Newsletter · 2026-07-25 · Ashutosh Agarwal

Autonomy Becomes a Contract as Capital Rewires Defense - Aerospace & Defense Weekly - Week of July 25, 2026

Shield AI turns military autonomy into an Air Force production contract, an ex-intelligence investor maps why cheap and adaptable now beats exquisite, and a paused cybersecurity rule reopens the factory gate. Coverage window: week ending July 25, 2026.

Aerospace & Defense Weekly

Week of July 25, 2026: Autonomy Becomes a Contract as Capital Rewires Defense

Last week the argument was about magazine depth, whether the West can physically make enough missiles, shells, and drones fast enough. This week the conversation moved on to a different, quieter question: now that the money is flowing and the drones work, who actually gets paid, and what has to change behind the scenes to let them deliver?

The answers came from an unusually mixed cast: the engineer who built the autonomy software now flying on other people's aircraft, a former intelligence official turned venture investor, an air-taxi founder pivoting into weapons, and a machine-shop owner cheering a piece of regulatory news most people never heard about. Put together, they describe a sector that has stopped debating whether cheap autonomous systems matter and started fighting over the plumbing: contracts, capital, and the small factories that turn ideas into hardware.

One note that runs through everything, because it changes how much weight to give each claim. We flag whether a speaker is a company operator (someone actually running the business and building the thing) or a pundit/analyst/investor commenting or allocating money from the outside. An executive telling you they just won a production contract is a different kind of evidence than an investor sketching a thesis. This week leaned heavily on operators.

1. Autonomy just graduated from science project to signed contract

The single most concrete development of the week came on The Road to Autonomy (July 22), in a long interview with Nathan Michael, the engineer who built Shield AI's autonomy software, called HiveMind (operator). For years, the pitch for military drone autonomy has been a promise. Michael spent the conversation explaining that it is now a delivered product with a paying customer.

Start with the scale. HiveMind, he said, is now deployed "on 30 plus platforms at this point, across air, moving into sea and even space." He was a robotics professor at Carnegie Mellon until about three years ago (more than 20 years in AI and robotics) and he first pitched the HiveMind concept to the Shield AI team back in the summer of 2017. The hard part, in his telling, was never the algorithms; it was "getting the business to a place where we could go through the full motion" and "deliver outcomes and impact… in days, if not weeks."

The proof point he offered is the kind of specific, checkable claim this sector usually lacks. Shield AI's VBAT (a medium-sized reconnaissance drone) has been flying in Ukraine "for the last two to three years," he said, and it is built with HiveMind so it can keep working in "degraded and denied conditions," meaning places "where GPS is either jammed or spoofed" and communications may cut out entirely. That matters because, as an investor noted elsewhere this week (more on her below), electronic jamming has been quietly killing Western precision weapons on the Ukrainian battlefield. Michael's headline number: when Shield AI was asked to help in Ukraine, "we were brought in within 48 hours," had the software integrated on a platform, and "within two weeks, we were running missions and operations and having effects where likelihood of success went from about 5% to about 75%." A jump from a 1-in-20 chance to 3-in-4 is the difference between a gadget and a weapon.

The genuinely new news, though, is the contract. "In June, 2026, the U.S. Air Force awarded Shield AI a production contract to provide the mission autonomy for the collaborative combat aircraft, CCA, program," Michael confirmed. Collaborative Combat Aircraft is the Pentagon's term for uncrewed jets, "loyal wingmen," that fly alongside a human-piloted fighter, get sent ahead into danger, and take on the risky work so the pilot doesn't have to. Shield AI will supply the "AI pilot" flying on board.

He also described a structural shift that investors should file away, because it decides who wins these programs long-term. The Air Force is building what it calls an Autonomy Government Reference Architecture (AGRA), a common set of standards so it can swap different companies' "AI pilots" in and out like apps on a phone. The point, Michael explained, is that the Air Force "don't want to have to have vendor lock-in. They want flexibility. They want transparency… They want to be able to bring in the best of breed… and not be fundamentally locked into a specific vendor." That is the opposite of the traditional defense model, where one prime contractor owns a program for decades. If it holds, it favors the fastest, most capable software, and it means the autonomy layer becomes a competitive market rather than a captive one.

On the fear that machines will simply be turned loose, he was careful. The model is a "teammate," with humans "on the loop" holding "decisive oversight." In a fully jammed environment where a person can't reach the drone, "these systems have to be able to think for themselves," but the design is that they "come back and… ask for permission" before pursuing a mission, rather than acting entirely on their own.

Why it matters: this is what the autonomy thesis looks like once it stops being a slide and becomes a purchase order: a real deployment record in a real war, a real Air Force production award, and a procurement structure being rewritten specifically to let software companies compete. The debate is no longer whether autonomy works. It's who owns the standard it runs on.

2. The investor's map: two "labs," one lesson: cheap, adaptable, and made with allies

For the wide-angle view, the sharpest voice of the week was Anne Neuberger on What's Your Number? (July 15). She is now head of global affairs at the venture firm Andreessen Horowitz (an investor/pundit), but she spent almost 20 years in government, 15 of them in the intelligence community, so she is describing a market she used to be the customer of.

Her central image is that Ukraine and Israel are "labs" where defense technology is being tested for real, not in theory, and that the biggest lesson is about adaptation, not raw firepower. Before Russia's invasion, she recalled sitting in the Situation Room hearing "confident briefings" that Russia's "massive, scaled military… [would] just… overrun Ukraine in a few days." It didn't happen, partly because Ukraine "scaled commercially available tech, cheap tech quickly and adapted very, very quickly." Her takeaway: "it's not only the ability to build and deploy, it's the ability to scale and the ability to quickly adapt."

She also delivered the warning that sits underneath Nathan Michael's jamming-resistant pitch above: "advanced precision targeting capabilities that came from the U.S. and its allies were disabled pretty quickly by Russian electronic warfare." In plain terms, some of the West's expensive smart weapons stopped working when the enemy jammed them, forcing a scramble to adapt. From Israel she drew a parallel lesson about the Iron Dome air-defense system, "the software and the hardware working together and adapting" to decide what a projectile is and which interceptor to use, while being candid that the system "somewhat choke[d] in the first hours of October 7th as it faced thousands of projectiles."

Her most useful passage for investors was the history of why venture money is suddenly pouring into defense. A decade ago, she said, "investment in defense tech… that was a non-starter." She described being handed, on her first day in government in 2007, a fold-out chart of the Pentagon's weapons-acquisition process so large it "covered most of my desk" (a multi-year maze with "only one set of customers"). No venture fund would touch it. What changed is that commercial technology, "satellite networks… drones and autonomy, AI," turned out to have military uses, and being "first to market and adapting" became the edge. Now, she said, venture money funds "that first level of development so the military can evaluate technologies," making the process "not only much faster, but potentially much cheaper."

She grounded the economics in a number worth remembering: because an unmanned system has "far less care abouts in terms of safety," the cost collapses. "The cost of an unmanned ship, we're talking in the single digit millions," cheap enough to "scale them much more across a long waterway." She then dropped a live example, delivered almost in passing: "the U.S. just used an autonomous ship for its first combat operations in the last 24 hours in the Straits of Hormuz." She also flagged the "mix and match" world this creates for smaller nations (noting "the UAE buying BrahMos missiles a few weeks ago from India") and argued that small, software-savvy countries (she cited the Nordics protecting undersea cables and Taiwan hardening its own resilience) can now punch far above their weight, while only the wealthiest keep buying "exquisite" one-off systems like the F-35.

One newsy aside for anyone tracking the tech-industry-meets-Pentagon friction: asked about Anthropic's clash with the government over contract terms, she argued the AI company "clearly overstepped" by trying to write rules into a contract, insisting the place "to legislate new laws… is Congress, not in a contract," and that "the military is ultimately responsible for our safety."

Why it matters: this is the investor's mental model for the whole sector in one sitting: cheap beats exquisite, adaptable beats perfect, and the money now flows before the requirement instead of after it. It's also a reminder that the West's expensive kit can be neutralized by cheap jamming, which is exactly why the jamming-resistant autonomy from section 1 is worth a production contract.

3. The air-taxi company that decided it's actually a weapons company

The most telling sign that the capital and attention have shifted toward defense came from an unexpected corner. On Squawk on the Street (July 20), reporting from the Farnborough Air Show, Archer Aviation CEO Adam Goldstein (operator) unveiled a new aircraft called Thunder, built in partnership with the defense startup Anduril.

Archer has spent nearly a decade and a lot of money building an electric air taxi (a small, quiet aircraft that takes off vertically like a helicopter; the industry calls these eVTOLs). Goldstein's pitch was that the same technology now has an obvious second life. Thunder, he said, "is actually a dual-use platform that will have both a defense application as well as a civil application." And the reason the defense version can go first is a regulatory quirk worth understanding: a military aircraft doesn't need the years-long civilian safety certification that a passenger air taxi does. "This is an autonomous aircraft… that has the ability to be fielded when the aircraft can show that it can meet the performance," he said, so "you can fast-track a lot of these platforms." The line that captures the whole pivot: companies like Archer, he argued, "won't be looked at as just an air taxi company anymore, but actually a diversified A&D [aerospace and defense] company."

He was refreshingly blunt about the skepticism eVTOLs have earned after years of hype. "Hardware is hard," he said. "Investing in hardware and aviation is not for the faint of heart." On the original air-taxi dream, he stuck to the ambitious goal of operating by the 2028 Los Angeles Olympics, helped by a government pilot program (the eVTOL Integration Pilot Program) letting the industry fly demonstration routes around Texas, Florida, and New York to build public and regulator trust. But even he conceded the timeline is "ambitious," and it was clear which half of the business he thinks pays off first.

Why it matters: when a company that spent ten years chasing flying taxis reintroduces itself as a defense contractor, that tells you where the demand, and the patient capital, has gone. The Anduril tie-up is also another data point in this year's recurring story: the fast-moving newcomers are teaming up to muscle into markets the old primes once owned.

4. The quiet policy reversal that just widened the factory gate

The least glamorous story of the week may be the most consequential for the industrial-base problem that has dominated this sector all month. On my TRUE POSITION (July 18), a working machine-shop owner (operator) walked through a piece of Pentagon news that never made the front page but reshapes who gets to build America's weapons.

The news: on July 13, 2026, the Department of Defense suspended the rollout of "CMMC Phase 2." CMMC (the Cybersecurity Maturity Model Certification) is a program that would have required defense suppliers to pass an independent cybersecurity audit before winning certain contracts. Phase 2 was due to begin November 10. It's now on hold.

Why pause a cybersecurity rule? Because, as the host put it, the government "cannot rebuild America's arsenal while simultaneously burying the small manufacturers needed to build it under six figures of compliance costs." The dollar figures explain the panic. The government's own estimate was roughly $101,752 for a small business to get through a single Level 2 assessment, and about $104,670 across the three-year cycle, and that assumes the shop has already installed all the required security controls. In the real world, the host said, his own shop has been getting consultant quotes "from as little as 20 grand to as much as 400 grand," before hardware, secure servers, cloud environments, and staff time. For "a 25-person machine shop," where the same handful of people already handle quoting, quality, purchasing, and payroll, that is often unaffordable, so many were simply quitting defense work.

His framing is the one to take away: "It is not simply a cybersecurity story. It's a capacity story. It's a supply chain story. It's an acquisition speed story." The country is trying, all at once, to replace weapons sent to Ukraine, expand missile and ammunition output, meet NATO stockpile requirements, and build new drones and satellites, and, in his words, critical shortfalls in "explosives, energetic materials, castings, forging, electronics, motors, and specialty components" have "increased significantly" in government analysis between 2019 and 2023. You cannot, he said, "surge a supply chain that barely exists," and "just because somebody gives you a multi-million dollar purchase order… does not instantly create a qualified supplier."

He was careful about what did not change. This is "not… a get-out-of-jail-free card": the underlying data-protection rules (NIST 800-171, DFARS) still apply, and the big prime contractors will keep demanding security questionnaires, secure file exchanges, and incident reporting regardless. His read on who wins in a suddenly more crowded field: the pause "opened the floodgates to a lot of new competition," which will squeeze pricing on "lower complexity… price sensitive components," so established shops should "keep the high end stuff." The durable advantages, he argued, are the unglamorous ones (quality certifications, export-control registration, disciplined inventory systems, traceability, and "financial stability") because when the regulatory gate gets lower, buyers "need other ways to separate legitimate suppliers from opportunists."

Why it matters: this is last month's "America forgot how to run a factory" anxiety being addressed with an actual policy lever, and it's a rare piece of good news for the thousands of small shops that are the hidden bottleneck in every rearmament headline. It also quietly hands an edge to the disciplined, well-run suppliers over the fly-by-night ones now rushing back through the wider gate.

5. One number for the capital-flow watchers

Worth logging for anyone tracking where defense dollars land: on The Rundown (July 24, pundit/journalist), the host flagged that the Pentagon announced a software deal with Oracle worth up to $7 billion over the next decade, covering "software licenses, maintenance, and consulting" for government data centers serving the U.S. intelligence community and the Coast Guard. The important caveat: "this isn't necessarily $7 billion of new business." It mostly consolidates existing Oracle contracts into one agreement, which the Pentagon says will "save taxpayers at least $441 million over the life of the contract." The context that makes it matter to Oracle: the stock has been "getting hammered this year," down "around 38%" on worries about the company's debt and its older software being disrupted by AI, so locking in a long-term government anchor is a bright spot on an otherwise rough tape for the shares. (Fun fact the host couldn't resist: the CIA was Oracle's very first customer.)

Why it matters: defense spending isn't just missiles and drones. The back-office software that runs the intelligence community is a large, sticky, recession-proof revenue stream, and, as with the hardware primes, it's a market where incumbency and existing footprint still count for a great deal.

The bottom line

If last week was about whether the West can make enough hardware, this week was about the machinery that decides who gets to. The throughline is that autonomy has crossed from promise to procurement: Shield AI's software is flying in a real war and now has an Air Force production contract, the Air Force is rewriting its rules so autonomy becomes a competitive market instead of a captive one, and even a flying-taxi startup is rebranding as a weapons company to get in. The investor's map, drawn by someone who used to sit on the government side of the table, says the same thing from 30,000 feet: cheap and adaptable is beating expensive and perfect, and money now moves before the requirement, not after it. And the least glamorous story, a paused cybersecurity rule, may do more for actual production than any of it, by letting the small shops that forge the parts back through the gate. The demand was never the question. This week was about the plumbing, and the plumbing is finally getting fixed.