OpenAI’s Codex Cloud, Voice Controls, and Dots Unveiled at AI Engineer Code Summit

OpenAI’s “New Ways of Working with Codex” workshop at the AI Engineer Code Summit highlighted three practical shifts: a revamped cloud execution model, voice-first interaction, and the broader rollout of personal agents called “dots” Latent Space. The presenters, Dominik Kundel and Charlie Guo, framed the changes as responses to persistent friction in earlier releases.

Cloud environments become background executors

Codex launched in May 2025 as a cloud-based agent, but early users reported clunky handoffs between local and remote contexts. Kundel noted that version 2 of Codex Cloud, announced at the September DevDay, now allows a single published environment to spawn multiple isolated tasks. Users select GitHub repositories to grant access, and the system can keep work running after the laptop is closed. Kundel demonstrated moving a local task to the cloud, preserving the Appshot capture, and described the cloud executor as “just a background” that merges with the local instance. The team acknowledges that managing context differences remains a work-in-progress.

Voice as a multidimensional control plane

During a live demo of a fictional record-label site, Kundel used voice dictation to iterate UI changes, emphasizing speed when the end goal is unclear. Guo expanded the concept, treating voice not as a single workflow lane but as a “control plane” for thread management, data retrieval, and spawning new tasks across the codebase.

Dots: safety nets, delegation sinks, and ubiquitous presence

Personal agents—OpenAI’s “dots”—were presented as a nascent paradigm. Guo admitted that power users find them “a little bit awkward,” yet identified three core uses: flagging missed issues (safety net), handling delegation via Slack (delegation sink), and providing an “everywhere” presence across dev boxes, laptops, and cloud environments. Kundel confirmed he often initiates tasks from his dot because it already holds relevant context and can push PRs forward.

Our read: OpenAI is stitching together cloud, voice, and personal agents into a single, fluid developer experience, but the transition points—especially context handoff and dot ergonomics—still demand refinement.