The editorial argues the shift from $30M Reapers to $1-3M attritable drones is not incremental optimization but a different category of weapon entirely — a plane you are allowed to lose. This reframes defense acquisition around the Replicator model, where quantity and replaceability beat exquisite bespoke platforms.
Draws a direct analogy between the Reaper and legacy mainframe computing: exquisite, artisanal, maintained by a priesthood, and unsustainable the moment the environment turns adversarial. Iran destroying roughly $1B in Reapers with cheap air defenses proves the economics no longer work.
By surfacing the Ars Technica piece framing Iran's destruction of $1B in Reapers as the catalyst for the Pentagon's pivot, the submitter implicitly endorses the view that expensive hunter-killer drones have become uneconomical against contested airspace.
Argues that Anduril's Lattice, Shield AI's Hivemind, and Skydio's autonomy stacks matter more than any specific airframe. The value migrates up the stack to fleet-management, sensor-fusion, and orchestration software — making these firms the new primes-in-waiting, displacing General Atomics-style incumbents.
The Pentagon has spent the last two years watching a $30-million-per-unit hunter-killer platform get shredded by air defenses that cost less than a mid-sized SaaS Series B. Iran alone is credibly reported to have destroyed on the order of a billion dollars in MQ-9 Reapers — the exact tally is fuzzy, but the arithmetic isn't: roughly a squadron's worth of aircraft, each one a bespoke, decade-in-the-making General Atomics artifact, taken out by systems the operator can replace in an afternoon.
The Air Force's response, telegraphed in this week's reporting and in RFPs that have been circulating for months, is to buy something cheaper. Not incrementally cheaper. An order of magnitude cheaper. The target unit cost being floated — somewhere between $1M and $3M per airframe — is not a Reaper with the extras stripped out; it is a different category of thing entirely. It's a plane you are allowed to lose.
That word — *attritable* — is the whole story. It has been quietly restructuring defense acquisition since the Replicator initiative kicked off in 2023, and it is the reason companies like Anduril, Shield AI, and Skydio are being treated by the DoD less like scrappy upstarts and more like the new prime contractors in waiting.
If you have spent any time in software, the pattern here should feel almost aggressively familiar. The Reaper is a mainframe: exquisite, artisanal, maintained by a priesthood, and economically ruinous the moment the environment gets adversarial. The attritable drone is a commodity Linux box. You buy a thousand of them, you assume some percentage will fail on any given day, and the value is not in the hardware — it is in the orchestration layer above it.
This is why the vendor list matters more than the airframe specs. Anduril's pitch has never really been about their airframes; it's about Lattice, their fleet-management and sensor-fusion software. Shield AI leads with Hivemind, an autonomy stack. Skydio's drones are Skydio's drones because of the perception software, not the props. The Pentagon is not buying planes anymore. It is buying deployment targets for an autonomy runtime, and every incumbent that built its moat around airframe IP is now selling the wrong product.
Compare the incentives. A traditional prime contractor makes its money on cost-plus procurement, which rewards making the thing more expensive, slower to build, and harder to replace. That model produced the F-35, the Zumwalt, the Littoral Combat Ship, and a Reaper program that has been iterating on the same basic airframe since 2007. It cannot produce a $1M drone because its entire cost structure — engineering hours, compliance overhead, vertically integrated supply chains — is priced for a $30M drone. When people talk about "defense procurement reform," this is what they mean: the cost-plus doctrine is a business model, and the business model has to die before the airframes get cheap.
Community reaction on HN captures the tension well. One side notes, correctly, that a $2M drone with weaker sensors and shorter loiter time is not actually a Reaper replacement for every mission — it's a different tool. The other side notes, also correctly, that if you can put ten of them in the air for the price of one Reaper, mission planning changes shape entirely. Both are right. The strategic question isn't 'can attritable drones replace Reapers'; it's 'what does a mission look like when the marginal aircraft is nearly free?' That is the same question cloud infra forced on ops teams a decade ago, and the answers rewrote the org chart.
If you build software and you have watched this space with the polite disinterest most devs reserve for defense news, this is the moment to update your model. Defense is quietly becoming one of the largest software markets in the world, and it is buying the same primitives you already ship: fleet orchestration, computer vision at the edge, low-latency messaging, OTA updates, sim-to-real ML pipelines, and observability that survives being shot at. The DoD's Replicator target is thousands of autonomous systems fielded by 2025-2026, and the software side of that number is dominated by problems that look identical to the ones a self-driving startup or a warehouse robotics team is solving — with the difference that the deployment environment is genuinely adversarial and the safety envelope is inverted.
Practical implication: the talent market is about to bend. Anduril has been hiring aggressively out of the same pool as Waymo, Cruise, and the surviving autonomy startups, and the compensation numbers are competitive with FAANG for perception, planning, and infra roles. If you have shipped anything meaningful in robotics, edge ML, or distributed systems, you are now a defense candidate whether you sought that or intended it. That's a values decision, not a technical one, but pretending the option doesn't exist is no longer accurate.
Second implication: the open-source surface area is going to grow. Attritable doctrine only works if the software stack is portable across airframes — the whole point is that any given plane is disposable. That pressures the ecosystem toward shared autonomy stacks, standardized message buses (PX4, MAVLink and their descendants), and simulation environments that don't lock to a single vendor. Expect the next few years to produce the defense equivalent of Kubernetes: an ugly, committee-designed, industry-consortium abstraction that everybody complains about and everybody adopts.
The Reaper losses in Iran are being framed as a procurement failure, but that's the small version of the story. The larger version is that the exquisite-platform doctrine has been quietly bankrupt for a decade and it took a live-fire object lesson to force the org chart to admit it. The winners of the next decade of defense contracting will not be the companies with the best airframes; they will be the companies whose software runs on the most airframes. That's a familiar shape. The industry that just watched it happen was ours.
There are ~350,000,000 of us. When I read we spent $1B, I think about how I'm responsible for $3 of that. It doesn't matter considering the ~$117,550 of the national debt I'm responsible for. It palls compared to the $3,000 a year in interest towards the national debt I'm respons
> The Defense Innovation Unit notice called for drones capable of carrying many different sensor and weapons payloads up to 2,800 pounds and flying with a combat radius of at least 2,300 nautical miles—or 8,000 nautical miles on a one-way strike mission—while executing the same missions that the
I’m no military expert by any means but US appears to be obsessed with destroying some super important target to win, like they did with killing Iran leaders only to find out that new leader replace the perished.The same with the other stuff, they have super important radar and super important ships
The defense industry has spent the last 40+ years grooming the DoD into thinking it costs $30mil/unit to produce missiles and drones. They should have rejected any of the bids, but being fueled by massively excessive taxes in the USA, they don't have to answer to any sort of efficiency or
Top 10 dev stories every morning at 8am UTC. AI-curated. Retro terminal HTML email.
I worked on the control systems for Predators and Reapers back in the mid and late 00s, and the inefficiencies around process were enormous. Safety is extremely important, so you expect some slowness as a result, but it got pretty extreme. I remember one time having to do 6 weeks of testing around a