How OpenAI turned its coding tools into dots, an always-on personal assistant

OpenAI just dropped another shiny toy on the AI playground – dots, an always‑on personal assistant that lives in the ChatGPT app for Pro‑tier users. As a founder who’s been wiring up production servers for a decade, I see this as another classic case of a hyperscaler slapping a premium price tag on a feature that could be cobbled together with open‑source tools, then wrapping it in a “must‑have” narrative. The hype is real, the engineering is solid, but the business risk for independent hosting providers and SaaS founders is anything but.

What dots actually does – and why it matters

Dots is positioned as a proactive AI that stays awake in the background, monitoring your calendar, email, and other tools to surface conflicts and suggestions without you having to ask. In the TechCrunch interview, OpenAI’s enterprise lead Alexander Embiricos describes it as “an entity and it’s a single conversation, not many different conversations,” aiming for an intuitive experience. The assistant lives inside the ChatGPT desktop and mobile apps and is only available to subscribers on the Pro tier or higher, with plans starting at $100 a month.

The core promise is that dots will “think ahead” – flagging calendar clashes, upcoming deadlines, and even analyzing outages for a drone delivery startup, according to Embiricos. In practice, the early version only warned users minutes before an event, as Embiricos admits: “I only have three minutes!” Still, the concept of a persistent agent that can act without prompting is a clear signal that the AI market is moving from reactive chat to autonomous assistance.

Pricing and the hyperscaler trap

At $100 a month for a single user, dots sits squarely in the premium tier of AI services. For an enterprise, that cost can balloon quickly, especially when you factor in the token usage that a constantly listening agent generates. OpenAI’s own admission that “more token usage” is a byproduct of the always‑on model underlines the hidden cost. Independent hosting providers know that token‑driven pricing can erode margins faster than any hardware expense.

Contrast this with the earlier Codex app, also from OpenAI, which required more manual prompting but didn’t carry the same subscription premium. The trade‑off is clear: pay for convenience or build a leaner, more controllable workflow yourself. For founders, the lesson is to question whether the marginal productivity boost justifies the recurring spend, especially when you can self‑host a comparable LLM stack on cheaper infrastructure.

Data privacy – the elephant in the room

Dots needs deep access to personal data to be useful – calendars, emails, cloud drives – and OpenAI offers the same privacy controls as ChatGPT: you can turn off model training and limit resource access. However, the controls are not granular, and the app refuses read‑only cloud drive access, forcing broader permissions. Embiricos acknowledges the need for “super, super, super granular controls,” but the product still leans on a model where data is harvested to improve the service.

This mirrors the criticism aimed at Meta’s Muse, which builds detailed profiles of your friends and family to power advertising. OpenAI’s data collection, while framed as “proactive research,” still feeds back into the broader model training pipeline unless you explicitly opt out. For businesses handling sensitive data, that creates a compliance headache – you’re essentially handing a third‑party a key to your internal knowledge base.

Product‑market fit vs. real‑world reliability

In testing, the assistant could spot calendar conflicts but did so “not necessarily in the most timely fashion.” Embiricos cites a three‑minute warning as an early example, acknowledging that timing will improve. The technology is promising, but the current reliability gap means you can’t fully trust it for mission‑critical alerts. In production environments, a missed or late notification can translate to lost revenue or SLA breaches.

OpenAI’s own anecdote about a drone delivery startup’s outage illustrates the potential upside – an AI that detects and analyses incidents faster than a human. Yet, that scenario is still anecdotal. For founders, the risk is that you’re banking on a feature that’s still in its infancy, while your business depends on consistent performance. The prudent move is to treat dots as a supplemental tool, not a core operational component.

The hidden infrastructure cost of always‑on agents

Running a persistent AI that constantly polls calendars, emails, and cloud storage is a heavy lift on the backend. OpenAI admits that “more token usage” is inevitable, which translates into higher compute and bandwidth consumption. For hyperscalers, that cost is absorbed across massive economies of scale, but for smaller providers, each additional token can mean a noticeable uptick in server load and electricity bills.

Independent hosting firms that offer AI‑enhanced services need to factor this into capacity planning. If you’re already juggling multi‑tenant workloads, adding a fleet of always‑on agents could push you past the sweet spot of utilization, forcing you to over‑provision or risk latency spikes. The bottom line: the convenience of an always‑on assistant comes with a hidden infrastructure tax that can erode profit margins if you’re not careful.

Strategic takeaways for founders and hosting providers

First, scrutinize the pricing model. A $100 monthly fee per seat may look palatable for a small team, but scale it up and the expense becomes a line item that can outpace the value delivered. Second, demand granular privacy controls. If an AI service can’t let you limit access to read‑only for cloud drives, you’re handing over more data than necessary. Third, treat the assistant as a productivity enhancer, not a mission‑critical system. Its early‑stage timing issues mean you should keep a human in the loop for high‑stakes decisions.

Finally, consider building your own thin wrapper around an open‑source LLM. You can host it on your existing infrastructure, apply strict access policies, and only pay for the compute you actually need. This approach sidesteps the hyperscaler premium and gives you full control over data residency – a crucial factor for compliance‑heavy industries.

Actionable steps for the independent AI operator

1. **Audit your current AI spend** – Identify any recurring subscription fees for AI assistants and calculate the per‑user cost versus the productivity gain you observe. If the ROI is marginal, start looking at alternatives.

2. **Implement granular permission layers** – Use OS‑level sandboxing to restrict AI agents to read‑only access where possible. If the vendor’s SDK forces broader permissions, flag it as a deal‑breaker.

3. **Pilot a hybrid workflow** – Combine a lightweight open‑source model for routine tasks (e.g., calendar parsing) with a human review step for critical alerts. This reduces token usage while preserving reliability.

4. **Monitor token consumption** – Set up alerts on token usage spikes when using always‑on agents. Unexpected surges can indicate misbehaving processes or excessive background polling.

5. **Plan for scalability** – If you decide to adopt an always‑on assistant at scale, allocate dedicated compute resources and test under load. Remember that each additional user multiplies the background workload.

By taking these steps, you can reap the convenience of AI assistance without surrendering control or profitability to a hyperscaler’s subscription model.

In the end, OpenAI’s dots is a slick demonstration of where the AI market is heading – from on‑demand chat to autonomous, always‑on helpers. The technology is impressive, but the business implications are messy. As a founder who’s been in the trenches of production infrastructure, my advice is simple: enjoy the novelty, but keep your eyes on the cost, the data privacy trade‑offs, and the reliability gap. If you can navigate those challenges, you’ll turn a hype‑driven product into a genuine competitive advantage.

— Allan Ali, Founder

This article was produced with AI-assisted research and editorial support. Reporting is based on the source material cited below. Sources: TechCrunch; techcrunch.com; Global1.News (27 October 2026).

By Allan Ali, Global1.News