xAI rolled out Automations in Grok on July 16, 2026, a feature that lets users describe a job once and have the chatbot carry it out on a schedule or the moment an email arrives. The company said the tool is live on grok.com and inside the Grok app for iOS and Android. It marks a clear step from chatbots that wait for a prompt toward software that acts without being asked each time.
The move puts xAI alongside a widening field of agent-building labs. Google, Microsoft, and OpenAI have each shipped or previewed systems that can take a goal and work toward it across many steps. xAI's version is narrower and consumer-facing: it is built to run small, repeatable jobs — a morning news scan, a rent reminder, a triage of the inbox — rather than open-ended research.
"Describe the job once and Grok runs it on its own," the company wrote in its July 16 announcement on x.ai. Users set instructions in plain language, attach files for context, add connectors and skills, and pick a cadence. From that point every run is a fresh request: same instructions, current data.
How the feature works
Automations offer two ways to fire. A schedule runs once, daily, on weekdays, weekly, monthly, or yearly, at a time the user chooses in their own timezone. xAI gave the example of a morning brief at 8:00 before the workday starts, or a bill reminder on the first of the month.
The second trigger watches email. When an incoming message matches filters the user sets — by sender, recipient, or subject — the automation runs with that email as its context, and Grok responds to the actual message. Instructions can also point at connected tools directly; typing "@" mentions a connector, and Grok uses it on every run.
Each run opens a real conversation. Grok does the work and saves the result to a run history the user can open to read the full thread or pick up where the assistant left off. The user chooses how the automation reports back: email, in-app notification, both, or neither. A "Run now" button lets someone test an automation right after building it.
xAI said scheduled automations are available to everyone, while email triggers are included with SuperGrok, the paid tier. Users can also spin up an automation straight from a chat by asking Grok to set one up, and the Automations page ships with suggested templates. Any automation can be paused, resumed, edited, or deleted at any time.
A model built for the work
The launch landed the same day xAI introduced Grok 4.5, which the company called its smartest model for coding, agentic tasks, and knowledge work. The pairing is not accidental. Scheduled agents need a model that can follow instructions reliably across many independent runs, pull current information, and write a useful summary without a human watching.
Grok 4.5 follows a busy stretch for the lab. In the weeks before, xAI opened Grok Build, a terminal-native coding agent, to open source; added 21 new flagship voices; and shipped a voice agent builder. The company has also pushed Grok onto third-party platforms including Amazon Bedrock, Databricks, and Microsoft's Office add-ins. Automations is the consumer-facing layer that ties those pieces together: a place where a non-developer can tell the system to do something repeatedly.
The bigger shift to software that runs itself
Agentic AI has moved from demo to product across the industry in 2026. The appeal is plain. Most knowledge work is repetitive: checking feeds, summarizing threads, flagging conflicts, drafting replies. A model that can be pointed at those chores and left to run changes how people think about the tool. It is no longer a search box; it is a junior colleague who starts before you wake up. The AI category on our site has tracked this creep from chat to action all year.
That promise carries real limits. An automation that reads email and acts on it needs broad access to a person's inbox, and a wrong summary can send someone chasing a problem that does not exist. xAI's design keeps each run as a separate conversation and lets users inspect the thread, which helps, but the underlying risk of autonomous actions on personal data remains. Regulators in the European Union are preparing full enforcement of the AI Act in August 2026, with fines reaching 35 million euros or 7 percent of global revenue for the worst violations.
Cost is another constraint. Running a model on a schedule means paying for compute every time the job fires, even when nothing changed. xAI has not published pricing for the scheduled runs beyond the SuperGrok gating of email triggers, and heavy users will watch their usage climb. For now the free tier covers schedules, which lowers the barrier to trying it.
Where it fits among the AI labs
xAI is late to some races and early to others. OpenAI shipped GPT-5.6 earlier this month and has pushed its own agent and operator features; Anthropic's Claude handles long-running tasks and is preparing for a public offering; Google's Gemini line added computer use and live translation. What xAI brings is a clean, scheduled-job interface inside a mainstream consumer app, rather than a developer framework.
The data centers behind these models are straining under the load. Training and serving frontier models now consumes enough electricity that utilities and governments track AI compute as a grid issue. Our coverage of cloud and edge computing has tracked metro data centers pulling inference work closer to users as demand surges. That build-out is the silent prerequisite for features like Automations: without spare capacity, a chatbot that runs on its own a thousand times a day is just a bill.
What users can try today
The fastest path is grok.com/automations. xAI suggests starting from a template or from chat: tell Grok to check the news every morning and flag anything about a topic, and it sets the job up. Email-triggered automations require SuperGrok. The company said every automation reports back the way the user picks, and each run is saved so it can be audited later.
For a lab that spent 2025 racing on model size, the July 16 releases show a turn toward usefulness. A chatbot that waits is a toy for many people; one that does the small stuff on a timer is closer to a tool they will keep open. Whether users trust it with their inbox is the real test, and that answer will take more than a launch day.