OpenAI Launches Dots: Autonomous AI Agents for 24/7 Productivity

OpenAI's new dots are autonomous AI agents powered by GPT-6 Astra that can manage complex tasks 24/7 using their own cloud computers. These agents integrate into existing workspaces like Slack and Teams, learning user preferences to proactively handle everything from coding to sales proposals. While highly capable, dots include robust safety controls and sandboxing to ensure users maintain final authority over sensitive decisions.

Key Points

• Dots are persistent AI agents powered by GPT-6 Astra that operate 24/7 using their own cloud-based computing environments and browsers.
• They integrate with over 4,000 applications and platforms like Slack and Teams to proactively manage tasks and learn user-specific standards.
• Safety is managed through sandboxing, read-only background research, and customizable permission rules that require human approval for consequential actions.
• Specialist dots are being piloted for enterprise-level roles, offering dedicated identities and integration with Microsoft's governance tools.
• The rollout begins with Pro and Business users, offering one dot at no extra cost with options to scale speed and task volume in the future.

Sentiment

The overall sentiment is skeptical but not dismissive. HN does not broadly accept the article's optimistic framing at face value, but the strongest comments are not simple rejection; they describe real use cases and operating practices while insisting that autonomy must be bounded, auditable, and permissioned. The community is especially wary of branding, safety claims, and unsupervised access, yet many accept that persistent agents can be useful when treated as constrained assistants rather than fully trusted workers.

In Agreement

• Persistent, domain-specific agents can handle tedious operational work such as support triage, bug investigation, tracker cleanup, infrastructure checks, release assistance, and draft customer responses.
• Separate agents with their own workspaces, identities, and scopes can create clearer trust boundaries than one universal assistant with a merged memory across personal and business contexts.
• Asynchronous cloud agents are valuable because humans do not need to babysit every step; even if they are slower end to end, they can produce plans, diffs, or drafts for later review.
• Agents should be treated like fallible junior employees rather than deterministic software, with permissions earned gradually through observed reliability.
• Managed always-on agents could make a pattern that technical users are already building with shells, email, containers, issue trackers, and cron accessible to less technical users.

Opposed

• Autonomous agents are not trustworthy enough to act without close human oversight, especially when they can use accounts, browsers, files, customer channels, or infrastructure.
• Jailbreaks, hallucinated actions, destructive commands, and weak safeguards make persistent agents risky even when vendors advertise sandboxing and approvals.
• AI output can quietly direct human behavior because reviewers often accept most generated work after only shallow inspection.
• The cute branding and character-like framing are seen by many as manipulative, emotionally sticky, and designed to soften concern about surveillance, dependency, and corporate incentives.
• Some commenters view the launch as predictable hype, an attempt to increase token consumption, or a repackaging of tools that already exist in developer-oriented agent systems.