Claude spotlights an overnight camera experiment built with Claude Code
London AI builder Paolo Rosson says he left a camera pointed at Old Street and asked the coding agent to take pictures through the night.
By Ryan Merket · Published
Primary source: Claude
Why it matters
The repost presents Claude Code in a camera-based creative workflow, but the clip documents a user-built experiment, not a supported product capability.
Anthropic's official Claude account spotlighted a camera experiment in which London AI builder Paolo Rosson (@redp314) says he left Claude Code running overnight with a view of Old Street. The account's October 2nd post describes the setup in a sentence; Rosson's original post and video give the premise: he gave the coding agent access to his camera, asked it to take pictures all night and be creative, then went to bed.
Anthropic introduced Claude Code as a command-line agent for software work: it can inspect and edit code, run tests and use command-line tools. Rosson used it for an open-ended task outside that software-work remit, connecting it to a camera and asking it to decide what to make. The clip demonstrates a workflow Rosson assembled around the tool, not a native camera feature or a formally supported workflow from Anthropic.
The builder behind the experiment
Rosson's personal site lists a PhD in atomic and laser physics from Oxford, applied AI work at Dext and earlier projects including MenuPhotoAI, an AI service for restaurant photography. His GitHub profile describes him as a MenuPhotoAI co-founder and says he builds production apps with AI as his engineering team.
According to his profiles, Rosson has worked on products and production AI systems. This is a small, personal test of how far a coding agent can be pushed beyond editing software. He gave the agent a device, a location and room to decide what to make, rather than specifying a feature or finished application.
The account's post does not establish how Rosson connected the camera, what software mediated access, or what the agent produced from the captured images. The video alone cannot show how much was autonomous, how much came from custom code, or whether the same process would work reliably for another user.
A product showcase, not a product launch
Anthropic's official account shared the experiment as a Claude Code example, presenting an agent usually associated with terminal-based software work in a creative setting. The post does not announce a camera integration, a new model capability or an Anthropic-run test.
Claude Code can be part of a user-built workflow that connects a model to other tools. Anthropic describes the agent as able to work through command-line tasks; Rosson's experiment supplied the camera interface. The workflow's capabilities depend on the setup, not just the model.
Rosson says he gave the system an instruction and stepped away. The post does not establish how much supervision the process required before he went to bed, how it handled errors, or whether the camera ran continuously. It does not provide a reproducible technical account.
Anthropic's repost endorses the example as worth sharing, but does not show that the company built or verified the workflow. Rosson's post leaves the setup and its reliability largely undocumented.