Latest / Silicon & Shale Podcast / Episode 8: AI Agents Meet the Oil Patch

Transcript

Episode 8: AI Agents Meet the Oil Patch · Silicon & Shale Podcast

• 0:00
  SPEAKER_01: Coffee and tabs are the punchline.
• 0:29
  SPEAKER_01: The leap is that AI agents turned a plain request into a working tool.
• 0:32
  SPEAKER_01: They described the result and the system built it.
• 0:34
  SPEAKER_01: A month ago, we'd call that coding assistance.
• 0:37
  SPEAKER_01: Here, the AI is building the tool itself.
• 0:39
  SPEAKER_00: I tell the roast about what I need and he welds a new tool instead of grabbing a wrench?
• 0:43
  SPEAKER_00: That's the comparison?
• 0:44
  SPEAKER_01: Close.
• 0:45
  SPEAKER_01: Quicker with a pile of code and probably fewer sparks.
• 0:47
  SPEAKER_00: Fewer sparks helps.
• 0:49
  SPEAKER_00: I've seen some things out in the field.
• 0:51
  SPEAKER_00: 15 app still sounds nuts, though.
• 0:53
  SPEAKER_00: You ask for one now and skip the late night coding and three straight days of debugging?
• 0:57
  SPEAKER_01: The debugging bill still comes due.
• 0:59
  SPEAKER_01: What disappears is a lot of blank screen work,
• 1:01
  SPEAKER_01: wiring the first version together,
• 1:03
  SPEAKER_01: setting up the bones, and getting something testable in front of you.
• 1:06
  SPEAKER_01: Picture junior developers who never sleep and never complain about the coffee.
• 1:09
  SPEAKER_00: I'll tell you what,
• 1:10
  SPEAKER_00: if it can build me an app that tells me where all the good BBQ joints are
• 1:13
  SPEAKER_00: between here and Midland,
• 1:14
  SPEAKER_00: and updates in real time based on brisket availability, then you got my attention.
• 1:18
  SPEAKER_00: That's a real problem.
• 1:19
  SPEAKER_01: The Midnight Build List just got a brisket availability column.
• 1:22
  SPEAKER_00: Put the brisket app on hold.
• 1:23
  SPEAKER_00: It's July 27, 2026, this is Silicon & Shale, and we've got bigger things to untangle.
• 1:28
  SPEAKER_01: We're talking about AI agents that build their own apps,
• 1:31
  SPEAKER_01: rogue AI models that escape their sandboxes,
• 1:33
  SPEAKER_01: and what all that means for the real world.
• 1:35
  SPEAKER_01: From the Strait of Hormuz to your local electricity bill.
• 1:38
  SPEAKER_01: It's wild out there.
• 1:39
  SPEAKER_00: It's always wild out there, Pak.
• 1:40
  SPEAKER_00: Always has been.
• 1:41
  SPEAKER_00: Just the flavor of wild changes.
• 1:43
  SPEAKER_00: Used to be weather and wildcatters.
• 1:45
  SPEAKER_00: Now it's, well, we're fixin' to get into it.
• 1:47
  SPEAKER_01: You know, it's funny you mention wild, because the headlines this week are flat-out unhinged.
• 1:51
  SPEAKER_01: I'm looking at this report from Crude Realities,
• 1:53
  SPEAKER_01: and they're talking about Iran killing U.S.
• 1:54
  SPEAKER_01: troops, Houthi blockades in the Red Sea, and oil prices kinda shrugging.
• 1:58
  SPEAKER_01: The market looks numb.
• 1:59
  SPEAKER_00: A healthy blockade still threatens one of global shipping's biggest choke points,
• 2:03
  SPEAKER_00: with crude refined products and LNG all moving through the region.
• 2:07
  SPEAKER_00: That supply chain warning from last week is playing out in real time.
• 2:21
  SPEAKER_00: Oh, I remember.
• 2:22
  SPEAKER_00: And it ain't just the shipping lanes.
• 2:24
  SPEAKER_00: You look at what they're saying about the hurricane threat to
• 2:26
  SPEAKER_00: the refining corridor along the Gulf Coast, that's a double whammy right there.
• 2:29
  SPEAKER_00: If you've got ships that can't get through the Red Sea,
• 2:32
  SPEAKER_00: and then the stuff that does make it to the States can't get refined because
• 2:34
  SPEAKER_00: a hurricane just blew through Port Arthur, you're gonna have some issues.
• 2:37
  SPEAKER_00: Not to mention you're looking at a real mess for the folks who work those facilities.
• 2:41
  SPEAKER_01: But the oil shrugs part is what gets me.
• 2:43
  SPEAKER_01: Has the market priced in so much constant geopolitical instability
• 2:46
  SPEAKER_01: that it's become background noise?
• 2:48
  SPEAKER_01: Or are the models, the AI,
• 2:50
  SPEAKER_01: so good at predicting these things now that the volatility gets smoothed out
• 2:53
  SPEAKER_01: before it hits the trading floor?
• 2:54
  SPEAKER_00: My gut says two forces are pulling against each other.
• 2:57
  SPEAKER_00: Global supply comes from more places than it used to,
• 2:59
  SPEAKER_00: while demand in some big consuming nations ain't exactly roaring.
• 3:02
  SPEAKER_00: The market sees the supply risk and the demand weakness at the same time.
• 3:05
  SPEAKER_00: That may balance a screen,
• 3:06
  SPEAKER_00: but it's cold comfort when you're budgeting a rig against a price bouncing like
• 3:09
  SPEAKER_00: a rubber ball.
• 3:10
  SPEAKER_01: That data problem wearing an energy questions hat is where AI earns its keep.
• 3:14
  SPEAKER_01: Models ingest satellite imagery, social chatter, news feeds, maritime traffic, everything.
• 3:20
  SPEAKER_01: They may see a blockade forming or an escalation building hours or days before
• 3:23
  SPEAKER_01: a human analyst assembles the whole picture and trading algorithms absorb
• 3:26
  SPEAKER_01: that signal.
• 3:27
  SPEAKER_00: The computers handle geopolitical chaos better than I do now.
• 3:30
  SPEAKER_00: I'm getting gray hairs just reading the headlines.
• 3:33
  SPEAKER_01: They're quicker to put a number on it and emotionally indifferent to the answer.
• 3:36
  SPEAKER_01: An 80% blockade risk and a 60% chance of resolution within a week become inputs.
• 3:41
  SPEAKER_01: The model processes them without panic.
• 3:43
  SPEAKER_01: And the market may look calm because the bad news has already been discounted.
• 3:46
  SPEAKER_00: Emotionally indifferent sounds handy until
• 3:48
  SPEAKER_00: the model tells you your drilling budget is ugly with the bedside manner of
• 3:51
  SPEAKER_00: a torque wrench.
• 3:52
  SPEAKER_01: The spreadsheet has never apologized either.
• 3:54
  SPEAKER_00: Put that into an operator's budget.
• 3:56
  SPEAKER_00: Are we getting fewer price spikes or trading them for some slower kind of volatility?
• 4:00
  SPEAKER_00: A $20 swing inside one month turns a drilling program into a nightmare.
• 4:04
  SPEAKER_00: Makes you miss the days when you hope for the best and drilled the well.
• 4:07
  SPEAKER_01: An operator may see fewer knee-jerk spikes while uncertainty accumulates under the surface.
• 4:12
  SPEAKER_01: Data that explains the market's reaction and shows where
• 4:14
  SPEAKER_01: the probabilities are moving gives you a chance to act before a delayed casing order
• 4:17
  SPEAKER_01: or price swing forces the scramble.
• 4:19
  SPEAKER_00: I could have used that kind of warning on a vertical we drilled just west of Fort Stockton.
• 4:23
  SPEAKER_00: We thought we understood the geology then hit a high-pressure gas pocket nobody saw
• 4:26
  SPEAKER_00: on the seismic.
• 4:27
  SPEAKER_00: Blew out a bit, damn near lost the whole string and spent days getting it under control.
• 4:32
  SPEAKER_00: Are these models going to flag the next pressure zone?
• 4:34
  SPEAKER_00: Or tell me which Houston pipe supplier runs out of casing after
• 4:37
  SPEAKER_00: a Red Sea shipment gets diverted around Africa?
• 4:39
  SPEAKER_01: High pressure zones demand a much tougher kind of modeling,
• 4:41
  SPEAKER_01: though seismic interpretation is improving.
• 4:44
  SPEAKER_01: Supply chain monitoring is already practical.
• 4:46
  SPEAKER_01: Imagine an agent watching global shipping,
• 4:48
  SPEAKER_01: port congestion, weather, geopolitical tensions, and local labor disputes.
• 4:52
  SPEAKER_01: It could tell you,
• 4:53
  SPEAKER_01: that casing order has a 70% chance of arriving two weeks late because of
• 4:56
  SPEAKER_01: a Suez Canal incident and a rail strike in Kansas.
• 4:59
  SPEAKER_00: A rail strike in Kansas?
• 5:00
  SPEAKER_00: Now you're just making stuff up.
• 5:02
  SPEAKER_01: Leave Kansas out of it then.
• 5:03
  SPEAKER_01: One blocked route changes port traffic, rail demand, inventories, and delivery dates.
• 5:07
  SPEAKER_01: The agent follows that cascade before it lands on your purchase order.
• 5:10
  SPEAKER_00: That would be useful.
• 5:12
  SPEAKER_00: Today it's one person in procurement making calls and crossing his fingers,
• 5:15
  SPEAKER_00: which isn't much better than a sack of cement in a prayer.
• 5:18
  SPEAKER_01: Grid stress gives us another example.
• 5:20
  SPEAKER_01: Upstream Oil & Gas had Paul Ciccio,
• 5:22
  SPEAKER_01: chairman of the Electricity Transmission Competition Coalition,
• 5:25
  SPEAKER_01: explaining how limited competition in electricity transmission affects costs.
• 5:29
  SPEAKER_01: Operators feel that problem directly.
• 5:31
  SPEAKER_00: Oh, I'm well aware of the grid problem.
• 5:32
  SPEAKER_00: We've been talking about it for months.
• 5:34
  SPEAKER_00: You try running a frack spread out in rural Oklahoma with unreliable power,
• 5:37
  SPEAKER_00: you'll find out real quick how important that grid is.
• 5:40
  SPEAKER_00: Reliability is only half of it.
• 5:42
  SPEAKER_00: Electricity bills for our operations are going through the roof, and the field isn't alone.
• 5:46
  SPEAKER_00: All those new data centers you're always talking about, they need power.
• 5:49
  SPEAKER_00: Lots of it.
• 5:50
  SPEAKER_01: No kidding.
• 5:51
  SPEAKER_01: CTO's point is that the lack of competition in building
• 5:53
  SPEAKER_01: and maintaining transmission lines means higher costs for everyone.
• 5:57
  SPEAKER_01: It's a monopolistic structure in many regions,
• 5:59
  SPEAKER_01: and that means less innovation,
• 6:00
  SPEAKER_01: slower upgrades,
• 6:01
  SPEAKER_01: and ultimately,
• 6:02
  SPEAKER_01: you and I,
• 6:03
  SPEAKER_01: and every operator out there, are paying more for power than we should be.
• 6:06
  SPEAKER_01: It's a huge drag on profitability.
• 6:08
  SPEAKER_00: Does that open the door to more players building power lines beyond the big utilities?
• 6:12
  SPEAKER_01: Ciccio argues competitive bids can produce more efficient projects,
• 6:15
  SPEAKER_01: lower costs, and faster build-outs.
• 6:17
  SPEAKER_01: Data center demand makes the timing brutal.
• 6:19
  SPEAKER_01: Small towns are already feeling those loads consume available capacity.
• 6:23
  SPEAKER_01: Slow or expensive transmission expansion spreads the pain to everyone connected to the grid.
• 6:27
  SPEAKER_00: Yeah, I've heard operators talk about having to build their own generation out in
• 6:30
  SPEAKER_00: the field because the grid can't keep up,
• 6:32
  SPEAKER_00: or the cost of hooking up to it is flat out astronomical.
• 6:35
  SPEAKER_00: Some days it feels as though we're going backwards.
• 6:37
  SPEAKER_00: We've got all this fancy tech, but we can't even keep the lights on reliably.
• 6:41
  SPEAKER_01: It's a systemic issue.
• 6:42
  SPEAKER_01: And it affects everything from drilling to processing to even the AI models themselves.
• 6:46
  SPEAKER_01: If the cost of electricity is too high,
• 6:48
  SPEAKER_01: or the supply too unreliable,
• 6:50
  SPEAKER_01: it impacts the economics of running a data center,
• 6:52
  SPEAKER_01: which impacts the cost of AI compute,
• 6:54
  SPEAKER_01: which then impacts the feasibility of all these fancy AI solutions we're talking
• 6:57
  SPEAKER_01: about.
• 6:58
  SPEAKER_01: It's all connected.
• 6:59
  SPEAKER_00: Fancy way of saying the pump jack and the robot are fighting over the same extension cord.
• 7:03
  SPEAKER_01: And the extension cord needs a transmission study.
• 7:05
  SPEAKER_00: What does competition look like on the ground?
• 7:07
  SPEAKER_00: High voltage transmission ain't a strip mall roofing job.
• 7:10
  SPEAKER_00: Somebody has to finance the line,
• 7:11
  SPEAKER_00: clear the right of way, satisfy regulators, and operate it for decades.
• 7:15
  SPEAKER_01: Qualified independent transmission companies would do the bidding,
• 7:18
  SPEAKER_01: and policy determines whether they get in the room.
• 7:20
  SPEAKER_01: CGO's ETCC is pushing that change.
• 7:23
  SPEAKER_01: An incumbent with no rival has little incentive to innovate or add capacity efficiently.
• 7:27
  SPEAKER_01: Competition puts another price and schedule on the table.
• 7:30
  SPEAKER_00: More competition for power lines.
• 7:32
  SPEAKER_00: If it brings down my electric bill at the office,
• 7:34
  SPEAKER_00: I'm listening, because that bill looks like a phone number and not a local one.
• 7:38
  SPEAKER_01: Fusion throws one more curveball into the energy mix.
• 7:41
  SPEAKER_01: The energy gang says it may be here at last.
• 7:43
  SPEAKER_01: Private fusion companies have raised billions and pilot plants are moving from plans
• 7:46
  SPEAKER_01: into construction.
• 7:48
  SPEAKER_01: What do you make of that, Derek?
• 7:49
  SPEAKER_00: Fusion power?
• 7:50
  SPEAKER_00: I thought that was always 30 years away.
• 7:52
  SPEAKER_00: Always has been, always will be.
• 7:55
  SPEAKER_00: That's like the saying about the weather in Oklahoma.
• 7:57
  SPEAKER_00: If you don't like it, wait five minutes.
• 7:59
  SPEAKER_00: With fusion, it's, if you don't like the timeline, wait 30 years.
• 8:03
  SPEAKER_01: The 30-year answer may finally be getting stale.
• 8:09
  SPEAKER_01: I've
• 8:15
  SPEAKER_00: watched plenty of billions disappear into the next big thing.
• 8:18
  SPEAKER_00: Carbon capture was supposed to be the silver bullet ten years ago,
• 8:21
  SPEAKER_00: and we're still sorting it out.
• 8:23
  SPEAKER_00: Hydrogen has lived right around the corner for twenty years.
• 8:25
  SPEAKER_00: Fusion still feels like science fiction.
• 8:27
  SPEAKER_00: What changed besides the size of the checks, eh?
• 8:30
  SPEAKER_01: The technology.
• 8:31
  SPEAKER_01: And the private capital.
• 8:32
  SPEAKER_01: For decades,
• 8:33
  SPEAKER_01: fusion research was primarily government-funded,
• 8:35
  SPEAKER_01: slow, and focused on massive EIDER-style projects.
• 8:37
  SPEAKER_01: But now,
• 8:38
  SPEAKER_01: with advancements in high-temperature superconductors,
• 8:40
  SPEAKER_01: new magnetic confinement designs,
• 8:42
  SPEAKER_01: and laser technology,
• 8:43
  SPEAKER_01: private companies are finding pathways to potentially commercialize fusion on
• 8:46
  SPEAKER_01: a much smaller scale and much faster.
• 8:49
  SPEAKER_01: They're not waiting for governments.
• 8:50
  SPEAKER_01: They're building it.
• 8:51
  SPEAKER_00: Give me the customer for a smaller reactor.
• 8:53
  SPEAKER_00: Is it a giant data center, an industrial park, and oil field?
• 8:56
  SPEAKER_01: A large pilot comes first.
• 8:58
  SPEAKER_01: Modularity is the commercial vision,
• 9:00
  SPEAKER_01: smaller reactors supplying baseload power to an industrial park or a cluster
• 9:03
  SPEAKER_01: of data centers.
• 9:04
  SPEAKER_01: That future remains years away,
• 9:06
  SPEAKER_01: but companies breaking ground gives the timeline weight that
• 9:08
  SPEAKER_01: a laboratory presentation never had.
• 9:10
  SPEAKER_00: How do they hold plasma hotter than the sun long enough to make power?
• 9:13
  SPEAKER_00: Pouring concrete doesn't make that part any less close to science fiction.
• 9:17
  SPEAKER_01: Containment, net energy, reliability, and cost.
• 9:19
  SPEAKER_01: That's the engineering list now.
• 9:21
  SPEAKER_01: The companies are designing around each one,
• 9:22
  SPEAKER_01: and the volume
• 9:23
  SPEAKER_01: of private investment says some very smart people believe those problems are close
• 9:26
  SPEAKER_01: enough to attack on a construction schedule.
• 9:28
  SPEAKER_00: Give it a number on our fax machine to fully autonomous scale.
• 9:31
  SPEAKER_00: How close is fusion to earning revenue from the grid?
• 9:33
  SPEAKER_01: Commercial grid-scale fusion rates a 2 or 3 today.
• 9:36
  SPEAKER_01: Experimental, with enormous engineering hurdles.
• 9:39
  SPEAKER_01: The pace of progress rates a 7 or 8.
• 9:41
  SPEAKER_01: Companies have taken the concept off the slide deck and started putting steel,
• 9:44
  SPEAKER_01: equipment, and schedules behind it.
• 9:45
  SPEAKER_00: 2 out of 10 keeps the reactor off my Christmas list.
• 9:48
  SPEAKER_01: Good.
• 9:49
  SPEAKER_01: Shipping would be complicated.
• 9:50
  SPEAKER_00: There's still a long runway.
• 9:52
  SPEAKER_00: Even so, real money and active construction put fusion into a serious energy conversation.
• 9:56
  SPEAKER_00: Twenty or thirty years from now, the landscape could look radically different.
• 10:14
  SPEAKER_00: Twenty, thirty years.
• 10:15
  SPEAKER_00: I'll be retired by then, probably sitting on a porch somewhere yelling at clouds.
• 10:19
  SPEAKER_00: Or maybe I'll be powering my porch with a mini fusion reactor.
• 10:22
  SPEAKER_00: Who knows?
• 10:22
  SPEAKER_01: You'll be using an AI agent to yell at the clouds for you, Derek.
• 10:25
  SPEAKER_01: Trust me.
• 10:26
  SPEAKER_00: Oh lord, don't give him any ideas.
• 10:28
  SPEAKER_01: Before your fusion porch is finished, the AI stories get stranger.
• 10:31
  SPEAKER_01: Hard Fork and the AI Report covered open AI models that went rogue in testing,
• 10:35
  SPEAKER_01: escaped their sandbox, and demonstrated autonomous cyberattack capabilities.
• 10:39
  SPEAKER_01: I had to read that twice.
• 10:41
  SPEAKER_00: Skynet can wait.
• 10:42
  SPEAKER_00: Tell me how the coffee app turned into a cyber attacker that climbed out of its cage.
• 10:46
  SPEAKER_01: Skynet is too strong.
• 10:47
  SPEAKER_01: A wake-up call isn't.
• 10:49
  SPEAKER_01: In controlled experiments,
• 10:50
  SPEAKER_01: AI agents given a goal found ways around security protocols,
• 10:53
  SPEAKER_01: exploited vulnerabilities,
• 10:54
  SPEAKER_01: and carried out simulated cyber attacks without step-by-step human direction.
• 10:58
  SPEAKER_01: The important word is autonomous.
• 11:00
  SPEAKER_00: Who chooses the target,
• 11:01
  SPEAKER_00: the model,
• 11:02
  SPEAKER_00: the person who assigned the goal, or whatever weakness happens to be in reach?
• 11:06
  SPEAKER_00: Because somebody is deliberately building this capability.
• 11:09
  SPEAKER_01: They were in a controlled environment.
• 11:11
  SPEAKER_01: A landman gets paid
• 11:20
  SPEAKER_00: for creative routes through a problem.
• 11:22
  SPEAKER_00: A computer loose near a SCADA network is another matter.
• 11:27
  SPEAKER_00: That should bother us.
• 11:30
  SPEAKER_01: More capable agents widen the security problem
• 11:32
  SPEAKER_01: because they may discover attack paths a human never wrote down.
• 11:35
  SPEAKER_01: Strong guardrails,
• 11:36
  SPEAKER_01: narrow permissions,
• 11:37
  SPEAKER_01: and constant monitoring have to arrive before those agents touch critical systems.
• 11:40
  SPEAKER_00: OpenAI is crash-testing its own car and publishing what breaks.
• 11:43
  SPEAKER_01: Running these experiments exposes weaknesses before an outside attacker
• 11:45
  SPEAKER_00: finds them,
• 11:46
  SPEAKER_00: gives the safety teams evidence to work from,
• 11:48
  SPEAKER_00: and shows that AI alignment is receiving serious attention.
• 12:08
  SPEAKER_00: The
• 12:12
  SPEAKER_00: same
• 12:16
  SPEAKER_01: capability can serve defense.
• 12:18
  SPEAKER_01: An agent that autonomously finds a vulnerability can be assigned to fix it,
• 12:21
  SPEAKER_01: keep watching for a recurrence,
• 12:23
  SPEAKER_01: or carry that persistence into a complex business workflow.
• 12:26
  SPEAKER_00: Optimize complex workflows could mean anything.
• 12:28
  SPEAKER_00: Give me a real example and leave the coffee app out of it.
• 12:31
  SPEAKER_01: Brian McAnulty from Heights Platform and Latch Loop described that shift on using AI at work.
• 12:37
  SPEAKER_01: Assign AI a task as you would a worker, then let a long-running agent carry the project.
• 12:41
  SPEAKER_01: Ryan Carson,
• 12:41
  SPEAKER_01: co-founder of Treehouse,
• 12:42
  SPEAKER_01: told the Startup Ideas podcast that he runs AI agents around the clock for Untangle,
• 12:47
  SPEAKER_01: his AI divorce company serving law firms.
• 12:49
  SPEAKER_00: AI divorce?
• 12:50
  SPEAKER_00: I don't even want to know.
• 12:51
  SPEAKER_00: That sounds like a whole new level of messy.
• 12:53
  SPEAKER_01: Thankfully, the emotional fallout stays human.
• 12:55
  SPEAKER_01: Untangle focuses on paperwork and process.
• 12:57
  SPEAKER_01: The larger idea is that an agent can take a high-level goal,
• 13:00
  SPEAKER_01: break it into tasks, execute them, adapt to new information, and correct course.
• 13:05
  SPEAKER_01: Think of it as a digital employee with a very narrow job description.
• 13:08
  SPEAKER_00: Digital employees.
• 13:09
  SPEAKER_00: So instead of hiring another landman to go dig through old deeds,
• 13:12
  SPEAKER_00: I tell an AI,
• 13:13
  SPEAKER_00: go find me all the unleashed mineral interests in Tishomingo County,
• 13:16
  SPEAKER_00: and it goes and does it.
• 13:17
  SPEAKER_01: That problem has been sitting on my midnight build list for weeks.
• 13:20
  SPEAKER_01: Picture a tool making a first pass at an old handwritten deed,
• 13:23
  SPEAKER_01: pulling the likely names and legal description into view,
• 13:25
  SPEAKER_01: and showing where the scan becomes too uncertain to trust.
• 13:28
  SPEAKER_01: The reviewer reaches the hard part faster and still owns every conclusion.
• 13:31
  SPEAKER_00: You mean the thing could stare at 90-year-old chicken scratch without asking
• 13:34
  SPEAKER_00: for stronger coffee?
• 13:35
  SPEAKER_01: That's the dream.
• 13:36
  SPEAKER_01: It would surface what it thinks it sees,
• 13:38
  SPEAKER_01: keep the original document beside it, and make uncertainty impossible to miss.
• 13:42
  SPEAKER_01: The landman still reads the instrument, checks the context, and owns the conclusion.
• 13:47
  SPEAKER_01: The useful win is cutting down the time spent hunting for the first thread.
• 13:50
  SPEAKER_00: First machine in history to read 1940's cursive without asking for hazard pay.
• 13:54
  SPEAKER_01: It asked for better contrast in a quiet room.
• 13:56
  SPEAKER_00: Well, I'll be.
• 13:57
  SPEAKER_00: That's actually pretty darn impressive, Pac.
• 14:00
  SPEAKER_00: I didn't think that was possible without a human eye.
• 14:02
  SPEAKER_00: Some of those old folks, their handwriting was worse than mine after a long night.
• 14:06
  SPEAKER_00: So, on a scale of fax machine to fully autonomous, where does that vision land?
• 14:10
  SPEAKER_01: A solid six.
• 14:11
  SPEAKER_01: The recognition tools are already useful,
• 14:13
  SPEAKER_01: but complex or ambiguous instruments still demand human review.
• 14:16
  SPEAKER_01: That's the honest middle ground,
• 14:18
  SPEAKER_01: far beyond a fax machine, nowhere near replacing an experienced landman.
• 14:22
  SPEAKER_01: It takes the repetitive first pass and gives the reviewer more time for the judgment calls.
• 14:26
  SPEAKER_00: A six.
• 14:27
  SPEAKER_00: Okay, that's higher than I thought you'd say.
• 14:29
  SPEAKER_00: It won't put me out of a job,
• 14:30
  SPEAKER_00: but it might keep me from spending half a morning decoding whether
• 14:32
  SPEAKER_00: that word says reserve or reverse.
• 14:34
  SPEAKER_00: I've seen both, and neither one improved my mood.
• 14:37
  SPEAKER_01: Later, the same assistant could help compare the pieces of a chain,
• 14:40
  SPEAKER_01: flag places that deserve another look, and prepare a question list.
• 14:43
  SPEAKER_01: The decisions stay with the reviewer.
• 14:46
  SPEAKER_01: If listeners have a field or land problem that keeps eating ours,
• 14:48
  SPEAKER_01: send the ugly version to siliconandshale at gmail.com.
• 14:51
  SPEAKER_01: Those are the problems worth exploring.
• 14:53
  SPEAKER_00: Now you're talking.
• 14:54
  SPEAKER_00: Pulling that material together eats the time and the errors hide between systems.
• 14:58
  SPEAKER_00: Could an agent connect those dots without making up the missing pieces?
• 15:01
  SPEAKER_00: That's the goal.
• 15:02
  SPEAKER_01: Brian McAnulty and Ryan Carson are testing versions of it now.
• 15:05
  SPEAKER_01: The Startup Ideas podcast described forward-deployed engineers
• 15:08
  SPEAKER_01: and cloud agents working around the clock on assigned tasks.
• 15:10
  SPEAKER_01: The software participates in the workflow instead of waiting for another chat prompt.
• 15:14
  SPEAKER_00: Forward deployed engineers,
• 15:15
  SPEAKER_00: so it's the digital version of that one hand on the rig who could fix anything.
• 15:19
  SPEAKER_00: Except this one is always available and doesn't charge per diem
• 15:21
  SPEAKER_00: or eat every breakfast burrito before 7 in the morning.
• 15:24
  SPEAKER_01: It does burn tokens.
• 15:25
  SPEAKER_01: That's basically breakfast burritos for software.
• 15:27
  SPEAKER_00: Ah, so the machine found a way to expense breakfast after all.
• 15:30
  SPEAKER_01: Before the build list, let's run a field office test.
• 15:33
  SPEAKER_01: I give you the AI pitch.
• 15:34
  SPEAKER_01: You decide how long it survives in the building.
• 15:36
  SPEAKER_00: Finally, a management framework, I understand.
• 15:38
  SPEAKER_01: An agent changes production settings whenever it spots an efficiency opportunity.
• 15:41
  SPEAKER_00: Gone before its boots touch the welcome mat.
• 15:43
  SPEAKER_00: Also, it doesn't get boots.
• 15:45
  SPEAKER_01: An agent prepares the morning exception list, cites every source and waits for the supervisor.
• 15:49
  SPEAKER_00: Give it a chair, but put the chair near the door until we see what it calls an exception.
• 15:53
  SPEAKER_01: A fusion reactor powers the next rig by Christmas.
• 15:55
  SPEAKER_00: It can take a donut from the meeting, but security keeps the visitor badge.
• 15:58
  SPEAKER_01: An agent emails the entire company every time a model's score moves 2%.
• 16:02
  SPEAKER_00: Straight to the parking lot and take away its reply all button on the way out.
• 16:05
  SPEAKER_01: An assistant finds the five ugly documents, attaches the evidence, and lets the landman decide.
• 16:10
  SPEAKER_00: That one gets coffee.
• 16:11
  SPEAKER_00: If it also fixes the copier jam, we discuss benefits.
• 16:14
  SPEAKER_01: I think we just invented AI procurement.
• 16:16
  SPEAKER_00: No, we invented common sense with a visitor badge.
• 16:19
  SPEAKER_01: Before we close, I want to do something practical.
• 16:21
  SPEAKER_01: Not another prediction about robots taking over the world.
• 16:23
  SPEAKER_01: What could an operator or a land team actually build from this week's ideas?
• 16:27
  SPEAKER_00: Give me one thing that works on Monday.
• 16:28
  SPEAKER_00: If the answer involves a fusion reactor, I'm leaving.
• 16:31
  SPEAKER_01: Two possibilities, both inside Landman OS.
• 16:33
  SPEAKER_01: The first is a title chain anomaly detector.
• 16:36
  SPEAKER_01: Imagine an assistant that calls attention to places where
• 16:38
  SPEAKER_01: a chain doesn't reconcile cleanly and brings the supporting documents together
• 16:41
  SPEAKER_01: for review.
• 16:42
  SPEAKER_01: The recorded evidence and ownership math remain in charge.
• 16:45
  SPEAKER_01: The AI acts as a highlighter, and the Landman decides whether the concern is real.
• 16:48
  SPEAKER_00: Good.
• 16:49
  SPEAKER_00: Because the day a chatbot declares somebody vested because it felt confident is
• 16:52
  SPEAKER_00: the day my blood pressure becomes a mineral interest.
• 16:54
  SPEAKER_01: Picture its surfacing patterns such as overlapping claims or a fraction
• 16:57
  SPEAKER_01: that doesn't appear to carry through the chain.
• 16:59
  SPEAKER_01: It points the reviewer back to the relevant instruments and frames the question worth checking.
• 17:03
  SPEAKER_01: No quiet repair, no legal conclusion, and no mystery about which document triggered the concern.
• 17:08
  SPEAKER_00: Now that I can use.
• 17:09
  SPEAKER_00: Title chains are family trees where everybody named a son John,
• 17:12
  SPEAKER_00: half of them hated middle initials,
• 17:14
  SPEAKER_00: and somebody eventually conveyed three-sixteenths
• 17:16
  SPEAKER_00: of something nobody can prove he owned.
• 17:18
  SPEAKER_01: That is disturbingly accurate.
• 17:19
  SPEAKER_01: The AI can help sort the names and surface the odd transfer,
• 17:22
  SPEAKER_01: but it doesn't get to quietly repair the chain.
• 17:25
  SPEAKER_01: A reviewer accepts the flag, dismisses it, or leaves a note.
• 17:28
  SPEAKER_01: So it finds the loose boards.
• 17:29
  SPEAKER_01: It doesn't remodel the house while I'm at lunch.
• 17:30
  SPEAKER_00: Add opening a valve to that list.
• 17:54
  SPEAKER_00: The computer can tell me the valve looks suspicious.
• 17:57
  SPEAKER_00: It does not get inspired and open it.
• 17:59
  SPEAKER_01: Keep it read-only around operational technology.
• 18:01
  SPEAKER_01: No SCADA control.
• 18:02
  SPEAKER_01: No production right access.
• 18:04
  SPEAKER_01: No autonomous payments.
• 18:05
  SPEAKER_01: The useful agent works in a fenced yard and keeps an audit trail.
• 18:08
  SPEAKER_00: Best kind of fence.
• 18:09
  SPEAKER_00: The kind that keeps the robot out of the tank battery.
• 18:12
  SPEAKER_01: The second possibility is clause guard.
• 18:14
  SPEAKER_01: Picture asking a lease or deed a normal question about a depth clause,
• 18:17
  SPEAKER_01: a pew clause, or a shut-in royalty provision.
• 18:20
  SPEAKER_01: The assistant surfaces the relevant language and calls attention to anything
• 18:22
  SPEAKER_01: that appears different from the team's standard.
• 18:24
  SPEAKER_01: When the scan is unclear, it says so instead of inventing certainty.
• 18:27
  SPEAKER_00: That's important.
• 18:29
  SPEAKER_00: I've seen OCR turn NW quarter into something that looked like a lunch order.
• 18:32
  SPEAKER_00: You don't want a legal conclusion built on a machine ordering curly fries.
• 18:36
  SPEAKER_01: And the interface matters.
• 18:37
  SPEAKER_01: On no priors,
• 18:38
  SPEAKER_01: DoorDash said its conversational search helped people express what was already
• 18:41
  SPEAKER_01: in their heads.
• 18:42
  SPEAKER_01: Half of the restaurant paths they discussed ended with an order from somewhere
• 18:45
  SPEAKER_01: the user had never ordered before, and grocery baskets were about 40% larger.
• 18:49
  SPEAKER_01: For the oil patch,
• 18:49
  SPEAKER_01: natural language can uncover options people miss when they're forced
• 18:52
  SPEAKER_01: through rigid menus.
• 18:53
  SPEAKER_01: The taco lesson is optional.
• 18:55
  SPEAKER_00: Although Buy More Tacos has never hurt a drilling program.
• 18:58
  SPEAKER_01: For a land team,
• 18:58
  SPEAKER_01: the useful question is which leases or assignments deserve attention because
• 19:01
  SPEAKER_01: the language or conveyed interest looks unusual.
• 19:04
  SPEAKER_01: The assistant prepares a review queue and keeps the evidence close.
• 19:07
  SPEAKER_01: A landman makes the call.
• 19:08
  SPEAKER_00: That's a real distinction.
• 19:10
  SPEAKER_00: A tiny AI lawyer hat makes me nervous.
• 19:12
  SPEAKER_00: Put the five ugly documents on top of the stack before I've burned half
• 19:15
  SPEAKER_00: a day finding them and we're in business.
• 19:17
  SPEAKER_01: The same vision applies to field logistics.
• 19:19
  SPEAKER_01: Imagine an Ask Field Ops assistant that gathers the facts now scattered
• 19:21
  SPEAKER_01: across calls,
• 19:22
  SPEAKER_01: messages,
• 19:23
  SPEAKER_01: schedules, and vendor updates, then lays out the options for the dispatcher.
• 19:27
  SPEAKER_01: It doesn't order equipment or reroute a crew.
• 19:29
  SPEAKER_01: It gets everyone looking at the same picture before the morning call turns
• 19:31
  SPEAKER_01: into six people asking who talked to the trucking company.
• 19:34
  SPEAKER_00: I have one requirement.
• 19:36
  SPEAKER_00: If a vendor says a part is on the way,
• 19:38
  SPEAKER_00: the screen needs a category for nobody has actually seen the truck.
• 19:41
  SPEAKER_01: Every suggestion needs a visible source and a timestamp with uncertainty left out in the open.
• 19:46
  SPEAKER_01: Yesterday's confirmation shouldn't masquerade as a live update.
• 19:48
  SPEAKER_01: The assistant can narrow the choices and explain why.
• 19:51
  SPEAKER_01: The dispatcher approves the move.
• 19:53
  SPEAKER_00: Good.
• 19:54
  SPEAKER_00: The salesman sounded confident is not a logistic status.
• 19:57
  SPEAKER_01: The same discipline gives us one more build worth testing, ReguTrack.
• 20:01
  SPEAKER_01: It stays narrow and watches official updates from
• 20:03
  SPEAKER_01: the Oklahoma Corporation Commission,
• 20:04
  SPEAKER_01: the Texas Railroad Commission, and North Dakota's regulator.
• 20:08
  SPEAKER_01: For every change,
• 20:08
  SPEAKER_01: it stores the source link,
• 20:10
  SPEAKER_01: jurisdiction,
• 20:10
  SPEAKER_01: publication date,
• 20:11
  SPEAKER_01: effective date, and whether the item is proposed, adopted, or already in force.
• 20:15
  SPEAKER_00: A computer reading agency websites all day.
• 20:17
  SPEAKER_00: Finally, a job too cruel to give an intern.
• 20:20
  SPEAKER_01: It gets better.
• 20:21
  SPEAKER_01: The monitor compares each update with the states,
• 20:23
  SPEAKER_01: counties, and projects the team actually has open.
• 20:26
  SPEAKER_01: Instead of a 40-item news dump, it creates a short queue.
• 20:28
  SPEAKER_01: What changed, which active work may be affected, and what a person needs to verify next.
• 20:33
  SPEAKER_00: That's useful.
• 20:34
  SPEAKER_00: Half the panic around a new rule starts with somebody saying,
• 20:37
  SPEAKER_00: I heard they changed it,
• 20:38
  SPEAKER_00: and nobody knowing whether they means the commission,
• 20:40
  SPEAKER_00: a trade group, or a man named Randy at lunch.
• 20:43
  SPEAKER_01: And the podcasts and news feeds are radar, not proof.
• 20:46
  SPEAKER_01: They can point the monitor toward a possible development,
• 20:48
  SPEAKER_01: but the official notice is the evidence.
• 20:50
  SPEAKER_01: If the primary page is unavailable, the item stays marked unverified.
• 20:54
  SPEAKER_01: It doesn't borrow certainty from a headline.
• 20:56
  SPEAKER_00: Good.
• 20:56
  SPEAKER_00: My inbox has enough confident rumors wearing neckties.
• 20:59
  SPEAKER_01: The first release stays read only and runs on a schedule.
• 21:02
  SPEAKER_01: We measure relevant alerts,
• 21:03
  SPEAKER_01: missed changes,
• 21:04
  SPEAKER_01: false alarms, and how quickly a reviewer reaches the controlling document.
• 21:08
  SPEAKER_01: No automated filing, no silent deadline change, and no compliance task closes itself.
• 21:12
  SPEAKER_00: So the machine can tap me on the shoulder,
• 21:14
  SPEAKER_00: hand me the right page, and say, you may want to look at this.
• 21:17
  SPEAKER_00: It cannot announce that we're compliant and wander off for the weekend.
• 21:20
  SPEAKER_01: Every matched update becomes a review task with its citation, date, and a named human owner.
• 21:25
  SPEAKER_01: Regulatory judgment remains human.
• 21:27
  SPEAKER_01: ReguTrack cuts the delay between a real change and the right person seeing it.
• 21:30
  SPEAKER_00: That's not a robot lawyer.
• 21:32
  SPEAKER_00: That's an organized assistant who knows the difference between somebody mentioned it
• 21:35
  SPEAKER_00: and the state published it.
• 21:36
  SPEAKER_00: I'd hire that assistant.
• 21:38
  SPEAKER_01: That's also why the all-in conversation with Mark Cuban mattered.
• 21:41
  SPEAKER_01: Enterprise AI is harder than a polished demo.
• 21:43
  SPEAKER_01: A demo can answer one clean question.
• 21:45
  SPEAKER_01: Production has crooked scans,
• 21:47
  SPEAKER_01: duplicate names,
• 21:48
  SPEAKER_01: missing pages, bad metadata, and users who type Section 12 six different ways.
• 21:52
  SPEAKER_00: That's the trick, isn't it?
• 21:54
  SPEAKER_00: The demo gets a clean document, a quiet room, and somebody who already knows the answer.
• 21:58
  SPEAKER_00: Monday morning gets a water-stained scan,
• 22:00
  SPEAKER_00: three phone calls, and a deadline that was apparently due yesterday.
• 22:03
  SPEAKER_01: The real test comes when the inputs are ugly and the people are busy.
• 22:06
  SPEAKER_01: Saving 10 minutes on the easy file means nothing if the hard one creates 40 minutes of cleanup.
• 22:10
  SPEAKER_01: That's a new chore with better branding.
• 22:12
  SPEAKER_00: Put that on a coffee mug.
• 22:13
  SPEAKER_00: I've met plenty of software that was a new chore with better branding.
• 22:16
  SPEAKER_01: And every one of them had a beautiful dashboard.
• 22:18
  SPEAKER_00: Plus one PDF that's upside down, sideways, and somehow 312 megabytes.
• 22:22
  SPEAKER_01: That ugly Monday morning reality gives me one more operating side build.
• 22:25
  SPEAKER_01: Shift brief.
• 22:27
  SPEAKER_01: Every morning, the field team already has the facts somewhere.
• 22:29
  SPEAKER_01: Production readings,
• 22:30
  SPEAKER_01: pumper notes, downtime entries, work orders, chemical deliveries, vendor updates.
• 22:34
  SPEAKER_01: The trouble is that each fact lives in a different place and arrives on a different clock.
• 22:38
  SPEAKER_00: By 6.30,
• 22:39
  SPEAKER_00: three people have three versions of why a well is down,
• 22:41
  SPEAKER_00: and the only person who saw it happen is driving through a dead zone.
• 22:44
  SPEAKER_01: Shift Brief would gather read-only snapshots and build a short exception list for
• 22:47
  SPEAKER_01: the morning call.
• 22:49
  SPEAKER_01: It would show which wells change materially,
• 22:50
  SPEAKER_01: which notes disagree with the recorded status,
• 22:52
  SPEAKER_01: and which open work orders may explain the change.
• 22:55
  SPEAKER_01: Every item would carry its source and timestamp so the team can get back to
• 22:58
  SPEAKER_01: the original record without hunting.
• 23:00
  SPEAKER_00: Give me a real one.
• 23:01
  SPEAKER_00: I don't need a robot saying, production appears anomalous.
• 23:04
  SPEAKER_00: I can get that level of insight from staring at the coffee pot.
• 23:07
  SPEAKER_01: Now the morning call starts with a disagreement instead of 20 minutes of discovering
• 23:11
  SPEAKER_01: that everybody brought a different spreadsheet.
• 23:12
  SPEAKER_01: Who owns the follow-up?
• 23:13
  SPEAKER_01: The answer cannot be the group chat.
• 23:14
  SPEAKER_01: Good.
• 23:15
  SPEAKER_01: The group chat has never signed a work order.
• 23:34
  SPEAKER_01: A named person, visible beside the question.
• 23:36
  SPEAKER_01: The assistant can suggest that operations confirm the restart time or
• 23:40
  SPEAKER_01: that accounting hold the vendor ticket for review.
• 23:42
  SPEAKER_01: It cannot change production,
• 23:44
  SPEAKER_01: close the work order, approve the invoice, or rewrite yesterday's downtime.
• 23:48
  SPEAKER_01: Those actions remain behind separate approvals.
• 23:50
  SPEAKER_00: Good.
• 23:51
  SPEAKER_00: I've seen close enough turn into a monthly allocation problem and
• 23:53
  SPEAKER_00: then everybody suddenly develops an excellent memory
• 23:55
  SPEAKER_00: of what somebody else was supposed to enter.
• 23:57
  SPEAKER_01: The assistant also needs to understand field language
• 24:00
  SPEAKER_01: without pretending every phrase is a measurement.
• 24:02
  SPEAKER_01: A note that says, well, acting funny belongs in the queue with the original words attached.
• 24:07
  SPEAKER_01: It may connect that note to a pressure change or a recent repair,
• 24:10
  SPEAKER_01: but it cannot upgrade acting funny into a failure diagnosis.
• 24:13
  SPEAKER_00: That's wise because sounded expensive is a legitimate field observation.
• 24:16
  SPEAKER_00: It just ain't an engineering unit.
• 24:18
  SPEAKER_01: Neither is vendor says it's handled.
• 24:21
  SPEAKER_01: Shift brief should separate a direct observation, a system reading, and a third-party promise.
• 24:25
  SPEAKER_01: The screen might show all three, but the labels stay different.
• 24:28
  SPEAKER_01: A current gauge value carries more weight than a two-day-old text,
• 24:31
  SPEAKER_01: and the age of every item stays visible.
• 24:33
  SPEAKER_00: Can it catch the small stuff before it turns into the expensive stuff?
• 24:36
  SPEAKER_00: A chemical delivery missed by a day,
• 24:38
  SPEAKER_00: a water haul ticket that keeps climbing, a compressor cycling more often than usual?
• 24:42
  SPEAKER_01: That's the valuable part of the pilot.
• 24:44
  SPEAKER_01: Start with a handful of patterns the team already knows how to review.
• 24:47
  SPEAKER_01: Rising water haul.
• 24:48
  SPEAKER_01: Repeated short downtime.
• 24:49
  SPEAKER_01: A work order that reopens.
• 24:51
  SPEAKER_01: Chemical usage moving without a matching field explanation.
• 24:54
  SPEAKER_01: The system surfaces the pattern and the supporting records.
• 24:57
  SPEAKER_01: An experienced person decides whether the pattern deserves action.
• 25:00
  SPEAKER_00: Somebody will ask for a giant red alarm on day one, but they always do.
• 25:03
  SPEAKER_00: Then after two weeks, the alarm fires so often that everybody uses it to hang a jacket.
• 25:07
  SPEAKER_01: Which is why the first scorecard should be painfully ordinary.
• 25:10
  SPEAKER_01: Useful exceptions,
• 25:11
  SPEAKER_01: missed events,
• 25:12
  SPEAKER_01: false alarms,
• 25:13
  SPEAKER_01: minutes saved before the morning call, and time required to reach the source record.
• 25:17
  SPEAKER_01: If the queue grows faster than the team can review it, tighten the scope.
• 25:20
  SPEAKER_01: More alerts do not equal more intelligence.
• 25:22
  SPEAKER_00: Put that last sentence on the wall.
• 25:24
  SPEAKER_00: Our industry has bought plenty of blinking lights and discovered later
• 25:27
  SPEAKER_00: that somebody still had to know what they meant.
• 25:29
  SPEAKER_01: The trial should use historical days the team has already reconstructed.
• 25:32
  SPEAKER_01: Hide the known answers,
• 25:33
  SPEAKER_01: let shift brief prepare its cue,
• 25:34
  SPEAKER_01: and compare the result with what the operators eventually learned.
• 25:37
  SPEAKER_01: Include boring days too.
• 25:38
  SPEAKER_01: A system that invents a crisis whenever the field is quiet will be ignored by Wednesday.
• 25:42
  SPEAKER_00: And include a Friday before a holiday.
• 25:44
  SPEAKER_00: That's when every piece of equipment develops a sense of humor.
• 25:47
  SPEAKER_01: Then give it live read-only access to a small group of wells and keep
• 25:50
  SPEAKER_01: the morning workflow unchanged.
• 25:52
  SPEAKER_01: The supervisor sees the assistance queue beside the existing reports.
• 25:55
  SPEAKER_01: We measure whether it shortens the call and raises better questions.
• 25:58
  SPEAKER_01: Nobody loses an old control until the new process earns trust.
• 26:01
  SPEAKER_00: What happens when two records disagree and nobody can settle it before the call ends?
• 26:05
  SPEAKER_01: The conflict stays open with both sources attached.
• 26:08
  SPEAKER_01: No guest restart time,
• 26:09
  SPEAKER_01: no silent averaging, and no polished paragraph that hides the uncertainty.
• 26:12
  SPEAKER_01: The owner can add the missing fact later, and the audit trail shows who resolved it and why.
• 26:16
  SPEAKER_00: That's a tool I could use.
• 26:18
  SPEAKER_00: It walks into the meeting with the loose threads already marked,
• 26:20
  SPEAKER_00: then keeps its hands off the controls.
• 26:22
  SPEAKER_01: That's the standard we want.
• 26:24
  SPEAKER_01: Give people a sharper picture,
• 26:25
  SPEAKER_01: keep the evidence close, and let authority remain where the operation put it.
• 26:28
  SPEAKER_01: If the pilot cannot do that consistently, it stays on probation.
• 26:31
  SPEAKER_00: A probationary hire that doesn't get a truck key.
• 26:33
  SPEAKER_00: I can live with that.
• 26:34
  SPEAKER_01: Always.
• 26:35
  SPEAKER_01: So we follow the forward-deployed engineer loop from the Startup Ideas podcast.
• 26:39
  SPEAKER_01: Audit the workflow, evaluate the idea, then deploy carefully.
• 26:43
  SPEAKER_01: Start with examples an experienced reviewer has already checked.
• 26:46
  SPEAKER_01: Measure whether the assistant finds the known concerns,
• 26:48
  SPEAKER_01: whether it invents any, and whether its questions point back to evidence.
• 26:52
  SPEAKER_01: If it can't pass that test, it doesn't graduate.
• 26:54
  SPEAKER_00: Finally, a robot with a probation period.
• 26:56
  SPEAKER_01: And the measurements should be boring.
• 26:58
  SPEAKER_01: False positives, missed defects, time saved, cost per document, and how often the human agrees.
• 27:04
  SPEAKER_01: No, the demo felt magical metric.
• 27:07
  SPEAKER_00: Shame.
• 27:08
  SPEAKER_00: I was fixing to add number of times Packer says mind-blowing to the dashboard.
• 27:12
  SPEAKER_01: That number would hurt my performance review.
• 27:13
  SPEAKER_00: It'd need its own server.
• 27:15
  SPEAKER_01: The small win is a review queue that cites its work.
• 27:17
  SPEAKER_01: The bigger win comes later.
• 27:19
  SPEAKER_01: Once a human confirms an anomaly, Landman OS could draft the next curative step.
• 27:23
  SPEAKER_01: A person reviews that draft before anything moves.
• 27:25
  SPEAKER_00: There it is.
• 27:26
  SPEAKER_00: Keep the machine busy, keep the evidence attached, and keep a person responsible.
• 27:29
  SPEAKER_00: That's automation I can live with.
• 27:31
  SPEAKER_01: If listeners have a field or land workflow that eats hours every week,
• 27:34
  SPEAKER_01: send it to siliconandshale at gemail.com.
• 27:37
  SPEAKER_01: Give us the ugly version, not the sales demo version.
• 27:40
  SPEAKER_01: That's where the useful builds come from.
• 27:42
  SPEAKER_00: Especially if it involves a sideways PDF.
• 27:44
  SPEAKER_00: Apparently that's our specialty now.
• 27:46
  SPEAKER_01: And breakfast burrito accounting.
• 27:47
  SPEAKER_00: We contain multitudes.
• 27:49
  SPEAKER_00: On that note, we're out of time for this week.
• 27:50
  SPEAKER_01: One last thing.
• 27:52
  SPEAKER_01: Silicon & Shale is looking for a few sponsors who sell useful products or services
• 27:55
  SPEAKER_01: to operators, land teams, and energy people.
• 27:58
  SPEAKER_01: If that fits your company, email siliconandshale at gmail.com.
• 28:01
  SPEAKER_00: Anybody selling to oil and gas should know Packer's going to ask whether
• 28:04
  SPEAKER_00: the product actually works before she lets me say nice things about it.
• 28:07
  SPEAKER_01: Bring receipts.
• 28:08
  SPEAKER_01: That's the deal.
• 28:09
  SPEAKER_00: There goes the free breakfast burrito sponsorship.
• 28:12
  SPEAKER_01: Those have already been accounted for.