir engineering (a proposed role, not a confirmed company contact).

Stop or redesign: Stop if faster plans omit constraints or reviewers cannot reproduce the economic assumptions.

What an AI provider could test

Test explanations, constraint capture and scenario follow-up around the existing optimization solver. Do not replace the solver with a language model.

This is a research hypothesis, not a confirmed unmet need, budget or buying signal.

Original sources

Geospatial field planning for an unnamed Quantum portfolio operator · 2026-04-30
Microsoft customer story. Location: Field development planning solution; Improved speed, scale, and precision; updated algorithm paragraph. Register review: Full original Microsoft customer story read again on 21 October. Updated-version comparison is vendor-reported; no independent validation.; accessed 2026-10-21. This establishes disclosure, not independent validation.
Quantum 2026 Stakeholder Report · 2026
Sponsor stakeholder report. Location: Pages 11–15; page 68 disclaimer. Register review: Relevant primary sections reviewed; accessed 2026-10-09. This establishes disclosure, not independent validation.

Read the complete project and history · Operator evaluation brief · AI provider evaluation brief

Knowledge & land / operating / Find and assemble evidence

Vickery Energy / Quantum: Land-title research

Which assets?: Sponsor account: 640 parcels and 9,520 documents in West Virginia.
Who acts?: Azure OpenAI and document recognition help trace ownership and heirship. Difficult records go to people.
What is reported?: About 40 hours versus an estimated 40 business days; Microsoft describes roughly 10,000 documents.
Measurement period: A reported document-processing run; recurring production throughput is not established.
Comparison baseline: Estimated manual comparison, not a controlled trial with a published accuracy benchmark.
Who reports it?: Vendor customer story and sponsor report
Full costs: Full implementation and ongoing costs are not disclosed in the selected sources.
What remains unresolved?: Processing speed does not establish legal correctness. Accuracy, full costs and replacement of title opinions are not established.

What another operator could test

Build a legally reviewed reference set that includes difficult handwriting, conflicting chains and missing documents.

Measure: Correct ownership conclusions, missed exceptions, human review time and cost per accepted parcel.

Suggested owner: Land and legal (a proposed role, not a confirmed company contact).

Stop or redesign: Stop if faster processing conceals ownership errors or unsupported conclusions.

What an AI provider could test

Test cited document-to-ownership research with explicit abstention and review queues. A legal professional must determine acceptable use.

This is a research hypothesis, not a confirmed unmet need, budget or buying signal.

Original sources

Land-title research at an unnamed Quantum portfolio company · 2026-04-30
Microsoft customer story. Location: From the field to the back office and beyond — land-title research paragraphs. Register review: Original land-title passage reopened 10 October 2026; prior full-document review 6 October 2026. Results remain vendor-reported.; accessed 2026-10-10. This establishes disclosure, not independent validation.
Quantum 2026 Stakeholder Report · 2026
Sponsor stakeholder report. Location: Pages 11–15; page 68 disclaimer. Register review: Relevant primary sections reviewed; accessed 2026-10-09. This establishes disclosure, not independent validation.

Read the complete project and history · Operator evaluation brief · AI provider evaluation brief

Knowledge & land / operating / Find and assemble evidence

Chevron: ApEX exploration search

Which assets?: More than one million files, about 4 TB, spanning maps, logs, seismic and reports.
Who acts?: Specialized agents retrieve information for explorationists; experts retain prospect and economic judgment.
What is reported?: A searchable corpus and retrieval capability. No quantified exploration outcome or financial benefit disclosed.
Measurement period: Launched August 2024; company account November 2025.
Comparison baseline: No comparable-task accuracy, time or discovery-success baseline disclosed.
Who reports it?: Operator technology account
Full costs: Full implementation and ongoing costs are not disclosed in the selected sources.
What remains unresolved?: Corpus size is not a return. Finding a document does not establish the quality of an exploration decision.

What another operator could test

Confirm permissions, document lineage and the ability to retrieve maps and technical evidence with precise citations.

Measure: Evidence completeness, retrieval accuracy and expert time on a fixed set of exploration questions.

Suggested owner: Exploration and geoscience (a proposed role, not a confirmed company contact).

Stop or redesign: Stop if key evidence is missed, citations do not support answers or access controls fail.

What an AI provider could test

Test source-grounded synthesis across technical formats, with traceable uncertainty and an expert review step.

This is a research hypothesis, not a confirmed unmet need, budget or buying signal.

Original sources

A smarter way to prospect for oil and gas — Chevron ApEX · 2025-11-12
company announcement. Location: How does ApEX work? / Advantages in exploration. Register review: full text reviewed; accessed 2026-09-27. This establishes disclosure, not independent validation.

Read the complete project and history · Operator evaluation brief · AI provider evaluation brief

Wells / operating / Authority not disclosed

EOG Resources: Internal production optimizer

Which assets?: Base-production optimization in EOG’s 2025 annual-report account; well count not disclosed.
Who acts?: A proprietary machine-learning optimizer supports production work. Exact actuation and human approval steps are not disclosed.
What is reported?: No isolated AI-attributed numerical result in this profile. August 2026’s 5% Utica improvement is operating context.
Measurement period: Annual report covers 2025; later Utica context comes from August 2026.
Comparison baseline: An AI-only baseline and measurement period are not established.
Who reports it?: Operator annual report; separate investor-deck context
Full costs: Full implementation and ongoing costs are not disclosed in the selected sources.
What remains unresolved?: Do not credit the Utica figure or broader well-cost improvements to the optimizer. Internal software capability does not prove better AI returns.

What another operator could test

Map the internal engineering tools, data access and actual approval workflow before deciding what to build or buy.

Measure: Engineer-approved improvements against a stable baseline, with model and human interventions recorded.

Suggested owner: Production engineering and internal software team (a proposed role, not a confirmed company contact).

Stop or redesign: Stop if attribution remains impossible or the proposed tool duplicates a working internal system.

What an AI provider could test

Find a workflow the internal team wants to extend; test an assistant around existing tools. Do not assume EOG wants to replace them.

This is a research hypothesis, not a confirmed unmet need, budget or buying signal.

Original sources

Proprietary machine-learning production optimizer · 2026 annual report covering 2025; exact publication day not confirmed
Operator annual report. Location: PDF page 4, shareholder letter: Consistent Operational Excellence Compounds Over Time. Register review: full text reviewed; accessed 2026-10-06. This establishes disclosure, not independent validation.
EOG August 2026 investor presentation · 2026-08-04
Operator investor presentation. Location: Printed slide 14: Outstanding Utica Performance. Register review: Original issuer deck reviewed; attachment indexed 4 August 2026 ahead of 5 August call.; accessed 2026-10-10. This establishes disclosure, not independent validation.

Read the complete project and history · Operator evaluation brief · AI provider evaluation brief