OpenAI Launches Dots: Always-On Agents That Act, Not Chat

OpenAI shipped Dots today: always-on autonomous agents that pursue user goals across apps - managing projects, updating sales docs, building demos, reachable via Slack and Teams, powered by GPT-6 Astra. The pitch is head-to-head against Meta in the agent race, and it marks the enterprise shift from chatbots to software that acts for you. Reuters via The Manila Times

The launch had friction. Live demos failed on stage - voice updates kept breaking - and OpenAI pulled a more powerful Astra variant over safety concerns. That puts it in the middle of a tense week: the day before, Trump and six tech CEOs signed a voluntary frontier-safety accord with no consequences for breach. The 1 News

The agent economy owned Product Hunt

Today's Product Hunt feed reads like a press release for the same thesis. Dots by OpenAI is one of the top launches - always-on agents "built to handle everything," available to anyone with a GPT subscription. Product Hunt Alongside it: DSH Desktop, DeepSeek's official app for its open-source agent harness, legitimizing the local/open-source agent stack for self-hosters. Product Hunt And Polylane, an agent that monitors your code so you're not the one on call - a genuinely novel niche for teams burned out by 3am pages. Product Hunt

Frontier labs are shipping official agent surfaces, and startups are picking the niches in between.

The infrastructure race is where the real build is

GitHub's trending page tells the same story: agent context and memory infrastructure is the battleground. Three of the seven AI repos trending today are about how agents hold state - PageIndex, a vectorless, reasoning-based RAG index where the LLM reasons through document structure instead of embedding search VectifyAI/PageIndex; OpenViking from Volcengine, a self-evolving context database unifying memory, RAG, and skills volcengine/OpenViking; and needle, a 14MB foundation model that runs natively on phones, wearables, and robots - the edge end of the same trend cactus-compute/needle.

HN's counter-question: what value do we get from AI?

Not everyone is buying it. The freshest AI thread on Hacker News, posted two hours ago, asks bluntly "what value do we get from AI?" - and the top take is that LLMs are worth a lot as an "advanced search engine + autocomplete," less as magic. Hacker News The practitioner mood is skeptical and value-hunting: less "what can this do," more "prove the ROI." A useful frame for teams: position AI as a tool, not an oracle.

Signal to watch: The voluntary-accord era is meeting its first real test - agent safety incidents are already piling up, and Dots' failed on-stage demos may be the first of a long list.