Tech bros optimized war… and it’s working

0:00 Yesterday, the US Department of War

0:02 announced it's going allin on a new

0:04 primary operating system for the

0:06 battlefield.

0:07 A tool that's proven to be

0:08 so effective at blowing people up that

0:10 it's rolling out to every branch of the

0:11 military.

0:12 The Army, Navy, Marines, Air

0:14 Force, and even the Space Force are

0:16 going to be powered by the Maven Smart

0:18 System, an AI platform that shortens the

0:20 kill chain for kinetic operations.

0:22 You heard that right.

0:23 The same AI models

0:24 that can't spell strawberry are now

0:26 being used to turn people into a fine

0:28 mist faster than ever.

0:29 That might sound inhumane and terrifying, but don't

0:32 worry, there is still a human in the

0:33 loop who needs to click the accept all

0:35 cookies button before the missiles can

0:37 be launched.

0:37 The Peter Teal's Palanteer

0:39 is the main company behind Maven, but

0:41 all the big hyperscalers and AI labs are

0:43 cashing in on the taxpayer funded US war

0:45 machine.

0:46 In today's video, we'll take a

0:47 look at the future of war and find out

0:49 how prompt and destroy slop ops actually

0:51 work at a low level.

0:52 It is March 24th,

0:54 2026, and you're watching the Code

0:56 Report.

0:56 In Modern Warfare, you can't

0:58 just carpet bomb an entire city.

0:59 Instead, you need to first identify a

1:01 target, then verify it, then verify it

1:04 again.

1:04 Otherwise, you might end up

1:05 killing a school full of innocent

1:07 children.

1:07 Very bad intelligence.

1:09 I'm sorry.

1:10 And that's where the Maven Smart System

1:11 comes in.

1:12 It's an AI platform that uses

1:14 computer vision and sensor fusion to

1:16 automatically analyze surveillance data

1:18 like drone footage, then identify,

1:20 track, and prioritize targets.

1:22 There's still a human to push the kill button

1:24 today, but eventually this process could

1:26 become entirely autonomous.

1:27 Now, before we look at the technical details behind

1:29 the system, we first need to meet the

1:31 tech bros who created it.

1:32 The core platform is provided by Palunteer.

1:34 The current CEO is Alex Karp, and it

1:37 provides the operating system that glues

1:38 everything together.

1:39 We've got AWS and

1:40 Azure helping out with cloud

1:42 infrastructure.

1:42 And Google used to be

1:43 involved, too, but they had to back out

1:45 after their hippie employees started

1:47 protesting.

1:47 But the Maven system needs

1:48 tons of real world data and it gets much

1:50 of that data from Palmer Ly's Anderil

1:53 which provides terrifying kill machines

1:55 like the ghost drone, Amble Interceptor

1:57 and Ghost Shark underwater drone.

1:59 And then finally the system runs on multiple

2:01 large language models until recently

2:03 anthropics Claude was their champion.

2:05 But then their soyo Daario started

2:08 crying when he found out that his tech

2:09 might be used to harm humans.

2:11 War Chad Pete Hegsth, who can bench press 315

2:14 pounds, by the way, drank Daario's tears

2:17 and banned Anthropic as a national

2:18 security threat from all government

2:20 contracts.

2:21 Luckily, Sam Alman was happy

2:22 to step in for sloppy seconds.

2:24 And the web of people here goes way deeper, but

2:26 as a developer, I'm more interested in

2:28 how Project Maven actually works under

2:30 the hood.

2:30 The exact tech stack is

2:32 classified, but we have enough public

2:33 data and leaks to piece together a

2:35 similar system with open- source

2:37 software.

2:37 At the first layer, we need to

2:39 ingest tons of data in different formats

2:41 like video streams from our drones,

2:43 ecoms from our special ops teams, GPS

2:45 from our satellites, and so on.

2:47 And to do that, we're going to use a tool like

2:49 Apache Kofka.

2:50 But basically, Kafka allows us to stream multiple data

2:53 sources in one place, allowing this

2:55 entire complex system to stay updated in

2:57 real time.

2:57 And now that we have incoming

2:59 events, we can use a tool like Apache

3:01 Spark to subscribe to a Kafka topic.

3:03 So, we can start transforming that data into

3:05 something useful.

3:06 Like, we might send

3:07 drone footage to OpenCV to segment it

3:10 and detect actual objects in those

3:12 images.

3:12 But now, here's where things get

3:13 really interesting.

3:14 In order for AI to

3:15 blow people up, it needs to understand

3:17 the relationships between all the

3:19 different resources in our system.

3:20 At Palunteer, their secret sauce is called

3:22 the ontology, and the government is

3:24 paying them billions of dollars a year

3:25 to use it.

3:26 But what is it?

3:27 Well, basically it maps messy fragmented data

3:30 from different sources into a shared

3:31 structure while capturing the metadata

3:33 and relationships between these objects.

3:36 You can think of it like a digital clone

3:37 of an entire organization, which might

3:39 be a manufacturing plant, a hospital, or

3:42 in this case, the military.

3:43 At this point, we have data, but we don't

3:45 understand the relationships between our

3:46 data points.

3:47 Ironically, we won't use a

3:49 relational database here, but instead a

3:51 graph database like Neo4j, where people,

3:53 vehicles, and bombs become nodes, and

3:55 their movements turn into edges.

3:57 And now our entire battlefield is mapped in a

4:00 way that replicates the real world,

4:01 where it can be queried and visualized

4:03 by humans and AI.

4:04 And now that our

4:05 world's recreated, we need to set some

4:06 ground rules before we start taking

4:08 action.

4:08 A tool like Open Policy Agent

4:10 could help us do that by enforcing

4:12 policies across the entire stack.

4:13 It looks good to me.

4:14 Now we can start

4:15 dropping in AI agents with the model

4:17 context protocol.

4:18 From here you can grab

4:19 your favorite open Chinese model like

4:21 Kimmy or Quen.

4:22 Then use Heretic to

4:23 uncensor it.

4:24 And now it can start

4:25 performing actions based on this data.

4:27 But from there you just need to wire up

4:28 some tomahawk missiles and you'll be

4:30 blowing people up on pure vibes in no

4:32 time.

4:32 And what's really crazy is that

4:34 you don't need a trillion dollar defense

4:35 budget to build this thing thanks to

4:37 tools like Tracer, the sponsor of

4:39 today's video.

4:39 It's a spec driven

4:40 development tool that lets your whole

4:42 team work together with agents to build

4:44 real world software.

4:45 Just start by telling it what you want to build, and

4:47 Tracer's epic mode will ask you

4:49 follow-up questions to create a series

4:51 of specs and tickets that map out to

4:53 your requirements.

4:54 But from there, you

4:54 can invite your teammates to the project

4:56 and live edit the specs together.

4:58 It then assign people to specific tickets.

5:00 A tracer then passes all of that context

5:02 to your favorite coding agent and

5:04 validates the output along the way to

5:06 make sure it actually meets your

5:07 requirements without drifting.

5:09 So if your team wants to use agents to ship

5:11 software that actually works in

5:12 production, then try Tracer for free at

5:14 the link below.

5:15 This has been the code

5:16 report.

5:16 Thanks for watching and I will

5:18 see you in the next one.