Visor Dynamics · Geospatial Intelligence
All articles
Strategy 11 min read

Is mosaic warfare already here?

DARPA imagined a force of cheap, interchangeable pieces. Ukraine, Iran and the Pentagon suggest much of it is here, except the software to hold it together.

In our last article, Collapse the OODA loop, we looked at how militaries are trying to observe, orient, decide and act faster than their opponents. Mosaic warfare takes that idea a step further. Instead of one fast loop run by one big force, it imagines hundreds of small ones, run by pieces that can be lost, replaced and recombined in the middle of a fight.

When DARPA proposed the idea, it sounded like something for the 2040s. The news from the past six months suggests a lot of it is already happening.

Puzzle pieces and mosaic tiles

The concept came out of DARPA's Strategic Technology Office around 2017. Tom Burns, who ran the office at the time, put the problem this way: most weapon systems today are "more like pieces of a puzzle than tiles for a mosaic." A puzzle piece is cut to fit one place. If it goes missing, the picture has a hole in it, and nothing else fits there.

A mosaic is built from tiles that are simpler and largely interchangeable. A commander picks sensors, shooters, jammers, decoys and relays, crewed or uncrewed, and arranges them for the job at hand. When some are destroyed, others fill the gap. Burns summed up the goal as being able to "mass your firepower without having to mass your forces." With no aircraft carrier or headquarters at the center, there is no single target whose loss breaks the whole force.

In 2019 a Mitchell Institute study by David Deptula, Heather Penney, Lawrence Stutzriem and Mark Gunzinger laid out what this could mean for the US military, calling for a highly spread-out "kill web" with as few high-value nodes as possible. Later work by Bryan Clark and Dan Patt, now at the Hudson Institute, framed it as decision-centric warfare. The idea is to win less by destroying the other side's forces and more by giving its commanders so many problems at once that they can't keep up. In OODA terms, it aims to overload the other side's Orient step faster than they can clear it.

Chains become webs

A traditional kill chain is a fixed sequence: a particular satellite finds the target, a particular headquarters approves it and a particular aircraft strikes it. If any of those links is lost, the whole chain stops.

A kill web is a network where any sensor can cue any shooter through whatever path is still working. If the satellite is blinded, a drone or a ship's radar can take its place. If one launcher is destroyed, another one picks up the mission.

DARPA built early software for this. Its ACK program (Adapting Cross-Domain Kill-Webs) searched thousands of possible combinations of assets and recommended which ones to task. STITCHES linked systems that were never designed to talk to each other, without upgrading them. Both were used in a US Air Force battle management demonstration in 2020, which showed the concept could work in a controlled test. Doing the same thing at scale, while being jammed and shot at, is much harder.

Where the tiles are already on the board

Ukraine and Russia: a front line run by drones

The war in Ukraine is the closest thing to a working mosaic so far, on both sides of the line. Analysts cited by NPR in October 2026 estimate that drones now cause around 80% of casualties. Both armies have built a "kill zone" roughly 10 to 20 miles deep, where almost anything that moves can be seen and hit within minutes. Troops stay underground or spread out in small groups, and very few people hold the front line itself.

What makes this a mosaic rather than just a lot of drones is the way the pieces connect. Ukraine's Delta system shares sensor data, AI detections and target information across units, so a drone team can act on what another unit's camera saw. An August 2026 drone report described this as a horizontal kill web and contrasted it with the slower, more hierarchical process in most Western armies. Russia has adapted in the same direction, with its own drone units, interceptors and electronic warfare. Each side keeps reorganizing its tiles in response to the other.

Iran: command without a center

The biggest 2026 story under the word "mosaic" came from Iran. Its Islamic Revolutionary Guard Corps has been developing a doctrine called mosaic defence since about 2005. Under former commander Mohammad Ali Jafari, command was split into 31 provincial units, each with its own headquarters, communications, weapons and intelligence. Each critical post reportedly has a list of chosen successors.

This doctrine was tested in the war that began on 28 February 2026, after strikes killed Iran's Supreme Leader and many senior commanders. Iran's foreign minister said the system had been activated and that units were now acting on their own, following instructions issued in advance. Farzin Nadimi of the Washington Institute observed that units cut off from Tehran could still function as a cohesive force.

Analysts have drawn mixed conclusions. An assessment by Israel's Jerusalem Institute for Strategy and Security found that the doctrine prevented the institutional collapse many expected, but it also reduced strategic coherence and made the conflict harder to control and to wind down. IISS analysts were more cautious about how long it would hold.

Iran's version applies the mosaic idea to command structure rather than to weapons, so it isn't the same thing DARPA described. It does show the trade-off at the heart of the concept, though: a force without a center is very hard to knock out, and for the same reason it can be hard to steer or to stop.

The United States: a command built for autonomy

In the United States, the idea is turning into institutions and budgets:

  • On 30 September 2026, Defense Secretary Pete Hegseth announced an Autonomous Warfare Command, with a four-star commander and a target start date of 1 October 2027, subject to approval by Congress. An interim effort called Project Agincourt, led by Defense Innovation Unit director Owen West, is preparing the way. Hegseth said the lesson from Ukraine was the need for a "purpose-built system that maximizes the ability to adapt."
  • At the AUSA conference on 14 October, Hegseth said the Pentagon plans to field hundreds of thousands of one-way attack drones this year and millions in the years after.
  • The Defense Autonomous Warfare Group, which replaced the Replicator drone program, is being moved into a new task force under Special Operations Command. Its fiscal 2027 request is $54.6 billion, up from about $226 million. Its own chief has said he isn't sure how much of that will arrive.
  • In June, the Air Force awarded the first production contracts for Collaborative Combat Aircraft, the FQ-42A from General Atomics and the FQ-44A from Anduril. These uncrewed jets are meant to fly alongside fighters as extra sensors, shooters or decoys. The plan is for more than 150 by the end of the decade and around 1,000 over time.
  • In the Pacific, INDOPACOM's "Hellscape" concept would fill the Taiwan Strait with thousands of uncrewed air, surface and underwater systems in a crisis, and Taiwan has approved a large drone program of its own. Analysts disagree about whether expensive long-range drones or cheap short-range ones suit the plan better.

China: build a swarm, and plan to break one

China is working on both sides of the problem. In March 2026, a CETC demonstration showed one command vehicle controlling 96 fixed-wing drones launched seconds apart. In October, Chinese state television showed layered counter-drone defences using lasers, high-power microwaves, guns and missiles, designed to thin out a swarm before it arrives.

The People's Liberation Army's operational thinking centres on what analysts call systems destruction warfare: paralyse the opponent by hitting the key nodes that hold its forces together, using cyber attacks, electronic warfare, space assets and long-range strikes. Mosaic warfare is designed to have no key nodes. In our view, that makes the two ideas a direct test of each other. A US Air Force-affiliated research institute notes there is still little public evidence of PLA units using swarms outside testing, so how either approach would hold up against the other is an open question.

What's still missing

DARPA's full vision is more than lots of drones. It is a commander composing a fresh mix of tiles across air, land, sea, space and the spectrum, with software suggesting the combinations and adjusting them as the fight changes. That part isn't here yet, and most of the gaps are in software.

Someone has to arrange the tiles

Replicator, the Pentagon's earlier push for mass drones, reportedly struggled most with the software to command many different types of drone together. Its successor is tackling that directly. In January 2026, DIU, the Navy and the Defense Autonomous Warfare Group launched a $100 million prize for an "Autonomous Vehicle Orchestrator": software that takes a commander's intent, spoken or typed, and turns it into tasks for mixed fleets from different manufacturers, without touching each vehicle's own autonomy software. The Air Force is in a similar position with its new combat drones, and won't choose the main autonomy software provider for them until 2027. So the hardware is already on order, while the software that is supposed to tie it together is still a competition.

A million drones means a million video feeds

Every one of those drones is also a sensor. When militaries operate drones by the million, the amount of video they produce will be far beyond what any team of analysts can watch, and much of its value is lost if nobody can find it later.

Militaries will need systems like our Elite SPRINT to keep up: software that takes in live drone video, places it on the map as it arrives, records and indexes all of it, and archives it so that footage from any place and time can be found in seconds instead of being scrubbed through by hand. Otherwise most of what the drones record will never be looked at.

The web needs a network

A kill web depends on its links. A US Naval Institute Proceedings article, "Fighting When the Network Dies," argues that no force can count on high-bandwidth communications in a serious war, and that backups like HF radio carry far less and are easier to locate. RUSI points out that jamming affects your own communications as well as the opponent's. Ukraine shows how quickly electronic warfare changes which drones work on a given day. A mosaic has to keep working, in some degraded form, when large parts of the network are cut.

Tiles that don't fit together

Buying lots of cheap systems quickly from many suppliers produces a literal mosaic of equipment that may not share formats, protocols or symbology. RUSI warns of exactly this with electronic warfare kit. Common standards are what let one country's sensor cue another unit's launcher. They are also the dullest part of the whole concept, which is probably why they tend to be left until late.

Hard to stop

Iran's experience in 2026 raises a question that applies to every version of the concept. If authority is pushed down to many small units so the force survives losing its leadership, those units are also harder to recall when leaders want to pause, negotiate or de-escalate. As autonomy spreads, the same question applies to machines. We think the ability to call off an operation needs to be designed into these systems from the start, because it is very hard to add once thousands of them are deployed.

So, is it here?

Partly. Cheap drones are in service by the hundreds of thousands, Ukraine and Russia are fighting through something close to a kill web, Iran's command survived the loss of its leadership, and the US is building a command and production lines around autonomy. Almost every military now accepts that its forces should be spread out and hard to decapitate.

The part DARPA cared most about hasn't arrived yet. Nobody has shown a commander reliably assembling mixed drones, aircraft and sensors from different makers into one force on the fly, keeping it working while the network is jammed, and pulling it back when needed. Most of what stands in the way is software and data rather than hardware, and we expect that to be where militaries spend their effort over the next few years.

References

More articles

View All Articles
AI

Is a military AI agent just a chatbot with tools?

Building an AI agent now takes minutes. Building one a military can trust with real missions is a different problem, and most of the hard parts have little to do with the model.
Engineering

Would you trust vibe-coded defense software?

AI can write code faster than anyone can read it. In defense, that raises hard questions about security, accountability and who can still fix the system when it fails.
Strategy

Collapse the OODA loop

Whoever observes, orients, decides and acts faster than the opponent sets the pace. Where each step slows down, and how software is shortening the loop on the battlefield and in business.

© Visor Dynamics (Pvt) Ltd.

Visor Dynamics · Geospatial Intelligence