Agentic AI wins on tasks, not promises: the Snapdragon Summit lesson for photographers

By Zendoric · October 1, 2026.

Last week, at Qualcomm's Snapdragon Summit in Hawaii, the word heard most was not camera, chip or battery: it was "agentic". Jeremy Gray, senior technology editor at PetaPixel, recounted that during the first keynote he heard the same slogans repeated ad nauseam: a "personal" AI that "knows everything about you" but is "private and secure", and that will mean he "never wastes time again". His initial reaction, conveyed in a message to his editor-in-chief, was flat rejection: "Why would I want that?".

The relevant point is not the keynote itself, but Qualcomm's own acknowledgement, as Gray reports it: the biggest obstacle to agentic AI is not technical, it is public perception. A large share of users —Gray includes himself— distrust an assistant with simultaneous access to email, calendar, photos, messages and even a credit card. It is a reasoned distrust, not technophobia: the broader and more opaque the mandate given to an agent, the harder it becomes to verify what it does with those permissions.

And yet the article itself documents the turn. When Gray thinks about specific, narrowly defined tasks, the resistance drops. He gives a very personal example: to cover the Natural Landscape Photography Awards he had to download, resize and organize almost 100 photographs, and write the captions from data already available in another document. No step in that process requires creative judgement; it is digital plumbing that steals time from what does add value, which is looking at the photos and writing about them. It is not an isolated case: PetaPixel has already covered concrete moves in that direction, such as an Anthropic agent capable of executing resizing commands directly in Photoshop, or the AI assistant Adobe is building into its own editor to free up creative time.

The second area the article explores is more speculative: smart glasses. Gray imagines —and it is worth stressing that this is a hypothetical exercise by the journalist himself, not an announced product— an agent that, knowing you have your camera with you, suggests a nearby shooting spot in real time based on the light and the weather, shows you a preview with the framing of a 50mm lens, or acts as a tutor in composition and exposure for beginners. There are already real precedents for AI as photographic teaching support, such as Google's Camera Coach, so the extrapolation is not far-fetched, even if today it is the imagination of a sceptical user rather than a confirmed product roadmap.

What is really interesting here, beyond the anecdote of a journalist in Hawaii, is the pattern it illustrates about how trust in agentic AI is being won (or lost). It is not won with slogans about "I know everything about you": that is precisely what sets off the alarm over privacy and concentration of power that Gray voices bluntly. It is won by confining the agent to a verifiable task, with a result the user can check in seconds. It is the same principle we have been observing sector by sector in the labour market: the administrative and repetitive —organizing files, resizing images, generating metadata— is the first thing to fall to automation, while judgement, curation and the human relationship with the craft hold up and in fact gain relative weight. The photographer does not disappear; what disappears is the time spent on tasks that never required being a photographer in the first place.

It is worth keeping in mind, though, that the promise of "private and secure" remains just that, a Qualcomm marketing promise with no independent verification in the article itself. The underlying problem —who controls the agent, what data it touches, what happens if it gets it wrong or is manipulated— is not solved by promising it from a stage, but with verifiable architectures: processing on the device itself instead of in the cloud, granular and auditable permissions, and traceability of what the agent did and why. That is, in fact, the same debate running through agentic AI in the enterprise today, where containing the agent is increasingly being resolved through infrastructure rather than through the model's promise.

The underlying read, which connects with Zendoric's long-term thesis, is that the value of well-bounded agentic AI is not a technological luxury: it is human time returned to what really matters, whether that is photographing, writing or creating. That is exactly the kind of incremental abundance —minutes and hours freed from administrative friction— that, added up across the economy, points towards the horizon we defend: a society where tedious work is absorbed by the machine and people spend their time on what they are passionate about. But that horizon is only earned by first solving, with facts and not slogans, the trust problem that a sceptical journalist in Hawaii has just set down in writing with complete clarity.