Forrester's Farnborough Recap: Autonomy Is Still Mostly a Person With a Joystick

Forrester VP and Principal Analyst Paul Miller went to the Farnborough International Airshow to check autonomy claims against the hardware, and found a gap: “there’s a person on the ground, usually making almost every bank and turn with a joystick.” Despite marketing built around autonomy, Miller writes that most exhibited aircraft are still remotely piloted, calling current systems “more aspiration than reality.” The show wasn’t short on real progress — BAE Systems’ Brontanax collaborative combat aircraft is designed to fly alongside crewed F-35s, Vertical Aerospace flew its first public vertical-to-horizontal transition, Anduril showed its autonomous strike drone “Thunder” alongside Archer’s commercial variant “Halo,” and Loganair committed to five electric aircraft from Beta Technologies for Scottish island routes. But Miller’s frame for all of it is “physical AI” — AI that perceives, reasons, and acts in the real world — as a category distinct from consumer chatbots or enterprise software, and one where deployed capability is still catching up to the pitch.

That’s the same warning Forrester’s Craig Le Clair issued a week earlier about enterprise AI agents, just in a different domain. Le Clair’s tortoise-and-hare framing splits the agent market into fast-deploying horizontal “Hares” (Microsoft, Salesforce, ServiceNow-style SaaS agents good at email summarization and content drafting) and slower, custom-built “Tortoises” that solve substantive, process-specific problems behind the firewall. His argument — “the fastest runner doesn’t always win” — is Miller’s joystick observation restated for software: flashy, fast-to-demo capability and genuinely deployed, load-bearing autonomy are not the same thing, and mistaking one for the other is the recurring error buyers make in both aerospace and enterprise software.

For sales and GTM leaders evaluating any vendor’s autonomy claims — physical or digital — both Forrester analysts are pointing at the same question: is a human still making the decision that matters, dressed up as an agent, or has the system actually earned the handoff?