The top smart glasses on the market right now, the Meta Ray-Ban camera pair and the display version that followed it, land in a strange middle place: good enough to keep wearing, nowhere near good enough to stop wanting more. That gap between "good enough" and "actually good" is the whole story, and it is the reason a real DIY builder scene has grown up around this hardware instead of waiting politely for version four.
My first computer was an Apple II Plus, in middle school, and I went to computer camp the summer everyone else went to tennis camp. I mention it because it is the honest reason I don't take a company's product roadmap as the ceiling of what a device can do. If a manufacturer says the glasses can't do something, the useful question isn't whether to believe them. It's whether that's a hardware limit or a permission they decided not to give you.
What do the best smart glasses actually do right now?
The current generation, led by Meta's Ray-Ban lineup, does three things well: it takes a photo or video from your face without you fumbling for a phone, it puts an AI assistant in your ear that can identify objects and answer questions on the fly, and it translates spoken language close to real time. The camera model shoots at up to 3K, has a workable zoom, and the open-ear audio is genuinely clear enough for calls and podcasts. The AI wake word works. Object identification works. Basic offline translation covers a useful slate of languages even without a phone connection.
The display version is the more interesting animal. It puts an actual in-lens screen in front of one eye, controlled by a wrist-worn neural band that reads muscle signals instead of asking you to tap your temple like a search bar. Reviewers who have used it call it the first display glasses that are genuinely wearable for a full day without ear or nose fatigue, and that is not a small thing. Prior AR glasses were demo hardware. This is closer to a product.
The glasses are not lying to you about what they can do. They are just not telling you the full price of doing it.
Where does the display actually fail you?
Outdoors, in direct sun, is where every current in-lens display struggles. A projected image competing against Florida daylight loses that fight more often than the marketing photos suggest, and you end up tilting your head or waiting for shade to read what's on the screen. This is not a Meta-specific flaw. It's a physics problem shared by every microdisplay pushing light through a waveguide into a lens, and none of the current consumer options have solved it outright. Manufacturers know this, which is part of why the pitch leans so hard on "glanceable" information rather than anything you're expected to read for more than a few seconds.
Battery life is the second wall, and it is a harder one than most buyers expect going in. The camera-first glasses run around four hours of continuous active use, stretched to something like a day and a half with the charging case doing its job in your pocket between uses. The display model is worse on paper and worse in practice: rated for around six hours of mixed use, independent testing has clocked it dropping to 40 percent charge in an hour and a half of real use. That is not a rounding error. That is a device you plan your day around instead of one that disappears into your routine.
Is the battery anxiety real, or am I being dramatic?
It's real, and it's structural, not a bug that gets patched. A display, a camera, a microphone array, and a radio that's constantly talking to your phone are all drawing off a battery small enough to hide inside a temple arm that still has to look like normal eyewear. Every gram added for battery capacity is a gram fighting against the entire premise of the product, which is that it should look and feel like glasses you'd wear anyway. Until there's a real leap in energy density or a fundamentally different display technology that sips instead of gulps power, that tradeoff isn't going away. You will charge these every night. You may recharge them mid-afternoon if you actually use the display feature instead of leaving it dormant.
What about privacy? Are people right to be uneasy?
Yes, and the industry knows it, which is why every serious player has added a visible recording LED and capped continuous recording windows. Those are real mitigations, not cosmetic ones. But an LED only tells the person across the table that the camera might be on; it does nothing about the fact that a face-height camera on a device most people still read as "just glasses" is a different social contract than a phone held up and pointed. We are still working out what the etiquette is supposed to be, and I don't think that gets settled by a firmware update. It gets settled by norms, and norms take longer than product cycles.
The prescription question used to be a real dealbreaker and mostly isn't anymore. LensCrafters-fitted prescription lenses are available for the mainstream camera model, and some manufacturers now advertise a fairly wide diopter range built in. If you need corrective lenses, check the specific model before you buy rather than assuming; not every pair on the market handles high-index or progressive prescriptions, and that gap still trips people up.
So why hack your own instead of just buying one?
Because the commercial products are optimized for a mass-market default, and the interesting stuff happens at the edges the mass market doesn't ask for. Nobody wants a company deciding which data leaves the device, which model runs on it, or which features get gated behind a subscription six months after the hardware was paid for. A closed pair of glasses is a rental with extra steps. An open one is a tool.
This is not a fringe fantasy. There is a real, active builder scene right now, and it's more accessible than people assume. OpenGlass, an open-source project out of the Based Hardware community, turns almost any pair of glasses into an AI-capable device for around twenty to forty dollars in parts: a XIAO ESP32-S3 Sense microcontroller with a built-in camera, a small battery, and a 3D-printed mount that clips onto an existing frame. It records, identifies objects, recognizes people it's met before, and translates text, all running on hardware that costs less than a nice dinner. On the more finished end, Brilliant Labs sells the Frame, an open-source AR pair with an actual micro-OLED display, priced at $349, that ships with Python and Lua SDKs specifically so developers can write their own applications instead of waiting for the manufacturer's roadmap.
What does it actually take to build your own pair?
Less than you'd think, and more than a weekend. Here's the honest path:
- Skill level: comfortable following a wiring diagram, willing to flash firmware from a command line, patient with a hot glue gun or a 3D printer. You do not need to know how to design a circuit board from scratch. You do need to be unbothered by a project that doesn't work on the first try.
- Starting hardware: a XIAO ESP32-S3 Sense board (camera and microphone included), a small LiPo battery, and either a pair of glasses you don't mind modifying or a 3D-printed mount from the OpenGlass repository. Total cost lands around $20 to $40 if you're sourcing parts yourself.
- Software: the OpenGlass firmware is open on GitHub, along with a companion app; you're not writing an AI model from scratch, you're wiring existing capability (transcription, object recognition, translation) to a small always-on device and deciding what happens to the data it captures.
- A step up: if you want an actual display instead of just camera and audio, the Brilliant Labs Frame is the more finished starting point. It costs more, but you skip the soldering and get straight to writing the software that decides what shows up in your field of view.
- The real cost: not money, time. Budget a few evenings for the electronics build if you're starting from the ESP32 route, and expect the software side, deciding what the glasses actually do for you, to be the part that takes longer than the hardware.
What's the one thing worth remembering here?
Smart glasses in 2026 do a genuinely useful job of getting a camera and an AI assistant off your phone and onto your face, and they do it with real limits: sunlight beats the display, the battery beats your patience, and privacy is a social problem no LED fully solves. None of that is a reason to wait for version four. It's a reason to stop treating the roadmap a company hands you as the actual ceiling, and start finding out where the ceiling really is yourself.
Going to computer camp instead of tennis camp was never really about the computer. It was about wanting to know how a thing worked instead of just what it was sold as. Twenty dollars in parts and a free evening is still the fastest way to find out what a piece of hardware will actually let you do, once nobody's product manager is standing between you and the chip.
