OpenAI has started letting ChatGPT's mobile app do more than talk back. Plus and Pro subscribers can now use voice commands inside the app's Work tab to trigger agentic tasks — meaning ChatGPT can carry out multi-step actions like drafting documents, organizing information, or working through a project plan, instead of just responding to a question with text.
What happened
The Work tab, which OpenAI introduced earlier as a business-focused workspace inside ChatGPT, previously required typing out instructions. Now it accepts spoken commands and can act on them autonomously, chaining together steps without the user re-prompting after each one. This is a mobile-first rollout, restricted to paid tiers for now.
This matters because it closes a gap between two features OpenAI had been building separately: voice interaction and agentic task execution. Voice mode has existed since 2023, mostly as a way to have spoken conversations with the model. Agentic capabilities — where ChatGPT completes tasks across multiple steps with minimal supervision — have rolled out more recently through browser and desktop tools. Combining the two on a phone means a user can, in theory, dictate a task while walking between meetings and have it done by the time they sit back down.
The restriction to Plus and Pro subscribers follows a familiar OpenAI pattern: premium features launch to paying tiers first, then either expand downward to free users in stripped-down form, or stay locked behind a paywall as a retention tool. Advanced Voice Mode followed this exact path in 2024, launching to Plus users months before reaching free accounts in a limited version.
Why it matters
Every major AI vendor is racing toward the same destination: an assistant you talk to that then goes and does the work, rather than one that just answers. Google has been testing agentic browsing tools, Microsoft has pushed Copilot agents into Office apps, and Anthropic has experimented with letting Claude operate a computer directly. Voice-driven agentic tasks on mobile is a distinct front in that race — it targets the moments when someone isn't sitting at a laptop, which is most of a typical business owner's day.
What's genuinely new here isn't the voice interface or the agentic execution individually — both have existed in some form for over a year. It's putting them together in a phone app most people already have installed, lowering the friction to actually use agentic features rather than treating them as a novelty.
What this means for small businesses
For a business owner who already pays for ChatGPT Plus, this adds a capability without adding cost — provided the tasks it can handle overlap with real work, like summarizing a batch of emails, drafting a proposal outline, or pulling together notes from a voice memo into a structured document. The practical value depends heavily on how reliable the agentic execution proves to be; agentic AI tools across the industry still make mistakes on multi-step tasks that a human wouldn't catch until after the fact.
The bigger consideration is workflow trust. Handing off a task by voice, on a phone, without reviewing each step, means less oversight than typing instructions and checking output line by line. Businesses handling sensitive client data or financial information should treat this as a productivity tool to supervise, not a set-it-and-forget-it assistant.
What to watch
Watch whether OpenAI extends this to Team and Enterprise tiers, which would signal confidence in reliability for business use, and whether error rates on multi-step tasks get reported publicly or through user complaints. Also worth tracking: whether competitors with existing voice assistants — Google Assistant, Microsoft's ecosystem — respond with their own voice-to-agent features within the next two quarters.
The bottom line
Plus and Pro subscribers now have a new way to trigger multi-step ChatGPT tasks by voice on mobile, but the practical payoff depends on how well the agentic execution holds up under real business workloads — something worth testing on low-stakes tasks before trusting it with anything client-facing.