GPT-4o with scheduled tasks (jawbone) is available in beta

Original Article ↗ Hacker News Discussion ↗

OpenAI has quietly rolled out a beta version of GPT‑4o with “scheduled tasks,” letting users set recurring or one-off automations that trigger ChatGPT to act at specified times and optionally send notifications. Early users report unclear UX, missing or confusing documentation, timezone and reliability bugs, and behavior that feels closer to a basic reminder/todo system than a polished agent platform. The feature is nonetheless seen by some as an early step toward persistent AI agents and workflow automation, while others question its practicality, execution quality, and OpenAI’s broader product strategy.

Feature description & docs

  • New “GPT-4o with scheduled tasks (jawbone)” model adds automation that runs prompts on a schedule.
  • Tasks have three core parameters: title, prompt, and schedule, with schedules defined using iCalendar VEVENT syntax (e.g., RRULE, optional DTSTART).
  • Official help and changelog pages exist, but many find them sparse, confusing, or incomplete; some see “Invalid DateTime” in the changelog.

User experience & discoverability

  • Several users report the UI gives almost no hints about how to use scheduled tasks.
  • People discover functionality only after clicking avatar → “Tasks” or experimenting with phrases like “remind me…”.
  • Mobile and desktop apps behave inconsistently; some capabilities (e.g., editing tasks, Canvas) appear only on web.

Functionality & behavior

  • Tasks can send notifications via email and/or push (user-configurable).
  • They can run web searches as part of the scheduled job, though enabling “Search” during task creation can confuse the model.
  • Some see this as “scheduled tasks for the AI” (periodic research, summaries) rather than just user reminders.

Reliability & technical issues

  • Reports of incorrect time zones (defaulting to UTC), time deltas added to requested times, missed executions, and absent push notifications.
  • Some users receive “task couldn’t be completed” emails or generic errors.
  • Rollout appears flaky; the model sometimes doesn’t show up or errors in the UI (e.g., source map errors).

Use cases & value debate

  • Supporters see this as an early step toward agents, workflow automation, and “AI cron” for non-developers.
  • Critics dismiss it as a glorified reminder/todo app, especially given existing tools with better calendars and sync.
  • Some argue free or local models and simple cron + API are sufficient; others value having this inside ChatGPT they already pay for.

Comparisons & strategic context

  • Comparisons made to Siri, Google’s ecosystem (Gmail/Calendar/Maps), and other LLM providers; mixed confidence in each company’s execution.
  • Some view this as a tech demo to acclimate “normal users” to agents and long-term AI memory.
  • Skeptics question OpenAI’s product quality, QA, and long-term moat; others see steady capability improvements and inevitability of agent-based workflows.

Ethical / political concerns

  • At least one commenter refuses to use OpenAI for political/leadership reasons, noting similar issues likely exist across major tech platforms.