Autonomous flight is software that flies an aircraft from taxi to touchdown with no hands on the controls, and it is leaving the laboratory faster than aviation's rulebook can keep up.
The company making the loudest case for it right now is Merlin Labs, a Boston firm that went public on NASDAQ this March and is teaching a military C-130J transport to fly itself. Merlin is not the only one. The contest is a race to certify, not merely to fly, and the gap between the lab and the runway is closing fast.
From cargo runs to the C-130J: what Merlin Labs does
Merlin builds one product, the Merlin Pilot — an autonomy system designed to be aircraft-agnostic, which means the same core software can be fitted to different aircraft rather than rebuilt from scratch for each one. Matt George founded the company in 2018 with a plain goal: take the pilot's workload, and eventually the pilot, out of the cockpit.
Merlin started where the economics were friendliest, which is cargo. Merlin ran the Merlin Pilot on the Cessna 208 Caravan, a single-engine workhorse freighter, and flew test campaigns out of Kerikeri in New Zealand and bases in Rhode Island and Massachusetts. By 2024 the company had clocked more than a hundred hours of autonomous flight and begun the slow grind of certification testing with regulators — the part that actually decides whether any of this reaches commercial service.
Then the money and the mission got bigger. Merlin won a U.S. Special Operations Command contract worth up to $105m to bring autonomy to the C-130J Super Hercules, the four-engine tactical transport that special forces fly. In March 2026 Merlin cleared the programme's Preliminary Design Review, the engineering gate that lets it move into detailed design and, later, real takeoff-to-touchdown demonstrations. The aim is not an empty cockpit on day one but "reduced aircrew": fewer people doing more, with the software carrying the load.
Completing the PDR is a pivotal milestone for our C-130J program, bringing us closer to delivering highly assured autonomy that elevates mission performance, enhances safety, and expands operational flexibility for Special Operations Forces.
— Matt George, founder and CEO of Merlin Labs
The others racing to empty the cockpit
Merlin's clearest rival is Reliable Robotics, founded by Robert Rose, who led flight software for SpaceX's Falcon 9 and Dragon. Reliable has taken a parallel path: in November 2023 Reliable flew a Cessna 208B Caravan through a complete uncrewed flight — taxi, takeoff, cruise and landing — while a pilot supervised from a control room fifty miles away.
Reliable has since raised about $300m, won a $17.4m U.S. Air Force contract, and is assembling what it calls a "mountain of evidence" for the Federal Aviation Administration to certify an uncrewed Caravan under existing rules, with no special exemptions. Whoever clears that bar first writes the template everyone else follows.
Others have folded their bets into bigger players. Xwing, which flew the world's first fully autonomous gate-to-gate flight and logged 250 autonomous flights with its Superpilot software, sold its autonomy division to the air-taxi company Joby Aviation in June 2024. The talent and the code did not disappear; they moved up the food chain.
And the field is wider than cargo. California's Skyryse is attacking a different layer, simplifying the controls themselves so that one system can fly many aircraft types with far less training. The common thread across all of them is the same: the hard problem is not making a plane move on its own. The hard problem is proving it is safe enough to trust.
When the autonomous aircraft is a fighter jet
The most striking demonstration did not come from a startup at all. The breakthrough came from DARPA, the Pentagon's research arm, and its Air Combat Evolution programme. In September 2023 an AI agent flew a modified F-16 called the X-62A VISTA in a real within-visual-range dogfight against a human-piloted F-16 — the first time that had happened in the air.
The numbers are sobering. The team pushed more than 100,000 lines of flight-critical software across 21 test flights. The two jets closed to within 2,000 feet at 1,200 miles per hour. A safety pilot sat in the X-62A the whole time with a switch to kill the AI, and never had to use it. Officials would not say who won the dogfights, citing national security.
That work has a commercial heir in Shield AI, whose Hivemind "AI pilot" grew out of the same X-62 lineage. In February 2026 the U.S. Air Force picked Hivemind to fly the YFQ-44A, one of its Collaborative Combat Aircraft — the uncrewed wingmen meant to fly alongside crewed fighters. Shield AI has raised on the order of $2bn to scale Hivemind.
💡The autonomous-flight field at a glance
Five players, one race. • Merlin Labs: aircraft-agnostic autonomy, from cargo Caravans to the military C-130J. • Reliable Robotics: an uncrewed Cessna Caravan flown by a pilot on the ground. • Joby Aviation: Xwing's Superpilot autonomy, now inside an air-taxi maker. • Skyryse: simpler controls so one system flies many aircraft types. • DARPA and Shield AI: combat autonomy, from the X-62A dogfight to the Hivemind-flown YFQ-44A.
The real question is not whether the plane can fly itself
Here's what matters. Across every one of these programmes, the engineering is no longer the bottleneck. The machines can fly. The bottleneck is trust, and trust in aviation has a specific institutional form: certification. You cannot certify what you cannot explain, and you cannot explain a system you cannot audit. That is why Reliable Robotics talks about a "mountain of evidence" and why Merlin spends years on design reviews. The flying is the easy part.
Look closely and you see a spectrum of how much agency we are willing to hand over. Merlin's C-130J keeps a reduced crew in the cockpit. Reliable keeps a pilot on the ground. The X-62A removed the human from the decision loop entirely and trusted the AI to fight. The spectrum — human in command, human on the loop, human out of the loop — runs through every serious debate about Emergent Intelligence, except here the choice plays out at 1,200 miles per hour with live ordnance.
I have argued before that intelligence becomes real when it has to act in a body — that a mind without a body is only half a mind. An autonomous aircraft is exactly such a machine: Emergent Intelligence in a seventy-tonne body, moving through weather and traffic and consequence. And the aircraft raises the question I keep returning to. When the software is flying the plane, who is accountable when it goes wrong? The dignity-first answer is that accountability must never evaporate into the code. Someone owns the decision, even when no one is touching the controls.
There is a defence dimension that deserves naming. When Anthropic drew a line at certain military uses of its models, Anthropic was wrestling with the same question these aviation programmes answer in the affirmative: autonomy is going straight into the cockpit of a gunship and the wing of a fighter. Pretending the milestone is neutral helps no one. The decision is about what kind of intelligence we build, and what we let it do — a point I have made about treating these systems as neither ghost nor machine.
The planes can fly themselves. The unfinished work is human: the rules, the liability, and the decision about how much of the cockpit, and how many pilots, we are ready to give up. Merlin Labs and the others have made the machine the easy part. The hard part is us.
Frequently Asked Questions
These are the questions readers have been asking since Merlin Labs went public. Short answers follow, drawn from company filings, regulators, and the flight-test record.
What is autonomous flight?
In short, autonomous flight is software flying a plane without a human at the controls — handling taxi, takeoff, cruise and landing. The answer, simply put, is a spectrum: from a reduced crew that shares the work, to a remote pilot on the ground, to no human in the loop at all. Research and flight data from Merlin, Reliable Robotics and DARPA show all three already working today.
How does Merlin Labs' technology work?
Merlin Labs builds the Merlin Pilot, an aircraft-agnostic autonomy system. The key is that the same core software adapts to different airframes rather than being rebuilt for each. According to the company, the Merlin Pilot has logged more than 100 hours of autonomous flight and is now being matured for the C-130J under a U.S. Special Operations Command programme. Data from those test campaigns feeds the certification evidence regulators demand.
Why is autonomous flight different from autopilot?
Autopilot holds a heading and altitude that a human set; it cannot decide. According to engineers across these programmes, autonomous flight reverses the burden: the system plans the route, watches for traffic, and makes the calls a pilot would, while the human supervises or steps back entirely. The evidence shows the real difference is judgement, not just stick-and-rudder control.
Who is building autonomous flight besides Merlin Labs?
The answer is a crowded field. Reliable Robotics is certifying an uncrewed Cessna Caravan; Joby Aviation absorbed Xwing's autonomy team; Skyryse is simplifying flight controls; and Shield AI flies combat autonomy with Hivemind. Research and defence contracts show the money is flowing to whoever can prove safety to a regulator first. Even seasoned human pilots are watching this race closely.
What are the real risks of pilotless planes?
Analysis of the field reveals three durable risks. The first is certification: regulators must trust a system they cannot fully explain. The second is security, because a remotely supervised plane is a cyber target. The third is accountability — when software flies the plane, the data shows responsibility can blur unless someone owns the decision by design. In other words, the engineering is ahead of the governance, and that gap is where the danger lives.
