A robot that folds a towel, a screen that rolls up, an assistant that promises to replace your phone: at CES in Las Vegas, the future often fits into a three-minute demonstration. At home, it has to withstand an unreliable connection, a cluttered room and several years of use. In September 2026, then, the right question is not “what impressed us?” but “what would actually be worth buying, and then keeping?”
To answer that question, it is better to start with documented track records than to assume the latest announcements will succeed when that cannot yet be verified. The examples below are based on facts established before 2026; the outlook is presented as a set of hypotheses, not as a confirmed assessment of the products unveiled in Las Vegas this year.
The show demonstrates possibility, not reliability
A prototype is not a deception. It serves to explore a technology, persuade partners or gauge public interest. The misunderstanding begins when its presentation makes it look like a device that is almost ready. Between a single unit supervised by its designers and thousands of units shipped, there is still the work of scaling up production, certification, quality control and after-sales support.
At a booth, the lighting is controlled, the network is prepared and the scenario is rehearsed. A domestic robot may complete a task using a specific object, placed in the right spot, without being able to handle ordinary variations of that same task. What matters is not just what the machine can do, but the conditions under which it fails. A convincing demonstration should reveal its limitations, its setup time and the human interventions required.
First filter: the price of the full promise
The advertised price is only the beginning of the calculation. Essential accessories, consumables, any subscription and compatible equipment must also be included. A smart lock may require a hub; a camera may need a subscription to store videos. Over several years, an affordable device can therefore cost more than a conventional solution.
AI adds an element of uncertainty: some features rely on servers that cost the manufacturer money every time they are used. A generous launch offer may therefore be revised. Before buying, ask what works without a subscription, without the internet and without an account. If the answer is “almost nothing,” you are buying access to a service as much as a physical device.
Apple’s Vision Pro, which went on sale in the United States in 2024 starting at $3,499, illustrates another distinction: a technical achievement can become a real product without becoming an obvious financial choice. For certain professional uses, a high price may be justified. For watching an occasional film, the comparison with equipment already owned becomes less favorable.
Second filter: available, but for whom?
“Available this year” does not necessarily mean available in France, with a French-language interface, an applicable warranty and a repair provider you can reach. A preorder primarily proves that a manufacturer is accepting reservations. A crowdfunding campaign guarantees neither delivery nor the promised performance.
Reliable indicators are more down-to-earth: a specific retail model number, an accessible manual, understandable return policies and independent tests conducted on production units. Sales through an established retailer provide an additional point of contact, without guaranteeing quality on their own. A delayed release does not automatically invalidate an innovation; it is a reason to wait rather than fund uncertainty.
The checklist to keep handy
- Price: what is the total cost, including subscriptions and consumables?
- Availability: can you buy the finished product and return it?
- Repair: who handles repairs, with what parts and at what price?
- Use: what task genuinely becomes easier?
- Continuity: what remains if the online service shuts down?
Third filter: can you keep it for a long time?
Durability often comes down to details that are invisible onstage: a replaceable battery, an accessible charging port, available spare parts. An ultrathin device may be appealing, only to become uneconomical to repair at the first breakdown. Conversely, a less spectacular design may remain useful longer thanks to straightforward repairs.
The European regulatory framework gives buyers more to work with. Since June 2025, smartphones and tablets covered by the new European rules have been required, among other things, to carry labeling with information on battery life and repairability. Ecodesign requirements also govern their durability. This is not a universal “good product” label, but it is a tangible step forward: some promises can now be compared.
Repair is also a software issue. An intact device can lose its essential function when a server is no longer available. The case of the Humane AI Pin, launched in 2024, is instructive: its services were shut down in February 2025, following the announcement of HP’s acquisition of Humane assets. The physical device alone was not enough to preserve its usefulness. Cloud dependence should therefore be scrutinized as a hardware vulnerability would be.
Fourth filter: everyday life, not the ideal scenario
A useful innovation reduces effort, risk or expense without shifting the problem elsewhere. A robot vacuum can save time, but its overall value depends on the obstacles that need removing, the brushes that need cleaning and its ability to free itself when stuck. The right measure is not the number of advertised features: it is the work actually avoided.
To assess a device, imagine an ordinary week. Does it work when several people use it? With background noise, a poor connection or hesitant gestures? How long does setup take? For an aid intended for older people or people with disabilities, these questions determine real accessibility far more than a flawless demonstration does.
The smart home also exposes the limits of industry-wide promises. The Matter standard, launched in 2022, aims to make interoperability between ecosystems easier. The intention is useful, but its logo does not remove the need to check which features are actually supported. Compatibility with a basic command does not guarantee access to every option in every app.
Progress may be less photogenic
A reliable leak detector or a device that consumes less energy attracts fewer cameras than a humanoid robot. Yet their value may be easier to establish. Even so, they must be compared with the existing solution, and any added burdens must be counted. Replacing working equipment with a slightly more energy-efficient model is not automatically an environmental gain: manufacturing the new product counts too.
What next? In the months following September 2026, the decisive test will probably be less about a proliferation of announcements than the emergence of everyday evidence: consistent deliveries, extended testing, spare parts available to order and explicit software commitments. Manufacturers able to document these points may be better placed to turn curiosity into trust. For buyers, waiting for that evidence does not mean rejecting innovation: it means giving a chance to the kind that remains useful when the spotlights go out.
