Latest / Silicon & Shale Podcast / Episode 1: Bombs, Barrels, and the Bot That Said No to the Pentagon

Transcript

Episode 1: Bombs, Barrels, and the Bot That Said No to the Pentagon · Silicon & Shale Podcast

• 0:00
  SPEAKER_01: So I'm standing in the Reeves County clerk's office,
• 0:03
  and I mean standing because there are no chairs which tells you
• 0:05
  everything about
• 0:06
  SPEAKER_01: how they feel about visitors.
• 0:08
  SPEAKER_00: Wait, Reeves County?
• 0:09
  SPEAKER_00: That's deep Permian, right?
• 0:11
  SPEAKER_01: As deep as it gets.
• 0:13
  SPEAKER_01: And the deputy clerk slides a document across the counter to me and says,
• 0:17
  SPEAKER_01: this is the newest deed we have on this tract.
• 0:20
  SPEAKER_01: And I look at it, and it's from 1987.
• 0:22
  SPEAKER_00: Okay, that's not even that old by courthouse standards, though.
• 0:26
  SPEAKER_01: Packer, it was printed on a dot matrix printer.
• 0:29
  SPEAKER_00: Oh, no.
• 0:30
  SPEAKER_01: The ink had already started to fade,
• 0:32
  SPEAKER_01: and the legal description referenced a survey marker that,
• 0:35
  SPEAKER_01: I later found out,
• 0:37
  was a fence post that got pulled out sometime in the Clinton
• 0:40
  administration.
• 0:41
  SPEAKER_01: Gone.
• 0:42
  SPEAKER_01: Doesn't exist.
• 0:44
  The entire northeast corner of this lease is staked to a hole
• 0:47
  in the ground.
• 0:48
  SPEAKER_00: The fax machine is still warm.
• 0:50
  SPEAKER_01: The fax machine is not just warm,
• 0:52
  SPEAKER_01: the fax machine is actively transmitting,
• 0:55
  and this is a track that a very confident landman had already
• 0:58
  certified title on,
• 1:00
  SPEAKER_01: certified.
• 1:01
  SPEAKER_00: Which means someone is about to have a very bad Thursday.
• 1:04
  SPEAKER_01: Someone already had a very bad Thursday.
• 1:06
  SPEAKER_01: I was that someone.
• 1:08
  I had to call an operator and explain that the legal
• 1:11
  description they'd been
• 1:12
  drilling off of for six years was anchored to a ghost fence
• 1:15
  post in the West Texas
• 1:16
  SPEAKER_01: desert.
• 1:17
  SPEAKER_00: Okay,
• 1:18
  SPEAKER_00: but,
• 1:18
  SPEAKER_00: and I say this with love,
• 1:19
  SPEAKER_00: this is exactly the kind of problem that makes me want to scream,
• 1:23
  SPEAKER_00: okay,
• 1:24
  SPEAKER_00: but what if we just automated that into the void?
• 1:28
  Into the void,
• 1:28
  yes,
• 1:29
  because the void is where Reeves County's digitization budget
• 1:32
  went.
• 1:33
  SPEAKER_01: Welcome to Silicon and Shale,
• 1:34
  SPEAKER_01: 15 minutes where the oil patch meets the algorithm,
• 1:38
  SPEAKER_01: week of March 12th,
• 1:39
  SPEAKER_01: 2026.
• 1:39
  SPEAKER_01: We've got a lot this week.
• 1:43
  SPEAKER_00: The Strait of Hormuz situation keeps developing,
• 1:46
  and it is genuinely wild how differently the oil market is
• 1:49
  reacting versus what
• 1:50
  SPEAKER_00: anyone expected.
• 1:52
  SPEAKER_00: Plus the OpenAI Pentagon deal,
• 1:54
  SPEAKER_00: the anthropic blacklisting,
• 1:56
  and in the collision zone we are going somewhere I have been
• 1:58
  thinking about all
• 1:59
  SPEAKER_00: week,
• 1:59
  SPEAKER_00: and I cannot wait to get into it.
• 2:01
  SPEAKER_01: And Packer has things on her build list.
• 2:03
  SPEAKER_01: She always has things on her build list.
• 2:06
  SPEAKER_00: The build list is a living document, Derek.
• 2:08
  SPEAKER_00: It breathes.
• 2:09
  SPEAKER_01: It absolutely does not need to breathe.
• 2:12
  SPEAKER_01: Let's get into it.
• 2:13
  So the big story that won't quit,
• 2:15
  and we talked about this last week,
• 2:17
  the U.S.
• 2:17
  SPEAKER_01: strikes on Iran, oil barely flinching.
• 2:20
  SPEAKER_01: Well, it has continued to not quit in the most bizarre way possible.
• 2:24
  Because you'd think a live shooting war near the Strait of
• 2:26
  Hormuz would be a
• 2:27
  SPEAKER_00: five-alarm fire for oil markets,
• 2:29
  SPEAKER_00: right?
• 2:30
  SPEAKER_00: That's like the single biggest choke point in global energy supply.
• 2:33
  SPEAKER_00: 20% of the world's oil goes through that strait.
• 2:37
  SPEAKER_01: 20%.
• 2:38
  SPEAKER_01: And tanker traffic drying up,
• 2:39
  SPEAKER_01: shipping chaos,
• 2:40
  SPEAKER_01: the IEA starting to talk about emergency reserves,
• 2:44
  SPEAKER_01: and crude is doing what exactly?
• 2:46
  SPEAKER_00: Spiking and then selling off.
• 2:48
  SPEAKER_00: Wild overnight swings and then nothing sustained.
• 2:52
  SPEAKER_00: Which is insane.
• 2:54
  SPEAKER_01: And the reading I keep coming back to,
• 2:56
  SPEAKER_01: and David Purcell over at Apache has been really articulate about this,
• 3:00
  is that the market has decided OPEC's spare capacity is the
• 3:04
  real story here.
• 3:05
  SPEAKER_01: Like, the question isn't whether Iranian barrels get disrupted.
• 3:09
  The question is whether Saudi Arabia and the rest of them can
• 3:12
  actually cover it.
• 3:13
  SPEAKER_00: And the credibility question on that is massive.
• 3:16
  SPEAKER_00: Because OPEC has been saying they have spare capacity for years,
• 3:20
  SPEAKER_00: but can they actually turn that dial?
• 3:22
  SPEAKER_01: Right.
• 3:23
  SPEAKER_01: And here's my read after 30 years of watching these things.
• 3:27
  SPEAKER_01: Markets price what they believe, not what's true.
• 3:30
  And the market apparently believes that spare capacity is real
• 3:33
  enough to keep a lid
• 3:34
  SPEAKER_01: on prices,
• 3:35
  even with shooting happening near one of the most important
• 3:38
  pieces of water on
• 3:39
  SPEAKER_01: Earth.
• 3:39
  SPEAKER_00: Which is either very sophisticated or completely insane.
• 3:43
  SPEAKER_01: Historically, those two things have a lot of overlap.
• 3:46
  SPEAKER_00: Fair.
• 3:47
  SPEAKER_00: What about the SPR piece?
• 3:49
  SPEAKER_00: Because that story keeps coming up.
• 3:51
  So the IEA is basically treating the Strategic Petroleum
• 3:55
  Reserve and the emergency
• 3:56
  reserves across member countries less like a crisis tool and
• 4:00
  more like a price
• 4:01
  SPEAKER_01: management mechanism,
• 4:03
  which is a significant philosophical shift from what the SPR
• 4:06
  was designed to do.
• 4:08
  SPEAKER_00: It was designed for actual supply disruptions.
• 4:10
  Like,
• 4:11
  the pipeline blows up,
• 4:12
  the tanker sinks,
• 4:13
  you've got a genuine physical shortage.
• 4:16
  SPEAKER_00: Not oil went up $20 and it's inconvenient.
• 4:19
  SPEAKER_01: Exactly.
• 4:20
  And now you've got this situation where it starts to look more
• 4:23
  like a price cap
• 4:24
  SPEAKER_01: than an emergency backstop.
• 4:26
  And that has long-term implications for what anyone thinks the
• 4:29
  SPR actually is.
• 4:30
  Because if you drain it every time prices spike,
• 4:33
  it's not there when you actually need it.
• 4:35
  SPEAKER_01: And there's also the regime change dimension here,
• 4:38
  which remember in episode one I made the comparison to a title
• 4:41
  chain problem.
• 4:42
  SPEAKER_00: You called Iranian regime change a title chain problem,
• 4:45
  and every landman who heard that immediately understood exactly
• 4:48
  what you meant.
• 4:49
  SPEAKER_01: Because it is,
• 4:50
  SPEAKER_01: you've got competing claims of authority,
• 4:52
  SPEAKER_01: unclear chain of custody,
• 4:54
  documentation that may or may not be legitimate depending on
• 4:57
  who's asking,
• 4:58
  and transactions that happened under one regime that the next
• 5:01
  regime may or may not
• 5:02
  SPEAKER_01: honor.
• 5:03
  SPEAKER_01: That's a title problem.
• 5:05
  That's what happens when you drill through a formation and then
• 5:08
  find out the
• 5:09
  surface owner and the mineral owner had a falling out in 1962
• 5:12
  and nobody recorded
• 5:13
  SPEAKER_01: it.
• 5:14
  SPEAKER_00: Except the stakes are slightly higher.
• 5:16
  SPEAKER_01: Slightly.
• 5:17
  SPEAKER_01: And now throw in reservoir damage fears from the military activity,
• 5:22
  SPEAKER_01: which most of the people I've talked to think are mostly noise.
• 5:26
  SPEAKER_01: The subsurface isn't as fragile as the headlines suggest,
• 5:29
  SPEAKER_01: but the uncertainty itself has value as a market signal.
• 5:33
  SPEAKER_00: Okay,
• 5:33
  SPEAKER_00: I want to bring up the zombie wells thing,
• 5:35
  SPEAKER_00: because we touched on it last week and it connects to something bigger.
• 5:38
  SPEAKER_01: Hawk Dunlap,
• 5:40
  SPEAKER_01: Railroad Commission candidate,
• 5:42
  SPEAKER_01: 27 years of well control,
• 5:44
  SPEAKER_01: 100 countries,
• 5:46
  and he showed up talking about excavating 150 leaking wells on
• 5:50
  a single Texas
• 5:51
  SPEAKER_01: ranch.
• 5:51
  SPEAKER_00: 150 on one ranch.
• 5:54
  SPEAKER_01: And the primary was March 3rd,
• 5:56
  SPEAKER_01: so we'll see where that lands,
• 5:57
  but the conversation he forced about the scale of orphaned and
• 6:01
  zombie well
• 6:01
  SPEAKER_01: liability in Texas,
• 6:03
  SPEAKER_01: that's real regardless of who wins.
• 6:05
  SPEAKER_00: Can we just talk about the scope of this problem for a second?
• 6:09
  SPEAKER_00: Because I don't think people who aren't in the industry fully grasp it.
• 6:12
  SPEAKER_00: These are wells that were drilled, produced, and then just... left.
• 6:17
  SPEAKER_00: No proper abandonment.
• 6:19
  SPEAKER_00: No PNA.
• 6:20
  SPEAKER_00: Just a pipe sticking out of the ground in someone's pasture.
• 6:23
  SPEAKER_01: Or not even that.
• 6:25
  Sometimes it's just a depression in the soil and some dead
• 6:28
  grass and a faint smell
• 6:29
  of hydrogen sulfide if you're standing in the wrong spot on a
• 6:32
  warm afternoon.
• 6:34
  SPEAKER_00: And then years later,
• 6:35
  someone wants to drill a water well or run a pipeline or just
• 6:38
  sell the property and
• 6:39
  SPEAKER_00: suddenly this liability surfaces.
• 6:41
  SPEAKER_01: And the Railroad Commission is the backstop on all of this in Texas.
• 6:45
  SPEAKER_01: They've got the Orphan Well Fund,
• 6:47
  SPEAKER_01: they've got the plugging program,
• 6:48
  but the scale of the backlog versus the resources available to
• 6:52
  address it,
• 6:53
  SPEAKER_01: those two numbers are not living in the same universe.
• 6:56
  Okay,
• 6:56
  but what if...
• 6:57
  And I'm going to say this and you're going to make a face.
• 7:00
  SPEAKER_01: I haven't even made the face yet.
• 7:02
  SPEAKER_00: Satellite imagery, LIDAR, soil composition modeling.
• 7:07
  You could potentially identify probable well locations in areas
• 7:10
  where records are incomplete.
• 7:12
  Not definitively,
• 7:13
  but as a triage tool to tell inspectors where to look first.
• 7:17
  SPEAKER_01: I was going to make the face,
• 7:18
  SPEAKER_01: but that's actually...
• 7:19
  SPEAKER_01: I mean,
• 7:20
  SPEAKER_01: you'd need ground truth on top of it.
• 7:22
  SPEAKER_01: You can't condemn a piece of property based on a satellite anomaly.
• 7:26
  SPEAKER_01: Obviously not.
• 7:28
  But if you can narrow 100 suspected locations down to the 12
• 7:31
  that actually need a
• 7:32
  SPEAKER_00: physical inspection,
• 7:33
  SPEAKER_00: you just saved a lot of boots on the ground and a lot of state budget.
• 7:36
  SPEAKER_01: Okay, that I can't argue with.
• 7:39
  SPEAKER_00: All right, the OpenAI Pentagon thing.
• 7:42
  SPEAKER_00: because this is genuinely one of the stranger stories of the week,
• 7:45
  SPEAKER_00: and it connects to stuff we talked about last episode.
• 7:47
  SPEAKER_01: Lay it out.
• 7:48
  SPEAKER_00: So,
• 7:49
  SPEAKER_00: Anthropic,
• 7:50
  SPEAKER_00: who makes Claude,
• 7:51
  which is one of the two or three major AI systems anyone's
• 7:54
  actually using at scale
• 7:55
  SPEAKER_00: right now,
• 7:56
  SPEAKER_00: Anthropic refused to agree to Pentagon terms around two specific things,
• 8:01
  SPEAKER_00: domestic mass surveillance and autonomous weapons.
• 8:05
  SPEAKER_00: They said, no, we won't greenlight our system for those use cases.
• 8:09
  SPEAKER_00: And the response was, the Trump administration blacklisted them.
• 8:14
  SPEAKER_00: Pete Hegseth designated Anthropic a supply chain risk.
• 8:18
  SPEAKER_00: Federal agencies were ordered to stop using Claude.
• 8:21
  SPEAKER_01: That's a heavy word, supply chain risk.
• 8:23
  It's the kind of designation that gets you pulled from
• 8:25
  government contracts across the board.
• 8:28
  SPEAKER_00: It's not a slap on the wrist.
• 8:30
  SPEAKER_01: And then OpenAI steps in.
• 8:32
  SPEAKER_00: Within hours, Sam Altman announces they've got a deal with the Pentagon.
• 8:37
  And the deal apparently includes safety carve-outs for those
• 8:39
  same two issues.
• 8:41
  SPEAKER_00: No domestic mass surveillance, no autonomous weapons.
• 8:45
  SPEAKER_00: Which means OpenAI agreed to basically the same constraints Anthropic was
• 8:48
  SPEAKER_00: blacklisted for demanding.
• 8:50
  SPEAKER_01: So,
• 8:51
  Anthropic got punished for demanding something and OpenAI got
• 8:54
  rewarded for agreeing
• 8:55
  SPEAKER_01: to the same thing.
• 8:56
  SPEAKER_00: That is one reading of it, yes.
• 8:58
  SPEAKER_01: What's the other reading?
• 9:00
  SPEAKER_00: that it's about who controls the conversation.
• 9:02
  SPEAKER_00: Anthropic said no publicly and made it a confrontation.
• 9:06
  SPEAKER_00: OpenAI said yes and got to announce a deal.
• 9:09
  SPEAKER_00: Same outcome.
• 9:10
  SPEAKER_00: Completely different political optics.
• 9:13
  SPEAKER_01: I've been in lease negotiations that worked exactly like that.
• 9:16
  SPEAKER_01: You don't win by being right.
• 9:18
  SPEAKER_01: You win by letting the other person think they won.
• 9:21
  Which is a deeply Derek way to analyze a geopolitical AI story,
• 9:25
  and I respect it completely.
• 9:27
  Here's my question,
• 9:28
  though,
• 9:28
  and this is a genuine question,
• 9:30
  not a gotcha.
• 9:31
  SPEAKER_01: If the Pentagon is using AI in active operations,
• 9:34
  SPEAKER_01: which they apparently are,
• 9:36
  SPEAKER_01: and one of those operations involves Iran,
• 9:39
  and we're in an active military situation near the Strait of
• 9:42
  Hormuz right now,
• 9:43
  SPEAKER_01: what does that actually mean?
• 9:45
  SPEAKER_01: Like, what is AI doing in Operation Epic Fury?
• 9:48
  So the reporting suggests targeting support,
• 9:51
  intelligence synthesis,
• 9:52
  logistics optimization.
• 9:55
  SPEAKER_00: Not a robot pulling a trigger.
• 9:56
  SPEAKER_00: More like,
• 9:58
  imagine having an analyst who can process 50 intelligence feeds
• 10:01
  simultaneously and
• 10:03
  SPEAKER_00: surface the relevant pattern in 30 seconds instead of three hours.
• 10:06
  That's a different thing than what most people think when they
• 10:09
  hear military AI.
• 10:10
  SPEAKER_00: Way different.
• 10:12
  SPEAKER_00: And this is where I think the hype versus reality gap does real damage.
• 10:17
  SPEAKER_00: Because people hear AI in warfare,
• 10:19
  SPEAKER_00: and they picture Terminator,
• 10:20
  SPEAKER_00: and then they either panic or they dismiss it.
• 10:24
  SPEAKER_00: When the actual use case is closer to a really fast,
• 10:27
  SPEAKER_00: really thorough staff officer who never needs sleep...
• 10:30
  SPEAKER_01: Which is not nothing.
• 10:31
  SPEAKER_00: Not nothing at all.
• 10:33
  Fax machine to fully autonomous rating on current military AI
• 10:36
  deployment...
• 10:37
  SPEAKER_01: I'm going to say four.
• 10:39
  SPEAKER_01: It's real, it's meaningful, it's not science fiction.
• 10:43
  But there's a human in the loop on anything that matters and
• 10:45
  the systems are young.
• 10:47
  SPEAKER_00: I'd say five.
• 10:48
  SPEAKER_00: Maybe five and a half.
• 10:50
  The speed advantage in synthesis and targeting support is
• 10:53
  genuinely significant.
• 10:54
  SPEAKER_00: But I'll accept four as the conservative read.
• 10:57
  SPEAKER_01: And the anthropic thing,
• 10:58
  SPEAKER_01: here's what I keep coming back to,
• 11:00
  that is a company that built significant enterprise adoption
• 11:03
  because operators,
• 11:05
  SPEAKER_01: landmen,
• 11:06
  SPEAKER_01: people in our industry started using Claude for document processing,
• 11:10
  SPEAKER_01: for contract review,
• 11:12
  SPEAKER_01: for well file analysis.
• 11:14
  SPEAKER_00: The Claude use cases in oil and gas are real.
• 11:16
  SPEAKER_00: I know people using it for lease abstracting,
• 11:19
  SPEAKER_00: for title opinion drafting support,
• 11:21
  SPEAKER_00: for regulatory compliance checks.
• 11:24
  And now those people are in federal agencies that just got told
• 11:26
  to stop using it.
• 11:28
  SPEAKER_01: What happens to those workflows?
• 11:30
  SPEAKER_00: They either get rebuilt on open AI models or they get a waiver.
• 11:34
  SPEAKER_00: Or they just quietly keep using it anyway and hope nobody checks.
• 11:38
  Which,
• 11:39
  in my experience with government IT policy...
• 11:42
  The fax machine is still warm.
• 11:44
  SPEAKER_00: The fax machine is broadcasting on all frequencies.
• 11:47
  Let me bring up the data management angle because there was a
• 11:49
  conversation this
• 11:50
  SPEAKER_01: week that I thought was underrated.
• 11:53
  SPEAKER_01: The Orteca piece on data quality and metadata in oil and gas.
• 11:57
  Because this is the thing that never gets talked about when
• 12:00
  people talk about AI in
• 12:01
  SPEAKER_01: the industry.
• 12:02
  SPEAKER_00: Garbage in, garbage out.
• 12:04
  SPEAKER_01: You cannot build an intelligent system on top of bad data.
• 12:07
  SPEAKER_01: And bad data is,
• 12:09
  SPEAKER_01: I'm going to be very generous here,
• 12:11
  SPEAKER_01: the majority of what exists in E&P land operations records.
• 12:15
  SPEAKER_00: And the metadata problem is worse than the data problem.
• 12:18
  SPEAKER_00: Because it's not just that the record is wrong,
• 12:20
  SPEAKER_00: it's that you don't know which records to trust,
• 12:23
  SPEAKER_00: in what context,
• 12:24
  SPEAKER_00: with what caveats.
• 12:26
  SPEAKER_01: I have well files from the 1970s where the scout ticket says one thing,
• 12:30
  SPEAKER_01: the completion report says something different,
• 12:32
  SPEAKER_01: and the production data says a third thing.
• 12:36
  SPEAKER_01: All three are official records.
• 12:38
  SPEAKER_01: None of them fully agree.
• 12:39
  And an AI system ingesting all three simultaneously without
• 12:43
  context about which
• 12:45
  source to wait for which question is going to produce
• 12:47
  confidently wrong answers.
• 12:49
  Which in land is how you get a title opinion that says you own
• 12:52
  something you don't own.
• 12:53
  The metadata and data quality work has to come before the AI
• 12:57
  deployment work.
• 12:58
  SPEAKER_00: And nobody wants to fund the boring infrastructure project.
• 13:02
  SPEAKER_00: Everyone wants to fund the AI.
• 13:04
  SPEAKER_01: Everyone wants the rocket ship.
• 13:05
  SPEAKER_01: Nobody wants to build the launch pad.
• 13:07
  SPEAKER_00: That's a good line.
• 13:08
  SPEAKER_00: I'm taking that.
• 13:09
  SPEAKER_01: Okay, collision zone, hit me.
• 13:11
  SPEAKER_00: So here's the thread I've been pulling on all week.
• 13:14
  SPEAKER_00: We've got the zombie well problem,
• 13:17
  SPEAKER_00: massive scale,
• 13:18
  SPEAKER_00: incomplete records,
• 13:19
  physical infrastructure liability scattered across Texas and
• 13:22
  honestly every
• 13:23
  SPEAKER_00: producing state.
• 13:25
  And we've got these increasingly capable AI systems that are
• 13:28
  good at synthesis and
• 13:29
  pattern recognition and surfacing things that humans would miss
• 13:32
  in large datasets.
• 13:33
  SPEAKER_01: The problem being that the dataset for zombie wells is partly digital,
• 13:38
  SPEAKER_01: partly on paper,
• 13:39
  SPEAKER_01: partly in people's heads,
• 13:41
  SPEAKER_01: and partly just doesn't exist.
• 13:43
  SPEAKER_00: Right.
• 13:44
  SPEAKER_00: And that's actually the interesting part.
• 13:47
  SPEAKER_00: Because the problem isn't just, find the wells we know about.
• 13:51
  SPEAKER_00: The problem is, find the wells that got lost.
• 13:55
  SPEAKER_00: The ones where the company that drilled them went bankrupt in 1986?
• 13:59
  SPEAKER_00: The records went into a box.
• 14:01
  SPEAKER_00: The box went into a warehouse.
• 14:03
  SPEAKER_00: The warehouse got flooded.
• 14:04
  SPEAKER_01: I have stories.
• 14:05
  SPEAKER_01: Multiple stories.
• 14:07
  SPEAKER_00: I know you do.
• 14:08
  SPEAKER_00: But stay with me.
• 14:10
  Because what I think is buildable right now with tools that
• 14:13
  exist is essentially a
• 14:14
  SPEAKER_00: probabilistic well location recovery system.
• 14:18
  SPEAKER_00: Not, here is where the well is.
• 14:21
  SPEAKER_00: More like,
• 14:22
  SPEAKER_00: here are the 50 square miles in Reeves County where the probability of an
• 14:26
  SPEAKER_00: unrecorded well is above 30%,
• 14:27
  SPEAKER_00: ranked by risk profile.
• 14:30
  SPEAKER_01: Based on what inputs?
• 14:31
  SPEAKER_00: Historical production data where it exists.
• 14:34
  SPEAKER_00: Old plat maps, even scanned ones.
• 14:38
  OCR has gotten good enough to extract location information from
• 14:41
  terrible scans,
• 14:43
  SPEAKER_00: lease records that do exist and imply drilling activity,
• 14:47
  SPEAKER_00: satellite and aerial imagery going back as far as we have it,
• 14:51
  SPEAKER_00: soil disturbance patterns,
• 14:52
  SPEAKER_00: vegetation anomalies,
• 14:54
  cross-referenced against what's already in the Railroad
• 14:56
  Commission database.
• 14:58
  SPEAKER_01: The negative space, you're looking at what's not recorded.
• 15:01
  SPEAKER_00: Exactly.
• 15:02
  SPEAKER_00: And here's where it gets interesting for your world specifically.
• 15:05
  SPEAKER_00: If you're doing due diligence on a land acquisition in a mature basin,
• 15:09
  SPEAKER_00: you currently have no good way to assess unrecorded well liability.
• 15:13
  SPEAKER_00: You look at what's in the records and you hope the records are right.
• 15:16
  SPEAKER_01: And they are not always right.
• 15:18
  SPEAKER_00: They are not always right.
• 15:20
  SPEAKER_00: But what if there was a risk score?
• 15:23
  Like this tract has a historical drilling density and
• 15:25
  production history that
• 15:26
  SPEAKER_00: suggests a high probability of additional unrecorded well bores.
• 15:30
  Here's what that liability exposure might look like based on
• 15:33
  comparable plugging
• 15:34
  SPEAKER_00: costs in the area.
• 15:35
  SPEAKER_01: That changes an acquisition conversation significantly.
• 15:39
  SPEAKER_00: Right?
• 15:40
  SPEAKER_00: Because right now that liability is either zero because nobody found it,
• 15:44
  SPEAKER_00: or it's a complete surprise when someone does find it.
• 15:47
  SPEAKER_00: There's no middle ground.
• 15:49
  SPEAKER_01: The buyer assumes it,
• 15:50
  SPEAKER_01: the seller doesn't disclose what they don't know,
• 15:53
  SPEAKER_01: and the regulator can't enforce what they can't find.
• 15:56
  SPEAKER_00: And the landowner,
• 15:57
  SPEAKER_00: whose fence post got pulled out in 1993,
• 15:59
  SPEAKER_00: is still sitting on a well nobody's plugged.
• 16:02
  SPEAKER_01: You know what this reminds me of?
• 16:03
  SPEAKER_01: The Iranian regime changed title chain thing.
• 16:07
  SPEAKER_01: Same structural problem.
• 16:09
  SPEAKER_01: You've got valid activity that happened under a prior regime,
• 16:12
  SPEAKER_01: whether that's a prior corporate owner or a defunct operating company,
• 16:16
  SPEAKER_01: and the current system of record doesn't capture it.
• 16:19
  SPEAKER_01: The liability is real.
• 16:21
  SPEAKER_01: The documentation is gone.
• 16:23
  SPEAKER_01: And the only way you find it is when it surfaces somewhere inconvenient.
• 16:27
  SPEAKER_00: Sack of cement and a prayer, as Oklahoma would say.
• 16:30
  SPEAKER_01: Sack of cement and a prayer.
• 16:32
  SPEAKER_01: Although in Oklahoma, at least, they acknowledge the problem exists.
• 16:36
  In some of these Texas situations,
• 16:38
  the existence of the problem is itself the surprise.
• 16:41
  SPEAKER_00: So what would actually need to be true for this to work?
• 16:44
  Because I can see the shape of the solution,
• 16:46
  but the data access question is real.
• 16:49
  SPEAKER_01: Railroad commission data is public.
• 16:51
  SPEAKER_01: Historical USGS records, public.
• 16:54
  SPEAKER_01: Aerial imagery archives, some of that's accessible.
• 16:58
  SPEAKER_01: The gap is the private records,
• 17:00
  SPEAKER_01: the operator files that never got submitted anywhere,
• 17:03
  SPEAKER_01: that live in banker's boxes in someone's Fort Worth storage unit.
• 17:08
  SPEAKER_00: Which is a business model problem as much as a technology problem.
• 17:12
  SPEAKER_00: Someone has to go get those records, digitize them, ingest them.
• 17:16
  SPEAKER_00: It's not a glamorous job.
• 17:18
  SPEAKER_01: It is absolutely not glamorous.
• 17:20
  SPEAKER_01: It is the opposite of glamorous.
• 17:23
  It is a 60-year-old document covered in coffee stains in a room
• 17:26
  that smells like
• 17:27
  SPEAKER_01: mildew and broken dreams.
• 17:28
  SPEAKER_00: And yet it is the thing that makes the AI actually work.
• 17:32
  SPEAKER_01: Back to the launch pad.
• 17:33
  SPEAKER_01: Nobody wants to build the launch pad.
• 17:35
  SPEAKER_00: But whoever builds the launch pad owns the rocket program.
• 17:38
  SPEAKER_01: Now that I'll get excited about, Dee.
• 17:41
  SPEAKER_00: Okay, build list.
• 17:43
  I've been thinking about this since the zombie well
• 17:44
  conversation and honestly since
• 17:46
  SPEAKER_00: some of the data management stuff that came up this week.
• 17:49
  SPEAKER_01: What number are we on?
• 17:51
  SPEAKER_00: This is bringing us to 5,
• 17:52
  SPEAKER_00: 6,
• 17:52
  and 7 on the midnight build list because it's been a productive
• 17:55
  week for my
• 17:55
  SPEAKER_00: insomnia.
• 17:56
  SPEAKER_01: The list breathes.
• 17:58
  SPEAKER_00: The list has a pulse.
• 18:00
  SPEAKER_00: So first one, and this is the big one from the collision zone.
• 18:04
  Imagine a due diligence tool specifically for mature basin
• 18:08
  acquisitions that
• 18:09
  SPEAKER_00: surfaces unrecorded liability risk.
• 18:12
  SPEAKER_00: Not a definitive answer.
• 18:13
  SPEAKER_00: A risk score.
• 18:15
  SPEAKER_00: A probability map.
• 18:18
  Something that changes how buyers and sellers have the
• 18:20
  liability conversation
• 18:21
  SPEAKER_00: before closing,
• 18:23
  SPEAKER_00: not after.
• 18:23
  SPEAKER_00: The problem is real.
• 18:26
  SPEAKER_00: It affects every acquisition in every mature basin in the country.
• 18:30
  SPEAKER_00: And the tools to approximate an answer exist right now.
• 18:34
  The value prop is avoiding the surprise,
• 18:36
  not eliminating the risk,
• 18:37
  avoiding the surprise.
• 18:39
  SPEAKER_00: Which in deal terms is worth a lot of money.
• 18:42
  SPEAKER_00: Second one,
• 18:43
  SPEAKER_00: and this is more on the AI governance side from the Anthropic story,
• 18:47
  SPEAKER_00: there's a real need for operator-specific AI deployment policies.
• 18:51
  SPEAKER_00: Not corporate policies, not vendor terms of service.
• 18:55
  SPEAKER_00: Actual documented decisions about which AI tools touch which data,
• 18:59
  SPEAKER_00: under what conditions,
• 19:01
  SPEAKER_00: with what human review requirements.
• 19:03
  SPEAKER_00: Because right now,
• 19:04
  most operators are using AI tools in an improvised way and
• 19:08
  hoping nothing goes
• 19:09
  SPEAKER_00: wrong.
• 19:09
  And when something goes wrong,
• 19:11
  they have no documentation that they made a considered
• 19:13
  decision.
• 19:14
  SPEAKER_01: That's a liability problem.
• 19:16
  That's a liability problem that becomes a very expensive
• 19:19
  problem very quickly.
• 19:20
  SPEAKER_00: And the third one?
• 19:22
  SPEAKER_00: The metadata issue that came up in the data management conversation?
• 19:27
  SPEAKER_00: Someone needs to build a land records confidence scoring system.
• 19:30
  SPEAKER_00: Not a digitization tool.
• 19:32
  SPEAKER_00: There are plenty of those.
• 19:34
  A tool that helps you understand which of your existing
• 19:36
  digitized records to trust,
• 19:38
  SPEAKER_00: for which purposes,
• 19:40
  SPEAKER_00: with what caveats.
• 19:41
  SPEAKER_00: Because the records exist,
• 19:43
  SPEAKER_00: and they're in systems,
• 19:44
  SPEAKER_00: but nobody knows which version of the truth to use when they conflict.
• 19:48
  SPEAKER_01: That would have saved me a great deal of suffering in Reeves County.
• 19:51
  SPEAKER_00: That would have saved you three days and a very difficult phone call.
• 19:54
  SPEAKER_01: Four days.
• 19:55
  SPEAKER_01: It was four days.
• 19:56
  SPEAKER_01: I blocked it from memory.
• 19:58
  SPEAKER_00: These problems are real.
• 19:59
  SPEAKER_00: The tools to address them exist.
• 20:02
  SPEAKER_00: Someone should build them.
• 20:03
  And if you're listening to this thinking,
• 20:05
  I actually need that for my operation,
• 20:08
  reach out.
• 20:09
  SPEAKER_00: SiliconAndShaleEggMail.com.
• 20:11
  SPEAKER_00: We build this stuff.
• 20:13
  SPEAKER_01: We do.
• 20:13
  SPEAKER_01: That's not hypothetical.
• 20:15
  SPEAKER_00: Can I just make one more point about the fusion energy thing?
• 20:19
  because it came up in the AI energy demand conversation this
• 20:21
  week,
• 20:22
  and I cannot let it go.
• 20:24
  Briefly,
• 20:24
  AI data centers are going to double their power consumption by
• 20:28
  2030.
• 20:29
  SPEAKER_00: Double.
• 20:30
  And Commonwealth Fusion Systems is genuinely using AI to
• 20:33
  accelerate their timeline
• 20:34
  SPEAKER_00: on commercial fusion power.
• 20:36
  The same AI that's creating the energy demand problem is being
• 20:39
  used to solve the
• 20:40
  SPEAKER_00: energy supply problem.
• 20:42
  That is either the most elegant thing I've ever heard or a
• 20:44
  complete Ouroboros of insanity.
• 20:47
  SPEAKER_01: The snake eating its own tail.
• 20:49
  SPEAKER_00: Exactly.
• 20:49
  SPEAKER_00: And the oil and gas implication is,
• 20:52
  if you think the AI electricity demand build-out is going to be
• 20:55
  solved by new
• 20:56
  SPEAKER_00: natural gas plants,
• 20:58
  SPEAKER_00: which is the current conventional wisdom,
• 21:00
  SPEAKER_00: that's probably right for the next five years.
• 21:03
  SPEAKER_00: But the 10-year picture is much murkier.
• 21:05
  And anyone making 30-year infrastructure bets right now is
• 21:08
  making them on a lot of
• 21:10
  SPEAKER_00: assumptions about what the energy mix looks like in 2040.
• 21:13
  30-year infrastructure bets in oil and gas are made on
• 21:16
  assumptions that don't
• 21:17
  SPEAKER_01: survive contact with the following week,
• 21:20
  SPEAKER_01: let alone 30 years.
• 21:21
  Which is why the landmen and engineers who can think across
• 21:24
  these technology curves
• 21:25
  SPEAKER_00: are going to be...
• 21:27
  SPEAKER_00: Actually,
• 21:27
  SPEAKER_00: this is the whole show,
• 21:28
  SPEAKER_00: isn't it?
• 21:30
  SPEAKER_00: That's the whole show.
• 21:31
  SPEAKER_01: That might actually be the whole show.
• 21:33
  SPEAKER_00: Three episodes in, and we've accidentally described our own thesis.
• 21:37
  SPEAKER_01: Oh,
• 21:37
  SPEAKER_01: and if your company sells to operators,
• 21:39
  SPEAKER_01: landmen,
• 21:40
  SPEAKER_01: or anyone in the energy space,
• 21:42
  SPEAKER_01: and you want to reach the people who actually make buying decisions,
• 21:45
  SPEAKER_01: we're taking sponsors,
• 21:47
  SPEAKER_01: siliconeandshellagmail.com.
• 21:51
  We'll use your product before we talk about it,
• 21:53
  which is more than most podcasts can say.
• 21:55
  SPEAKER_01: That's true, we actually will.
• 21:57
  SPEAKER_00: Okay, I have to go.
• 22:00
  I just got a notification that the thing I deployed this
• 22:02
  morning is doing something
• 22:03
  SPEAKER_00: it was not supposed to do,
• 22:04
  SPEAKER_00: and I need to...
• 22:06
  SPEAKER_01: Is this the...
• 22:07
  SPEAKER_00: Don't ask.
• 22:08
  SPEAKER_00: It's fine.
• 22:10
  SPEAKER_00: It's probably fine.
• 22:11
  SPEAKER_00: Go find your fence post.