Opinion
AI is ready to act. Companies are not.
As intelligent systems move from pilots into vehicles, factories, and fleets, the bottleneck is no longer the model. It is the operating environment that lets the model earn its authority to act.
AI can already write, plan, recommend, and increasingly act. For a large organization, the hard question is no longer what the technology can do. It is what the organization is prepared to let it do. That question stays easy while AI lives in digital workflows. It gets much harder the moment intelligence enters a vehicle, a factory, a fleet, or a grid. There a wrong output is not a bad answer. It is an operating event.
For years I have worked between the startups building this technology and the large organizations that run real-world systems. The demo impresses. The technical team is strong. Then deployment slows, because the layer between the technology and the operation was never built. The startup built one side. The corporate built the other. The operating layer between them was nobody's job.
That missing layer is what I call the habitat. It is the environment around an intelligent system that lets it act, earn authority, and improve. A habitat does four things. It sets the boundaries of the job the system is allowed to do. It keeps an owned record of what the system did. It grants authority in stages rather than all at once. And it ties each action to a consequence the organization can see and answer for.
A brain is not an organism. A model can perceive, predict, and recommend. Acting in the world takes a body around it: workflows, safety rules, data feeds, oversight, and accountability. In a digital pilot, the missing body can stay hidden. In the physical world it shows up fast, because every action meets friction from regulation, safety, operations, liability, and human trust.
So the real work is answering a set of hard questions before a system is allowed to act. Which edge cases were tested? What counts as enough evidence? Who sets the minimum safety threshold? What does the system do when it is unsure? Every industry that puts intelligence into physical systems runs into them.
A system that predicts a failure or recommends a fleet action cannot jump from pilot to full autonomy in one step. It earns its way there. First it advises, with a person deciding. Then it works under observation. Then it leaves a record that can be reviewed. Then it improves as people override it and it learns why. Only after enough evidence accumulates should the organization let it act with more independence. Authority is granted against evidence, not against a demo.
This is why the next phase of adoption will reward companies that build memory around deployment. Every action, override, and outcome is evidence. Kept well, it teaches the next deployment and shortens the next rollout. It shows the organization where its systems can be trusted and where they still need limits. Most companies throw this evidence away. The ones that keep it deploy faster and with more confidence each time.
Startups should read the same shift in reverse. In physical industries, the strongest opening is often the missing layer inside a system that already exists. Large organizations know their biggest problems, and some are too broad for an early company to own. The better opening is the part they do not yet know they are missing: validation, monitoring, safety evidence, data translation, or the workflow logic that turns a capability into a repeatable operating process.
This is where Israel is well placed. It builds across the whole system, not only the model at the center. It makes the software layers that surround physical systems: intelligence, safety, data, cyber, and validation. It also builds parts of the organism itself: the sensing that lets a machine take in the world, and the smart, compact energy that powers it. That is where deep technology, software talent, and operational instinct meet the needs of global industry. The habitat is a build problem, and building around hard physical systems is a local strength.
The first decade of modern AI rewarded access to powerful models. The next decade rewards the harder thing: the habitat where a model can act, safely and repeatedly. The intelligence is arriving. The habitat it needs is the work still ahead of us.
Tal Cohen is Founder of Drive TLV and Drive Europe and General Partner at NextGear Ventures.
