The Verge's Kiwibit bird feeder test: the AI worked, the pigeon didn't care

A feeder that names its visitors

The Verge has published a hands-on account, dated October 10, of the Kiwibit Bird Feeder 2 Pro, a smart feeder shown at $269 in the piece (from $350) whose headline feature is a 4K HDR camera with built-in AI bird identification sold as lifetime-included. Writer Thomas Ricker mounted one in his garden, and within a few weeks small birds, great tits and Eurasian blue tits among them, arrived in large numbers to feed on the husked sunflower seeds it dispensed.

The AI layer did its advertised job at first. Phone alerts named each visitor, with a characteristic hedge: the notifications read "Looks like Great Tit!" and "Looks like Eurasian Blue Tit!" The companion app let Ricker read up on migration habits, nesting behavior and diet for each species, and the recorded footage was sharp enough that he began looking for distinguishing markings in the hope of naming individual birds.

The pigeon that ended the experiment

The turning point was an alert for a common wood pigeon, a large gray bird heavy enough to obscure the camera lens entirely. Ricker triggered the feeder's Nuisance Animal Alarm, an audible siren designed for squirrels and similar intruders. According to The Verge, the pigeon tensed, looked around, and went back to eating. It fed for 15 minutes straight until Ricker walked outside and shooed it away himself.

The pigeon, or pigeons, since he could not tell one from another, kept returning for sessions of 30 minutes or more, and mice moved in underneath to feast on the spilled seed. Ricker eventually conceded defeat, took the feeder down and stored it in a closet.

What the AI could and couldn't do

Notably, The Verge does not report the AI misidentifying anything; the labels appear to have been accurate. The failure the piece documents is subtler and arguably more instructive for anyone weighing AI-powered gadgets. Every premium feature, the identification, the species profiles, the high-resolution video, served the enjoyable part of the hobby, while the single feature aimed at the product's realistic failure mode, the deterrent alarm, proved useless against a determined bird. The lifetime AI, as the product listing promises it, contributed nothing to the problem that actually ended the review.

Ricker is explicit that he does not blame Kiwibit: he found the feeder genuinely pleasant to use right up until the pests took over, and the attraction of pigeons and mice is not the manufacturer's doing. His practical advice for anyone tempted by a smart feeder is to start with something simple, inexpensive and free of AI, because a used feeder cannot be returned simply on the grounds that it draws unwanted animals.

Why it matters

The piece works as a compact case study in how AI ships in consumer hardware today. The identification feature was genuinely used and, on this evidence, broadly accurate, but it arrives wrapped in hedged language, it adds real cost to the device, and it does nothing about the mundane physical problems that decide whether a product stays mounted or ends up in a closet. Buyers evaluating AI-included gadgets would do well to separate the feature list from the question of which failure modes the AI actually addresses. Here, the answer was none of the ones that mattered, and the technology that worked flawlessly was ultimately beside the point.