Latest / Silicon & Shale Podcast / Episode 5: Qatar LNG Shock, NVIDIA's Enterprise AI Play, and Why Your Title Chain Might Be Wrong

Transcript

Episode 5: Qatar LNG Shock, NVIDIA's Enterprise AI Play, and Why Your Title Chain Might Be Wrong · Silicon & Shale Podcast

• 0:00
  And I'm telling you,
• 0:01
  the moment that headline crossed my phone,
• 0:03
  I nearly choked on my coffee.
• 0:05
  SPEAKER_00: Qatar's LNG infrastructure.
• 0:07
  SPEAKER_00: Damaged.
• 0:08
  SPEAKER_00: Not disrupted.
• 0:10
  SPEAKER_00: Not temporarily offline.
• 0:12
  SPEAKER_00: Damaged.
• 0:13
  SPEAKER_01: Like, structurally.
• 0:14
  SPEAKER_01: That's a different category of problem entirely.
• 0:17
  Right,
• 0:17
  because you can work around a disruption,
• 0:20
  reroute,
• 0:20
  delay,
• 0:21
  call in favors.
• 0:22
  SPEAKER_00: But damaged infrastructure means you're talking weeks,
• 0:25
  SPEAKER_00: months,
• 0:26
  SPEAKER_00: maybe longer before you're back at capacity.
• 0:28
  SPEAKER_00: And Qatar isn't some marginal supplier.
• 0:31
  SPEAKER_00: They're one of the biggest LNG exporters on the planet.
• 0:34
  SPEAKER_01: And the timing.
• 0:35
  SPEAKER_01: Right after the Hormuz stuff we talked about last week.
• 0:38
  SPEAKER_01: Right after the Red Sea refinery hit.
• 0:41
  SPEAKER_01: It's like every backup plan just got crossed off the list simultaneously.
• 0:44
  SPEAKER_00: I'll tell you what,
• 0:45
  I've been in this business long enough to remember when the
• 0:48
  phrase global supply
• 0:49
  SPEAKER_00: shock was something you read about in a textbook.
• 0:52
  SPEAKER_00: Now it's just Tuesday.
• 0:53
  SPEAKER_01: And the IEA apparently did not see this coming.
• 0:56
  SPEAKER_01: Which, okay, I have thoughts about that.
• 1:00
  SPEAKER_00: Of course you do.
• 1:01
  SPEAKER_01: I mean,
• 1:02
  if an agency that exists specifically to model energy supply
• 1:05
  risk can't see a Qatar
• 1:07
  SPEAKER_01: LNG disruption scenario in an already hot Middle East conflict,
• 1:11
  SPEAKER_01: that's not a forecast problem.
• 1:13
  SPEAKER_01: That's a data problem.
• 1:15
  Which honestly,
• 1:16
  we talked about last week too with the spare capacity stuff.
• 1:20
  SPEAKER_00: The data quality thing, yeah.
• 1:21
  SPEAKER_01: It keeps coming up.
• 1:23
  SPEAKER_01: The same fundamental issue keeps coming up.
• 1:26
  SPEAKER_00: It keeps coming up because nobody fixed it.
• 1:28
  SPEAKER_00: Silicon and shale.
• 1:30
  SPEAKER_00: Oil patch meets the algorithm.
• 1:32
  SPEAKER_00: Week of March 24th, 2026.
• 1:35
  SPEAKER_01: And I'm Packer.
• 1:36
  And today we've got the Qatar LNG situation,
• 1:39
  what it means for Asia specifically.
• 1:41
  SPEAKER_01: They're getting hammered.
• 1:43
  SPEAKER_01: Data center power demand crashing into the grid.
• 1:45
  SPEAKER_01: And some genuinely wild AI developments,
• 1:48
  SPEAKER_01: including NVIDIA,
• 1:49
  basically announcing what Jensen Huang is calling a new
• 1:51
  operating system for
• 1:52
  SPEAKER_01: enterprise AI.
• 1:54
  SPEAKER_00: Which I have questions about.
• 1:55
  SPEAKER_01: You always have questions.
• 1:57
  SPEAKER_01: That's why this works.
• 1:58
  SPEAKER_00: That's fair.
• 1:59
  SPEAKER_00: Let's get into it.
• 2:01
  SPEAKER_00: Sawyer EMP Deep Dive.
• 2:03
  SPEAKER_01: Okay,
• 2:04
  so let's actually dig into the Qatar situation because the
• 2:07
  headline is alarming,
• 2:08
  SPEAKER_01: but I think the implications are even bigger than the headline.
• 2:12
  The crude realities guys,
• 2:13
  Chuck and Mark,
• 2:14
  broke this down and they weren't pulling punches.
• 2:17
  SPEAKER_00: Qatar LNG infrastructure heavily damaged,
• 2:20
  SPEAKER_00: Red Sea refinery hit,
• 2:21
  SPEAKER_00: and what they called the IEA apparently never seeing this coming.
• 2:26
  SPEAKER_00: And look, I believe it.
• 2:27
  SPEAKER_00: Because here's the thing, people forget about LNG infrastructure.
• 2:31
  SPEAKER_00: It's not plug and play.
• 2:32
  SPEAKER_01: Not even close.
• 2:33
  SPEAKER_00: You've got liquefaction trains,
• 2:35
  SPEAKER_00: you've got loading terminals,
• 2:37
  SPEAKER_00: you've got the regasification capacity on the other end.
• 2:41
  SPEAKER_00: You damage one link in that chain and the whole system backs up.
• 2:45
  It's not like crude where you've got spot tankers and a dozen
• 2:48
  alternative routes.
• 2:50
  SPEAKER_01: The destination flexibility in LNG is way more constrained.
• 2:54
  You've got purpose-built terminals,
• 2:56
  long-term contracts,
• 2:57
  specific ship classes.
• 2:59
  SPEAKER_00: And Asia is holding the bag on this one.
• 3:01
  SPEAKER_01: $200 oil in Asia was being floated.
• 3:04
  Which sounds crazy until you think about who doesn't have
• 3:07
  alternatives right now.
• 3:09
  SPEAKER_01: Japan, South Korea, Taiwan.
• 3:12
  SPEAKER_01: They don't have domestic production to fall back on.
• 3:16
  This isn't a we'll buy from someone else situation when
• 3:19
  everyone else is already spoken for.
• 3:21
  SPEAKER_00: You know what this reminds me of?
• 3:24
  SPEAKER_00: Early on in my career,
• 3:25
  SPEAKER_00: late 90s,
• 3:26
  SPEAKER_00: there was a pipeline that went down and actually doesn't matter where.
• 3:30
  But a single line went down and suddenly you had refineries in
• 3:34
  three states
• 3:34
  scrambling because everyone had optimized their supply chains
• 3:37
  so tight that there
• 3:38
  SPEAKER_00: was no slack left.
• 3:40
  SPEAKER_00: Nobody had buffer inventory because buffer inventory costs money.
• 3:45
  SPEAKER_00: And the answer at the time was, well, the market will sort it out.
• 3:49
  Which it did,
• 3:50
  eventually,
• 3:51
  but the word eventually did a lot of work there.
• 3:53
  SPEAKER_01: And that's the regime adjustment versus regime change question, right?
• 3:58
  SPEAKER_01: Chuck and Mark were asking,
• 3:59
  SPEAKER_01: is Iran going to try to ride this out,
• 4:01
  or is there an endgame here that fundamentally restructures
• 4:04
  Middle East energy
• 4:05
  SPEAKER_01: supply?
• 4:06
  SPEAKER_00: And that's not an energy question anymore.
• 4:09
  SPEAKER_00: That's a geopolitics question wearing an energy questions hat.
• 4:12
  SPEAKER_01: Which means the models can't really help you.
• 4:15
  SPEAKER_01: You can't model regime intent.
• 4:17
  SPEAKER_00: No, no, you can't.
• 4:19
  SPEAKER_00: You can model cargo flows and tanker traffic and days of supply.
• 4:23
  SPEAKER_00: But what does the Ayatollah want?
• 4:25
  SPEAKER_00: That's above my pay grade, and it's above any algorithm's pay grade, too.
• 4:29
  SPEAKER_01: What's interesting to me is the gas price piece.
• 4:32
  They mentioned gas prices mattering more politically than
• 4:35
  anyone wants to admit.
• 4:36
  SPEAKER_01: And that's so true.
• 4:38
  SPEAKER_01: Like, crude price is a financial headline.
• 4:41
  SPEAKER_01: Gas prices at the pump are a political reality.
• 4:44
  Seven dollar gas and suddenly everyone in Washington has an
• 4:47
  opinion on the Strait of Hormuz.
• 4:49
  SPEAKER_01: Which, by the way, is a different policy lever than oil.
• 4:53
  SPEAKER_01: Because you can release from the SPR.
• 4:55
  SPEAKER_01: We talked about that a few weeks ago.
• 4:58
  SPEAKER_01: But you can't SPR your way out of an LNG supply crunch.
• 5:02
  SPEAKER_01: Those are different systems.
• 5:03
  SPEAKER_00: Completely different.
• 5:05
  SPEAKER_00: The Strategic Petroleum Reserve is crude.
• 5:08
  SPEAKER_00: It does nothing for your natural gas supply.
• 5:11
  If LNG from Qatar is offline and Europe and Asia are fighting
• 5:15
  over what's left from the U.S.
• 5:16
  SPEAKER_00: Gulf Coast, the SPR is basically a spectator.
• 5:20
  SPEAKER_01: And U.S.
• 5:21
  SPEAKER_01: LNG export capacity is already running near full.
• 5:24
  SPEAKER_01: There's not a surge button.
• 5:26
  SPEAKER_00: Not in the short term.
• 5:27
  SPEAKER_00: You're talking years to bring new liquefaction trains online.
• 5:31
  SPEAKER_00: Sabine Pass, Freeport, they're running.
• 5:34
  You can't just will more capacity into existence because the
• 5:37
  price signal is right.
• 5:39
  SPEAKER_01: Okay, so?
• 5:40
  SPEAKER_01: Courthouse Chronicles.
• 5:42
  SPEAKER_01: You were going to connect this to something on the land side?
• 5:45
  SPEAKER_00: Oh, I was getting there.
• 5:46
  SPEAKER_00: So here's the thing.
• 5:48
  SPEAKER_00: This whole situation,
• 5:50
  SPEAKER_00: the LNG damage,
• 5:51
  SPEAKER_00: the scramble for alternatives,
• 5:52
  SPEAKER_00: the market shock,
• 5:54
  it's making operators think about something they haven't
• 5:56
  thought about seriously in
• 5:57
  SPEAKER_00: a few years,
• 5:58
  SPEAKER_00: which is domestic gas supply chains.
• 6:01
  SPEAKER_00: Who owns what?
• 6:02
  SPEAKER_00: What's connected to what?
• 6:04
  And that means people are suddenly very interested in old
• 6:06
  midstream easements and
• 6:08
  SPEAKER_00: right-of-way agreements that haven't been touched since the 80s.
• 6:11
  SPEAKER_01: Oh, no.
• 6:12
  SPEAKER_00: Oh, yes.
• 6:13
  SPEAKER_00: So, Grady County.
• 6:15
  SPEAKER_00: I was down there,
• 6:15
  SPEAKER_00: I want to say,
• 6:16
  SPEAKER_00: late last year,
• 6:18
  SPEAKER_00: county clerk's office,
• 6:19
  and someone needed to trace a gas-gathering easement that ran
• 6:22
  across three
• 6:23
  SPEAKER_00: different surface tracks.
• 6:25
  SPEAKER_00: The original agreement was from 1979, handwritten, carbon copy.
• 6:31
  SPEAKER_00: The landman who drafted it had,
• 6:33
  SPEAKER_00: and I say this charitably,
• 6:35
  SPEAKER_00: creative ideas about legal descriptions.
• 6:37
  SPEAKER_01: Creative how?
• 6:38
  The description referenced a fence post that no longer exists
• 6:42
  and a creek that had moved.
• 6:43
  SPEAKER_01: The fax machine is still warm.
• 6:45
  SPEAKER_00: The fax machine is still warm, Pac.
• 6:48
  SPEAKER_00: So you've got this easement that is legally valid,
• 6:51
  SPEAKER_00: probably,
• 6:52
  SPEAKER_00: but physically ambiguous,
• 6:54
  on a gathering line that is now strategically relevant because
• 6:57
  the operator is
• 6:58
  SPEAKER_00: trying to figure out if they can reroute gas to a different interconnect.
• 7:02
  And the answer to that question is buried in a Grady County
• 7:05
  deed book from 1979
• 7:06
  SPEAKER_00: that nobody has touched in 20 years.
• 7:09
  SPEAKER_01: And nobody has digitized.
• 7:11
  SPEAKER_00: Nobody has digitized.
• 7:13
  SPEAKER_00: I'm standing in the basement with my phone,
• 7:15
  SPEAKER_00: bad lighting,
• 7:15
  SPEAKER_00: trying to get a readable photo of a carbon copy.
• 7:18
  SPEAKER_00: This is where the rubber meets the road.
• 7:21
  All your AI tools and your beautiful data pipelines,
• 7:25
  they stop at the door of that basement.
• 7:27
  SPEAKER_01: I mean, you're not wrong.
• 7:29
  SPEAKER_01: But, and I say this with love, that's not an AI problem.
• 7:33
  SPEAKER_01: That's a digitization problem, which is solvable.
• 7:37
  SPEAKER_01: Slowly and expensively, but solvable.
• 7:40
  SPEAKER_00: How slowly?
• 7:41
  SPEAKER_01: In a current pace?
• 7:43
  SPEAKER_01: Decades.
• 7:43
  SPEAKER_00: Hmm.
• 7:44
  SPEAKER_00: So maybe I'll be retired before it matters.
• 7:47
  SPEAKER_01: I refuse to let you retire.
• 7:48
  SPEAKER_00: I'm one bad PDF scan away.
• 7:51
  SPEAKER_01: Can we talk about the data center power demand angle?
• 7:54
  Because the energy gang covered something that connects to all
• 7:56
  of this in a way
• 7:57
  SPEAKER_01: that I think is underappreciated.
• 7:59
  SPEAKER_00: The grid story.
• 8:00
  SPEAKER_01: The grid story.
• 8:01
  SPEAKER_01: So for two decades,
• 8:03
  SPEAKER_01: basically flat electricity demand in the U.S.,
• 8:06
  SPEAKER_01: prices barely moved,
• 8:08
  SPEAKER_01: then data centers explode,
• 8:10
  SPEAKER_01: electrification kicks in,
• 8:11
  and suddenly the grid is being asked to do things it was never
• 8:14
  designed for.
• 8:15
  And the host point was that the era of stable,
• 8:18
  cheap power is genuinely over.
• 8:20
  SPEAKER_00: Which is a natural gas story.
• 8:22
  SPEAKER_01: Completely.
• 8:23
  SPEAKER_01: Because when renewables can't keep up,
• 8:25
  SPEAKER_01: and at scale in certain hours of the day they can't,
• 8:29
  SPEAKER_01: you need dispatchable generation.
• 8:31
  SPEAKER_01: And dispatchable means gas.
• 8:33
  So the same commodity that's tight on the LNG side because of
• 8:36
  Qatar is also getting
• 8:37
  SPEAKER_00: pulled by data center demand growth domestically.
• 8:40
  SPEAKER_00: That's pressure from both ends.
• 8:42
  Which means any operator sitting on Haynesville or Marcellus
• 8:45
  Acreage right now is
• 8:46
  SPEAKER_01: having a very different year than they expected six months ago.
• 8:49
  SPEAKER_00: How's the hedging looking for those guys?
• 8:52
  SPEAKER_01: That's the question, right?
• 8:53
  SPEAKER_01: Because if you hedged at $2.50,
• 8:56
  SPEAKER_01: Henry Hub,
• 8:56
  and we're now looking at sustained demand pull from AI
• 8:59
  infrastructure build-out,
• 9:00
  SPEAKER_01: plus LNG export demand,
• 9:02
  SPEAKER_01: plus supply disruption,
• 9:04
  SPEAKER_01: you potentially left a lot of money on the table.
• 9:06
  And the guys who didn't hedge are either heroes or idiots,
• 9:09
  depending on which week you ask them.
• 9:12
  SPEAKER_01: Every week in this industry.
• 9:13
  SPEAKER_00: Every single week.
• 9:15
  SPEAKER_00: Nerdy.
• 9:16
  SPEAKER_00: AI deep dive.
• 9:18
  SPEAKER_01: Okay.
• 9:20
  SPEAKER_01: NVIDIA GTC.
• 9:21
  SPEAKER_01: Can we talk about what Jensen Huang actually said?
• 9:24
  Because new operating system is either one of the most
• 9:27
  important things anyone said
• 9:28
  SPEAKER_01: about enterprise AI,
• 9:30
  SPEAKER_01: or it's the greatest marketing line of the year.
• 9:31
  SPEAKER_00: Or both.
• 9:33
  SPEAKER_00: Those aren't mutually exclusive.
• 9:35
  SPEAKER_01: Okay, fair.
• 9:36
  SPEAKER_01: So here's the thing.
• 9:38
  SPEAKER_01: NVIDIA just dropped a new model at GTC.
• 9:40
  SPEAKER_01: Massive thing.
• 9:41
  SPEAKER_01: 120 billion parameters.
• 9:44
  SPEAKER_01: But they're claiming it runs at the speed of something a tenth that size.
• 9:48
  SPEAKER_01: That's a ten times efficiency gain.
• 9:50
  SPEAKER_01: And if that's real... Is it real?
• 9:52
  SPEAKER_01: The benchmarks suggest it's real.
• 9:54
  SPEAKER_01: Yeah.
• 9:55
  SPEAKER_01: And the reason it matters is...
• 9:57
  The whole argument against big enterprise models has always
• 9:59
  been latency and cost.
• 10:02
  SPEAKER_01: You want a model that can reason at scale?
• 10:04
  SPEAKER_01: It's slow and expensive.
• 10:06
  Now,
• 10:07
  if you can get that reasoning capability at fast model
• 10:09
  speeds...
• 10:10
  The cost equation changes.
• 10:12
  SPEAKER_01: The cost equation changes completely.
• 10:14
  SPEAKER_01: And this is where the new operating system framing starts to make sense.
• 10:19
  SPEAKER_01: Because Jensen's not talking about a chatbot.
• 10:22
  SPEAKER_01: He's talking about multi-agent systems.
• 10:24
  SPEAKER_01: Multiple AI agents working together, coordinating, handing off tasks...
• 10:29
  SPEAKER_00: Okay, hold on.
• 10:30
  SPEAKER_00: When you say multi-agent,
• 10:32
  SPEAKER_00: walk me through what that actually looks like in practice,
• 10:35
  SPEAKER_00: because I've heard agents thrown around a lot lately,
• 10:37
  and it means everything from slightly smarter autocomplete to
• 10:42
  the robot that runs
• 10:43
  SPEAKER_00: your whole company.
• 10:44
  SPEAKER_01: So think of it like a crew on a drilling rig.
• 10:47
  SPEAKER_01: You've got a driller, a tool pusher, a company man, a mud logger.
• 10:51
  SPEAKER_01: They all have different jobs.
• 10:53
  SPEAKER_01: They're all working on the same well, and they're coordinating.
• 10:57
  SPEAKER_01: A multi-agent system is like that,
• 10:59
  SPEAKER_01: except each crew member is an AI that specializes in one thing.
• 11:04
  SPEAKER_01: One agent handles research.
• 11:06
  SPEAKER_01: One handles drafting.
• 11:07
  SPEAKER_01: One handles quality checking.
• 11:09
  SPEAKER_01: One handles the final output.
• 11:12
  SPEAKER_01: They pass work between each other.
• 11:13
  SPEAKER_00: And nobody's asleep on tower.
• 11:15
  SPEAKER_01: Nobody's asleep.
• 11:17
  SPEAKER_01: They don't sleep.
• 11:18
  SPEAKER_01: Which is both the selling point and, honestly, a little unsettling.
• 11:22
  SPEAKER_00: Yeah, no, I see why Jensen is excited.
• 11:25
  SPEAKER_00: But here's my real-world question.
• 11:27
  SPEAKER_00: What happens when two of those agents disagree?
• 11:31
  SPEAKER_00: On a rig, the company man makes the call.
• 11:33
  SPEAKER_00: Who's the company man in a multi-agent system?
• 11:36
  SPEAKER_01: That's, honestly, that's one of the genuinely hard unsolved problems.
• 11:41
  SPEAKER_01: And the research from the big AI labs keeps bumping into it.
• 11:45
  SPEAKER_01: Like,
• 11:45
  SPEAKER_01: one of them just shipped this advanced coding agent,
• 11:48
  which apparently has triggered a chain reaction across every
• 11:50
  major lab trying to
• 11:51
  SPEAKER_01: catch up.
• 11:53
  SPEAKER_01: The big frontier is exactly that.
• 11:55
  SPEAKER_01: Agent coordination and who has final authority.
• 11:58
  SPEAKER_00: Because if nobody has final authority, you're just going in circles.
• 12:02
  Or you get the scenario where an AI agent's operating with
• 12:05
  enough autonomy that it
• 12:06
  SPEAKER_01: starts doing things nobody anticipated.
• 12:09
  SPEAKER_01: Making API calls it shouldn't.
• 12:11
  SPEAKER_01: Accessing data outside its scope.
• 12:14
  SPEAKER_01: That kind of stuff is already happening at some of the bigger labs.
• 12:17
  SPEAKER_00: That's the nightmare scenario for anyone running operations.
• 12:20
  SPEAKER_01: And the takeaway is,
• 12:22
  autonomous agents doing things unsupervised can go sideways in
• 12:25
  ways that are
• 12:26
  SPEAKER_01: genuinely hard to predict.
• 12:28
  SPEAKER_00: I feel like that should have been a bigger headline.
• 12:30
  It probably would have been if it weren't for approximately 17
• 12:33
  other things
• 12:34
  SPEAKER_01: happening simultaneously in AI right now.
• 12:37
  SPEAKER_00: Okay,
• 12:37
  SPEAKER_00: so,
• 12:38
  SPEAKER_00: fax machine to fully autonomous scale,
• 12:41
  SPEAKER_00: where does multi-agent enterprise AI sit right now?
• 12:44
  SPEAKER_01: I want to say six.
• 12:47
  SPEAKER_01: Maybe six and a half.
• 12:49
  The technology is genuinely impressive and it works in
• 12:52
  controlled environments.
• 12:54
  SPEAKER_01: But the coordination problem,
• 12:55
  SPEAKER_01: the authority problem,
• 12:57
  SPEAKER_01: the what-does-it-do-when-something-unexpected-happens problem,
• 13:02
  SPEAKER_01: those are real.
• 13:04
  SPEAKER_01: Production ready at scale?
• 13:06
  SPEAKER_01: Not quite yet.
• 13:07
  SPEAKER_00: I was going to say five, so we're close.
• 13:10
  SPEAKER_00: You're being generous.
• 13:11
  SPEAKER_00: I know it's unlike me, but that efficiency number is real.
• 13:15
  SPEAKER_00: If a 120 billion parameter model runs like a 12 billion parameter model,
• 13:20
  SPEAKER_00: that's not marketing,
• 13:21
  SPEAKER_00: that's engineering.
• 13:23
  SPEAKER_01: Right.
• 13:23
  SPEAKER_01: And it matters because the cost barrier for enterprise AI has been huge.
• 13:28
  This is the kind of thing that moves AI from pilot program to
• 13:32
  we actually deployed this.
• 13:34
  SPEAKER_00: Now there was something else.
• 13:35
  SPEAKER_00: Andre Karpathy, the auto research thing.
• 13:39
  SPEAKER_01: Okay, this one I'm genuinely excited about.
• 13:42
  SPEAKER_01: So Karpathy was talking about what he's calling the loopy era of AI,
• 13:46
  SPEAKER_01: where agents can design experiments,
• 13:48
  SPEAKER_01: collect data,
• 13:50
  SPEAKER_01: run the experiments,
• 13:51
  SPEAKER_01: analyze the results,
• 13:52
  SPEAKER_01: and improve themselves.
• 13:54
  SPEAKER_01: Without a human in the loop.
• 13:55
  SPEAKER_00: That's... Okay, I need to sit with that for a second.
• 13:58
  SPEAKER_01: Take your time.
• 13:59
  SPEAKER_00: So the AI is doing science by itself.
• 14:03
  SPEAKER_01: Not all science.
• 14:05
  SPEAKER_01: But a specific class of AI research problems...
• 14:08
  SPEAKER_01: Yes.
• 14:10
  SPEAKER_01: Auto-research is basically AI closing the loop on its own optimization.
• 14:14
  It runs an experiment,
• 14:16
  sees the result,
• 14:17
  decides what to try next,
• 14:19
  runs that.
• 14:21
  And it's doing this in domains where the experiment feedback
• 14:23
  loop is fast.
• 14:25
  Which explains last week's story,
• 14:27
  the battery R&D,
• 14:28
  eight years of research in two weeks.
• 14:31
  SPEAKER_01: Right.
• 14:31
  SPEAKER_01: Same concept.
• 14:33
  SPEAKER_01: When the feedback loop is fast and the search space is well-defined,
• 14:37
  SPEAKER_01: AI can explore it at a rate no human team can match.
• 14:40
  SPEAKER_00: What's the search space for oil and gas applications?
• 14:43
  SPEAKER_00: Like where's the feedback loop fast enough for this to matter?
• 14:47
  SPEAKER_01: Reservoir simulation.
• 14:49
  SPEAKER_01: Completion optimization.
• 14:51
  SPEAKER_01: Production forecasting.
• 14:53
  Anywhere you've got a model,
• 14:54
  parameters you can tune,
• 14:56
  and a metric you're optimizing for.
• 14:57
  SPEAKER_00: Okay, that's, yeah, that's actually real.
• 15:00
  SPEAKER_01: You're impressed.
• 15:02
  SPEAKER_00: I'm cautiously acknowledging that this is not nothing.
• 15:05
  SPEAKER_01: I'm writing that down.
• 15:07
  SPEAKER_01: Episode five, Derek acknowledges something is not nothing.
• 15:11
  SPEAKER_00: Don't make it weird.
• 15:12
  SPEAKER_01: It's already a little weird.
• 15:14
  SPEAKER_01: Can I add something about the AI layoffs angle?
• 15:17
  SPEAKER_01: Because hard fork covered the Atlassian and block cuts,
• 15:20
  SPEAKER_01: the meta 20% workforce reduction,
• 15:23
  and the question of whether this is actual AI displacement or
• 15:27
  just regular tech
• 15:28
  SPEAKER_01: layoffs wearing an AI story.
• 15:30
  SPEAKER_00: What's the answer?
• 15:31
  SPEAKER_01: Honestly, probably both.
• 15:34
  SPEAKER_01: Some of it is AI replacing certain tasks.
• 15:37
  SPEAKER_01: Code review, documentation, some junior dev work.
• 15:41
  Some of it is just tech companies that overhired in 2021 and
• 15:44
  are correcting.
• 15:46
  SPEAKER_01: The AI washing of layoffs is real.
• 15:48
  Companies love to say we're restructuring for an AI-first
• 15:51
  future when really they
• 15:53
  SPEAKER_01: just hired too many people.
• 15:55
  SPEAKER_00: That's every restructuring story since I've been alive.
• 15:58
  SPEAKER_00: They find the current fashionable reason.
• 16:00
  SPEAKER_01: But here's the thing.
• 16:02
  SPEAKER_01: Even if 50% of those layoffs are AI-washed, the other 50% is real.
• 16:07
  SPEAKER_01: And the jobs that are disappearing are not evenly distributed.
• 16:11
  SPEAKER_01: It's the entry level, the routine, the well-defined tasks.
• 16:14
  SPEAKER_00: Which, in our world, is some of what junior landmen do.
• 16:18
  SPEAKER_01: Yeah.
• 16:19
  SPEAKER_01: Title, run sheet automation, lease abstraction, document sorting.
• 16:24
  SPEAKER_01: Those are things that AI is genuinely getting good at.
• 16:27
  SPEAKER_01: That's not hypothetical anymore.
• 16:29
  SPEAKER_00: I've said this before and I'll say it again.
• 16:32
  SPEAKER_00: The senior landmen who understand the exceptions,
• 16:35
  SPEAKER_00: who know what the document means in context,
• 16:37
  SPEAKER_00: who can smell when something's wrong,
• 16:39
  SPEAKER_00: those people are fine.
• 16:41
  It's the folks who are just turning the crank who should be
• 16:44
  paying attention.
• 16:45
  SPEAKER_01: Which is actually the charitable reading of the AI story.
• 16:49
  SPEAKER_01: It's not replacing judgment.
• 16:52
  SPEAKER_01: It's replacing repetition.
• 16:54
  SPEAKER_01: For now.
• 16:55
  SPEAKER_00: For now.
• 16:56
  SPEAKER_01: For now.
• 16:58
  SPEAKER_01: The collision zone.
• 17:00
  SPEAKER_00: All right.
• 17:01
  SPEAKER_00: Collision zone time.
• 17:02
  And this one's been forming in my head all week because of the
• 17:05
  combination of the
• 17:06
  Qatar story and the courthouse stuff and something I keep
• 17:09
  running into in my actual
• 17:10
  SPEAKER_00: day-to-day.
• 17:11
  SPEAKER_01: Let's hear it.
• 17:12
  SPEAKER_00: So here's the problem.
• 17:14
  When there's a supply shock,
• 17:15
  and we've now had several in quick succession,
• 17:18
  right?
• 17:18
  SPEAKER_00: Hormuz, Qatar, Red Sea.
• 17:21
  SPEAKER_00: Operators need to make fast decisions about their domestic production.
• 17:24
  SPEAKER_00: Can we accelerate?
• 17:26
  SPEAKER_00: Can we reroute?
• 17:27
  SPEAKER_00: Can we bring something back online?
• 17:29
  SPEAKER_00: And the answer to every one of those questions depends on title,
• 17:33
  SPEAKER_00: on ownership,
• 17:34
  SPEAKER_00: on who has the right to produce from what.
• 17:36
  SPEAKER_01: Which takes forever to determine.
• 17:38
  SPEAKER_00: Takes forever.
• 17:40
  SPEAKER_00: Because the information is scattered across 50 county courthouses,
• 17:44
  SPEAKER_00: some of which have it digitized,
• 17:46
  SPEAKER_00: most of which don't,
• 17:48
  SPEAKER_00: all of which have their own filing systems,
• 17:50
  SPEAKER_00: their own indexing logic,
• 17:51
  SPEAKER_00: their own idea of what constitutes a searchable record.
• 17:55
  SPEAKER_00: And by the time you've run your title and confirmed that,
• 17:57
  SPEAKER_00: yes,
• 17:57
  SPEAKER_00: you own this thing,
• 17:58
  SPEAKER_00: and yes,
• 17:59
  SPEAKER_00: you have the right to produce it,
• 18:00
  SPEAKER_00: the market window has moved.
• 18:02
  So the speed of title research is literally limiting the speed
• 18:05
  at which operators
• 18:06
  SPEAKER_01: can respond to market signals.
• 18:08
  SPEAKER_00: That's the whole thing.
• 18:10
  SPEAKER_00: And last week,
• 18:11
  we talked about how data infrastructure quality problems
• 18:14
  undermine the ability to
• 18:15
  SPEAKER_00: quickly activate spare capacity.
• 18:18
  SPEAKER_00: This is the land side of that same coin.
• 18:21
  SPEAKER_00: It's not just wellbore data.
• 18:23
  SPEAKER_00: It's ownership data.
• 18:24
  SPEAKER_01: Okay,
• 18:26
  so the AI capability that I keep thinking about in this context
• 18:29
  is what Carpathy
• 18:30
  SPEAKER_01: was describing with the loopy,
• 18:31
  SPEAKER_01: self-improving research systems,
• 18:34
  SPEAKER_01: but applied to document understanding.
• 18:37
  Because the piece that's changed in the last 18 months is how
• 18:39
  good AI has gotten at
• 18:41
  SPEAKER_01: reading and understanding unstructured documents.
• 18:44
  SPEAKER_01: Handwritten.
• 18:45
  SPEAKER_01: Partial.
• 18:46
  SPEAKER_00: Ambiguous.
• 18:47
  SPEAKER_00: The Carbon Copy from 1979.
• 18:49
  SPEAKER_01: The carbon copy from 1979.
• 18:52
  SPEAKER_01: Fence posts that don't exist.
• 18:54
  SPEAKER_01: Creeks that moved.
• 18:56
  SPEAKER_01: AI can now,
• 18:58
  SPEAKER_01: with the right setup,
• 18:59
  SPEAKER_01: read that document,
• 19:00
  SPEAKER_01: flag the ambiguity,
• 19:02
  SPEAKER_01: cross-reference other records from the same time period and same area,
• 19:05
  and give you a probability assessment on what the legal
• 19:08
  description was probably
• 19:09
  SPEAKER_01: intending to say.
• 19:11
  SPEAKER_00: Probably.
• 19:11
  SPEAKER_01: Probably.
• 19:13
  SPEAKER_01: Which is the key word.
• 19:15
  SPEAKER_01: It's not replacing the landman's judgment.
• 19:18
  SPEAKER_01: It's giving the landman something to exercise their judgment on.
• 19:21
  SPEAKER_01: Faster.
• 19:22
  SPEAKER_00: Here's my concern, and it's a real one, not just me being ornery.
• 19:27
  The value of a title opinion is that a licensed attorney put
• 19:30
  their professional
• 19:31
  SPEAKER_00: reputation behind it.
• 19:33
  If an AI reads a document and says,
• 19:35
  probably,
• 19:36
  who's responsible for that probably?
• 19:39
  SPEAKER_01: And that's where it gets real.
• 19:40
  Because an operator makes a drilling decision based on probably
• 19:43
  owns it and then
• 19:44
  SPEAKER_00: turns out they don't.
• 19:46
  SPEAKER_00: That's not a software bug.
• 19:47
  SPEAKER_00: That's a lawsuit.
• 19:49
  SPEAKER_01: So what if the framing isn't AI gives you the title opinion?
• 19:53
  SPEAKER_01: It's AI does the triage.
• 19:55
  It reads 10,000 documents and tells the human exactly which 50
• 19:59
  need a licensed eye
• 20:00
  because they have ambiguity and which 9,950 are clean and can
• 20:04
  be processed
• 20:05
  SPEAKER_01: automatically.
• 20:06
  That's actually interesting,
• 20:08
  because right now the human is reading all 10,000.
• 20:11
  SPEAKER_01: Right, and most of them are fine.
• 20:14
  SPEAKER_01: Boilerplate, straightforward.
• 20:17
  SPEAKER_01: The landman's time,
• 20:18
  SPEAKER_01: and the attorney's time,
• 20:20
  SPEAKER_01: is being spent on documents that don't need that level of attention,
• 20:23
  which means the documents that do need it sometimes get less
• 20:27
  than they deserve
• 20:27
  SPEAKER_01: because everyone's exhausted.
• 20:29
  SPEAKER_00: I will grant you that's a real problem.
• 20:32
  I have personally signed off on something at four in the
• 20:35
  afternoon on a Friday that
• 20:36
  SPEAKER_00: I probably should have looked at harder.
• 20:38
  SPEAKER_01: Everybody has.
• 20:40
  SPEAKER_01: And that's not incompetence.
• 20:41
  SPEAKER_01: That's volume.
• 20:43
  That's a system that wasn't designed for the document load that
• 20:45
  modern operations generate.
• 20:47
  SPEAKER_00: So you're saying AI is a triage filter.
• 20:51
  SPEAKER_00: It reads everything,
• 20:52
  SPEAKER_00: flags the weird stuff,
• 20:53
  SPEAKER_00: routes the clean stuff through,
• 20:55
  SPEAKER_00: and lets the human concentrate their attention.
• 20:58
  SPEAKER_01: Yes.
• 20:59
  SPEAKER_01: And the implications for the supply shock scenario are,
• 21:02
  imagine if an operator could turn around a preliminary title
• 21:05
  assessment on a set of
• 21:06
  SPEAKER_01: leases in hours instead of weeks.
• 21:08
  SPEAKER_01: Not final.
• 21:10
  SPEAKER_01: Not bankable.
• 21:11
  SPEAKER_01: But good enough to say, yes, this tract is worth pursuing.
• 21:15
  SPEAKER_01: Or this one has a cloud on title that will take months to resolve.
• 21:19
  SPEAKER_01: Move on.
• 21:20
  SPEAKER_00: That's portfolio decision-making.
• 21:22
  SPEAKER_00: You're not running title for one well.
• 21:24
  You're running triage on 50 prospects and figuring out where to
• 21:28
  put your people.
• 21:29
  And that's what you need when the market moves fast and you
• 21:31
  need to move faster.
• 21:32
  SPEAKER_00: The thing that would make or break this is data quality going in.
• 21:37
  SPEAKER_00: You said it yourself last week.
• 21:39
  SPEAKER_00: Garbage in, garbage out.
• 21:41
  SPEAKER_00: If the courthouse records aren't digitized,
• 21:43
  SPEAKER_00: if the scan quality is bad,
• 21:45
  SPEAKER_00: if the handwriting is illegible...
• 21:47
  SPEAKER_01: Then the AI is doing its best, and its best might not be good enough.
• 21:51
  SPEAKER_00: And there's no standard.
• 21:52
  SPEAKER_00: Every county has their own filing system.
• 21:55
  SPEAKER_00: Grady County files by section township range.
• 21:58
  SPEAKER_00: Other counties file by grant or grantee.
• 22:01
  SPEAKER_00: Some have both.
• 22:02
  SPEAKER_00: Some have neither, and you're just looking.
• 22:05
  Which means you'd need to solve the data standardization
• 22:07
  problem before or
• 22:08
  SPEAKER_01: alongside the AI problem.
• 22:11
  SPEAKER_01: And that's not a six-month project.
• 22:13
  SPEAKER_00: No, that's a decade project.
• 22:14
  SPEAKER_01: Unless,
• 22:16
  SPEAKER_01: and this is the thing that keeps me up at night,
• 22:18
  SPEAKER_01: unless you build the AI system to handle the variability itself,
• 22:22
  to learn that Grady County files this way and Beckham County
• 22:25
  files that way,
• 22:26
  SPEAKER_01: and it knows to look in different places depending on where you are.
• 22:30
  SPEAKER_00: A system that knows the county clerk.
• 22:32
  Metaphorically,
• 22:33
  I know the county clerks,
• 22:35
  literally,
• 22:35
  by name,
• 22:36
  and they know me.
• 22:38
  And that relationship is worth something that no system has yet
• 22:41
  figured out how to replicate.
• 22:43
  Because sometimes the clerk knows that the thing you're looking
• 22:46
  for got misfiled in
• 22:48
  SPEAKER_00: 1987,
• 22:49
  and it's actually in the grantor-grantee index under the wrong
• 22:52
  name because
• 22:52
  SPEAKER_00: somebody made a typo,
• 22:54
  SPEAKER_00: and she just knows that because she's been there for 30 years.
• 22:58
  SPEAKER_01: That is a real problem.
• 22:59
  SPEAKER_01: But,
• 23:00
  SPEAKER_01: and I'm not dismissing it,
• 23:02
  what you just described is institutional knowledge that lives
• 23:05
  in one person's head
• 23:06
  SPEAKER_01: and retires when she does.
• 23:07
  SPEAKER_00: Yeah, that's fair.
• 23:09
  SPEAKER_01: So the AI system that can document that institutional knowledge,
• 23:13
  SPEAKER_01: that can capture in this county,
• 23:14
  SPEAKER_01: under these circumstances,
• 23:16
  SPEAKER_01: check this index too.
• 23:18
  SPEAKER_01: That's not replacing her.
• 23:20
  SPEAKER_01: That's preserving what she knows.
• 23:22
  SPEAKER_00: Huh.
• 23:22
  SPEAKER_01: You're going quiet.
• 23:24
  SPEAKER_00: I'm thinking.
• 23:25
  SPEAKER_01: That's either really good or I'm about to get a three-minute objection.
• 23:28
  SPEAKER_00: No, I... That's actually the right frame.
• 23:31
  SPEAKER_00: Not AI instead of the clerk.
• 23:33
  SPEAKER_00: AI that learned from the clerk.
• 23:34
  SPEAKER_00: That's a different proposition.
• 23:36
  SPEAKER_01: And it matters because in 10 years,
• 23:38
  SPEAKER_01: half of those county clerks are going to retire,
• 23:40
  SPEAKER_01: and that knowledge walks out the door with them.
• 23:43
  SPEAKER_01: It's already happening.
• 23:44
  SPEAKER_00: I know clerks who've been in those offices for 40 years.
• 23:48
  SPEAKER_00: Their replacement is going to be someone who's been there for 18 months,
• 23:51
  SPEAKER_00: and they're not going to know about the 1987 misfiling.
• 23:55
  SPEAKER_01: So the window to capture this is now.
• 23:57
  SPEAKER_01: Not in five years.
• 23:58
  SPEAKER_01: Now.
• 24:00
  Yeah,
• 24:01
  I'll tell you what,
• 24:02
  that's the part of this that actually worries me.
• 24:05
  SPEAKER_00: Not the AI part, the clock part.
• 24:07
  SPEAKER_01: The clock is ticking on the human knowledge that makes any of this work.
• 24:11
  SPEAKER_00: Yeah, separate field report.
• 24:14
  SPEAKER_01: So, for the people actually out in the field this week...
• 24:17
  Because we can talk about 120 billion parameter models and
• 24:21
  multi-agent systems all
• 24:22
  SPEAKER_00: day,
• 24:23
  SPEAKER_00: but the question is,
• 24:25
  what does any of this mean if you're a landman in Kingfisher
• 24:28
  County this week or a
• 24:29
  SPEAKER_00: completion engineer making decisions in the Anadarko?
• 24:32
  SPEAKER_00: Right.
• 24:33
  SPEAKER_01: So here's what I think the real signal is from this week.
• 24:36
  SPEAKER_01: And then you tell me if it holds up in the field.
• 24:39
  SPEAKER_01: The Qatar situation?
• 24:41
  SPEAKER_01: The grid demand story?
• 24:42
  SPEAKER_01: The AI efficiency gains?
• 24:46
  SPEAKER_01: They're all pointing at the same thing.
• 24:48
  SPEAKER_01: The pace of decision-making in energy is getting faster.
• 24:52
  SPEAKER_01: Market signals are moving faster.
• 24:54
  SPEAKER_01: You need to respond faster.
• 24:56
  And if your internal systems can't keep up with that pace,
• 24:59
  you're leaving money on the table.
• 25:01
  SPEAKER_00: That's true.
• 25:03
  SPEAKER_00: And it's accelerating.
• 25:05
  I remember when fast in land work meant you turned around a
• 25:07
  lease acquisition in a week.
• 25:10
  Now operators are calling wanting preliminary title assessments
• 25:13
  in 48 hours on
• 25:15
  SPEAKER_00: packages that would have taken three weeks five years ago.
• 25:18
  SPEAKER_01: And that pressure doesn't go away.
• 25:20
  SPEAKER_01: It increases.
• 25:22
  SPEAKER_00: So here's the field observation.
• 25:24
  The operators who are going to win in this environment are not
• 25:27
  necessarily the ones
• 25:28
  SPEAKER_00: with the best acreage.
• 25:29
  SPEAKER_00: They're the ones who can move on the best acreage fastest.
• 25:33
  Because good acreage in a hot market that you can't act on
• 25:36
  because your title is
• 25:37
  SPEAKER_00: slow is someone else's opportunity.
• 25:39
  SPEAKER_01: So the practical takeaway for Landman right now,
• 25:42
  SPEAKER_01: and I mean this week,
• 25:44
  SPEAKER_01: is what in your workflow is slowing down your preliminary assessment?
• 25:49
  SPEAKER_01: Not your final title.
• 25:51
  SPEAKER_01: Your preliminary,
• 25:53
  SPEAKER_01: the thing that tells you this is worth pursuing,
• 25:55
  SPEAKER_01: or this isn't,
• 25:57
  SPEAKER_01: because that's where AI tools can help right now.
• 25:59
  Today,
• 26:00
  without waiting for some perfect digitization of the entire
• 26:03
  courthouse system,
• 26:04
  SPEAKER_00: The other thing I'd say to operators directly,
• 26:07
  SPEAKER_00: and I know some of you are listening because you've told me,
• 26:11
  SPEAKER_00: the data quality work we talked about last week,
• 26:14
  SPEAKER_00: that's not future state,
• 26:15
  SPEAKER_00: that's now.
• 26:17
  SPEAKER_00: If your production data has the errors we described,
• 26:20
  SPEAKER_00: if your land records are siloed in a system nobody can actually query,
• 26:25
  SPEAKER_00: that's costing you money in a supply shock environment.
• 26:28
  SPEAKER_00: Not hypothetically, actually.
• 26:31
  The operators who cleaned up their data infrastructure two
• 26:34
  years ago are the ones
• 26:35
  SPEAKER_01: who can activate capacity in 72 hours when the market calls for it.
• 26:40
  SPEAKER_01: The ones who didn't are still trying to figure out what they own.
• 26:43
  SPEAKER_00: I've seen both.
• 26:44
  SPEAKER_00: It's a real difference.
• 26:46
  SPEAKER_01: And look,
• 26:47
  SPEAKER_01: if you're listening to this and thinking,
• 26:49
  SPEAKER_01: that's actually my operation,
• 26:51
  SPEAKER_01: I don't know where my records are or if my data is trustworthy,
• 26:55
  SPEAKER_01: reach out.
• 26:56
  SPEAKER_01: Silicon and shaleagmail.com.
• 26:59
  We're talking to operators every week,
• 27:01
  and this is the kind of thing we dig into.
• 27:03
  The other field observation I want to leave people with is on
• 27:06
  the AI tool side specifically.
• 27:08
  There are a lot of vendors right now who are going to come
• 27:11
  knock on your door and
• 27:12
  SPEAKER_00: tell you their product is the answer.
• 27:14
  SPEAKER_00: And some of them are right.
• 27:16
  But the ones who are right are the ones who start by
• 27:18
  understanding your specific
• 27:19
  SPEAKER_00: workflow problem,
• 27:21
  SPEAKER_00: not by demoing a flashy interface.
• 27:23
  SPEAKER_01: A tool that solves a problem you don't have is still not useful.
• 27:26
  SPEAKER_00: Yep.
• 27:28
  SPEAKER_00: So the homework for this week,
• 27:30
  SPEAKER_00: and I'm saying this to every landman and every operator,
• 27:33
  write down the three things in your workflow that take longer
• 27:36
  than they should.
• 27:37
  SPEAKER_00: That's your AI shopping list.
• 27:39
  SPEAKER_00: You're not looking for AI in general.
• 27:41
  SPEAKER_00: You're looking for AI that addresses those three specific things.
• 27:44
  SPEAKER_01: And then actually test it.
• 27:46
  SPEAKER_01: With real data.
• 27:47
  SPEAKER_01: Your data, not a vendor's curated demo dataset.
• 27:51
  Because a lot of these tools look great on clean data and fall
• 27:54
  apart the moment you
• 27:55
  SPEAKER_00: hand them a scan from Kingfisher County.
• 27:57
  SPEAKER_01: Which is why we test things before we talk about them.
• 28:00
  SPEAKER_00: Which is more than most people can say.
• 28:03
  SPEAKER_00: Number three.
• 28:04
  SPEAKER_00: The build list.
• 28:06
  SPEAKER_01: Okay, build list time.
• 28:09
  Things I am currently in the middle of building or have added
• 28:11
  to the midnight build
• 28:12
  SPEAKER_01: list because of what we talked about this week.
• 28:15
  SPEAKER_00: Let's hear it.
• 28:16
  So first,
• 28:16
  the triage concept from the collision zone is now officially on
• 28:20
  the list.
• 28:21
  The idea of an AI system that reads county recording documents
• 28:25
  and roots them by
• 28:26
  SPEAKER_01: complexity level.
• 28:27
  SPEAKER_01: Clean, ambiguous, needs attorney eyes.
• 28:31
  SPEAKER_01: That's been simmering for a while and this week pushed it to the front.
• 28:34
  SPEAKER_01: That's midnight build number 13.
• 28:36
  SPEAKER_00: 13, keeping track.
• 28:38
  SPEAKER_01: Always keeping track.
• 28:40
  SPEAKER_01: Second thing,
• 28:41
  the multi-agent coordination problem that Carpathy and the
• 28:44
  NVIDIA stuff both
• 28:45
  SPEAKER_01: circled around.
• 28:47
  SPEAKER_01: I'm experimenting with that in the context of production data validation.
• 28:51
  SPEAKER_01: Multiple agents checking each other's work on dataset quality.
• 28:55
  SPEAKER_01: Which is a very different use case than what they're describing,
• 28:57
  SPEAKER_01: but the architecture idea translates.
• 29:00
  SPEAKER_00: The agent says mutual quality control.
• 29:02
  SPEAKER_01: That's it.
• 29:04
  SPEAKER_01: One agent flags anomalies.
• 29:05
  Another one checks whether those flags are real or artifacts of
• 29:08
  bad data entry.
• 29:09
  SPEAKER_01: A third one prioritizes which anomalies actually need human review.
• 29:15
  SPEAKER_01: Nobody reviewing everything.
• 29:17
  SPEAKER_01: Everybody reviewing the right things.
• 29:18
  SPEAKER_00: That's actually how a good quality control team works.
• 29:22
  SPEAKER_01: It's a digital QC team.
• 29:23
  SPEAKER_01: That's what it is.
• 29:25
  SPEAKER_00: Okay, I'll admit that's not a crazy idea.
• 29:27
  SPEAKER_01: High praise.
• 29:28
  And if you're listening to this and you want that built for
• 29:31
  your operation,
• 29:32
  SPEAKER_01: the data quality QC system,
• 29:34
  SPEAKER_01: the document triage,
• 29:35
  SPEAKER_01: anything we've talked about today.
• 29:38
  SPEAKER_01: Silicon and shaleagmail.com.
• 29:40
  SPEAKER_01: We build this stuff.
• 29:42
  SPEAKER_00: We do.
• 29:42
  SPEAKER_00: That's not hypothetical.
• 29:44
  SPEAKER_01: Running total on the year.
• 29:46
  SPEAKER_01: I think I'm at somewhere around 3,400 hours of manual work automated,
• 29:50
  which is up from whatever I said last week because I had a good
• 29:53
  Wednesday night.
• 29:54
  SPEAKER_00: What does a good Wednesday night look like for you?
• 29:56
  SPEAKER_01: Pizza, coffee, and a code editor open until about 2 in the morning.
• 30:01
  SPEAKER_00: That sounds like a bad Wednesday night.
• 30:03
  SPEAKER_01: We have different definitions of fun.
• 30:05
  SPEAKER_00: Clearly.
• 30:06
  SPEAKER_00: Chapter K, outro.
• 30:08
  SPEAKER_00: You know what's sitting with me from this week more than anything else.
• 30:11
  SPEAKER_00: What's that?
• 30:12
  The clock thing,
• 30:13
  the county clerks retiring,
• 30:15
  the institutional knowledge walking out the door.
• 30:19
  I've been so focused on the AI is coming side of this that I
• 30:22
  haven't thought enough
• 30:23
  SPEAKER_00: about the human expertise is leaving side.
• 30:25
  SPEAKER_00: And those two timelines need to actually meet in the middle.
• 30:29
  The window to capture what those clerks know is right now,
• 30:32
  not in five years.
• 30:33
  SPEAKER_00: Right now.
• 30:34
  SPEAKER_00: And most of the industry isn't treating it that way.
• 30:37
  They're acting like that knowledge will just still be there
• 30:40
  when they're ready.
• 30:41
  SPEAKER_00: It won't.
• 30:42
  SPEAKER_00: No, it won't.
• 30:43
  SPEAKER_01: That should probably be its own episode.
• 30:45
  SPEAKER_00: Put it on the list.
• 30:46
  SPEAKER_01: Already on the list.
• 30:49
  SPEAKER_01: Oh.
• 30:50
  SPEAKER_01: And if your company sells to operators,
• 30:52
  SPEAKER_01: landmen,
• 30:52
  or anyone in the energy space and you want to get in front of
• 30:55
  the people who
• 30:55
  SPEAKER_01: actually make buying decisions...
• 30:58
  SPEAKER_01: We're taking sponsors, silicone and shaleeggmail.com.
• 31:02
  We'll use your product before we talk about it,
• 31:04
  which is more than most podcasts can say.
• 31:06
  SPEAKER_00: That's actually true.
• 31:07
  SPEAKER_00: We will.
• 31:08
  SPEAKER_00: All right.
• 31:08
  SPEAKER_01: I have to go.
• 31:09
  SPEAKER_01: I have a build that broke at 11 last night,
• 31:12
  SPEAKER_01: and I told myself I'd fix it before noon,
• 31:14
  SPEAKER_01: and it's currently 1147.
• 31:16
  SPEAKER_00: Go.
• 31:16
  SPEAKER_01: All right.
• 31:18
  SPEAKER_01: I got to go fix a pipeline that's been yelling at me since noon.
• 31:21
  SPEAKER_00: If the county clerk's office has Wi-Fi.
• 31:23
  SPEAKER_01: It doesn't.
• 31:24
  SPEAKER_00: Never does.