Artificial Intelligence (AI) no longer just answers questions in a chat window: it can create and send emails, manage messages, document meetings, and help change software. This affects anyone who regularly works with Gmail, Apple Messages, calendars, video calls, or Slack. Several integrations announced in August 2026 show AI assistants moving closer to the work itself and increasingly taking action on a user’s behalf.
From the chat window to the work app
The most notable step comes from Anthropic: Claude can access Google Workspace through a connector, meaning a link between the assistant and an external service. According to the report on Claude’s expanded Google connector, that connection covers Gmail, Google Calendar, and Google Drive. The main change is that Claude can now create, forward, and send emails rather than merely search for them.
In Calendar, Claude can create appointments, edit existing entries, and delete events that are no longer needed. When setting up a meeting, the assistant can also check every participant’s availability, according to the report. A practical workplace example would be organizing a project call: Claude finds an open time for the group, adds the appointment, and prepares the invitation.
The interface is not always the work app itself. In many cases, you give the instruction to an AI assistant, which then accesses a connected inbox, calendar, or other service. It can still feel like working inside your familiar environment because you no longer need to transfer information manually between several windows.
This development is not limited to large companies. HoneyBook has released a Claude connector for its platform and aims to bring agentic AI, meaning systems that can complete several steps of a task, to independent businesses. However, the brief report on the HoneyBook connector provides neither a detailed feature list nor pricing, so its practical scope remains unclear from the supplied information.
Messages and speech become interfaces
OpenAI is taking a similar approach to Apple Messages with ChatGPT. Its new plug-in can sort, analyze, and edit messages as well as search through conversation history. It can also draft texts, delete messages, and send them on a user’s behalf; according to the report, the connection works with Codex and ChatGPT Work as well.
OpenAI provides a straightforward everyday example: ChatGPT can review messages received the previous day and suggest suitable follow-ups for your contacts. Instead of opening every conversation separately, you could ask it to find unanswered messages and prepare drafts. The coverage of the Apple Messages plug-in also makes clear that automatic sending can remove your last opportunity for a manual review.
Meta is additionally treating speech as a universal input method. Its new Meta AI app for Mac includes system-wide dictation that, according to the company, works across applications. The app can also inspect what is currently visible on the screen and answer questions using that context.
For business owners, Meta AI can connect Instagram and Facebook accounts, advertising campaigns, and Google Workspace apps. The report about the new Mac app says the assistant can summarize campaign performance and audience engagement while also drafting presentation decks, documents, and spreadsheets. A marketing team could, for example, compare successful posts and turn the available data into an initial presentation draft.
Meetings and coding get dedicated assistants
AI is also intended to reduce the work left after a meeting. Calendly’s new note-taking assistant joins calls, records audio and video, and produces a transcript, which is a written version of the conversation. It can then generate a summary, action items, and drafts for follow-up emails.
Calendly is also testing a feature that uses system audio for transcription and plans to release an assistant called Callie. Callie is intended to use Calendly’s existing meeting and scheduling data to check availability, arrange meetings, and retrieve context from earlier conversations. According to the report on Calendly’s note taker, the target users include people in sales and marketing whose working days are packed with calls.
The market is already crowded, however. The report names Granola, Fireflies, Read AI, Otter, and Fathom, alongside note-taking features from Notion, ClickUp, and Wispr. Calendly is therefore trying to distinguish its product through the work that happens after transcription, not through recording alone. Participants are notified about the recording in the meeting chat and can ask the assistant to leave the conversation.
Slack applies the same idea to programming tasks. Slack Code provides open, project-specific channels where teams can work with AI agents; in this context, an AI agent is an assistant that carries out a task across multiple steps. With so-called vibe coding, people describe the desired result largely in natural language while an assistant creates or modifies the program code.
A team can tag an agent such as Claude or Devin to open a channel for a new feature, a website update, or a bug fix. Inside that channel, team members can compare changes, preview the output, give feedback, and approve the result before release. Slack Code is available on every Slack plan, according to the report; its channels archive themselves when an assignment is complete and retain an audit log.
Pros and Cons of direct AI actions
Pros:
- Less manual transfer – You no longer need to copy information from emails, calendars, meetings, and messages into a separate chat by hand.
- Faster follow-up – Summaries, action lists, and reply drafts can be created directly from the existing work context.
- Clearer collaboration – Visible code channels with previews, feedback, and approval steps make an assistant’s work easier for a team to inspect.
- Simpler input – System-wide dictation can speed up longer texts and instructions without requiring a separate speech feature in every app.
Cons:
- No final review – Automatically sent emails or messages can deliver incorrect claims, an unsuitable tone, or content to the wrong recipients.
- Broad data access – Inboxes, message histories, calendars, screen contents, and meeting recordings often contain sensitive information.
- Unclear processing – OpenAI says its Messages plug-in runs locally and does not build an index of all messages, but the precise implications remained unclear in the report.
- Errors become actions – A misunderstood instruction can do more than produce a poor answer; it can delete an appointment, send a message, or change code.
What this means for you
If you are a beginner, start with a reversible task. You could ask an assistant to prepare a reply from an email or summarize a meeting while keeping control over sending, deleting, and calendar changes. This lets you judge how reliably it understands your context and tone before an error can affect other people.
If you already have experience with AI assistants, you can connect clearly bounded stages of a workflow: prepare a meeting based on availability, extract tasks afterward, and draft the follow-up email. You will get more value when responsibilities, permitted data sources, and approval points are defined in advance. Slack Code’s visible changes and audit logs illustrate how such controls can be embedded in the workflow itself.
The sources provide no Switzerland-specific details about availability, supported languages, or data processing. They also give almost no concrete pricing information; Slack Code is the sole feature described as being available across all Slack plans. For business inboxes, customer information, and meeting recordings, Swiss users and organizations therefore still need clarity about their internal privacy rules and where connected services process data.
AI assistants are becoming more useful because they no longer stop at making a suggestion and can instead complete real steps of a task. That is also the central risk: a convincing error can become an actual message, a deleted appointment, or modified software code. For now, the balanced approach is automation with visible intermediate steps and human approval for anything that is difficult to reverse.
Sources
- Claude in Gmail: So verschickt die KI jetzt Mails für dich – Publication unknown, 2026-08-19
- ChatGPT can now send texts for you with new Apple Messages plug-in – Publication unknown, 2026-08-20
- Meta AI’s new Mac app wants you to talk to your apps – Publication unknown, 2026-08-20
- HoneyBook bets on agentic AI to streamline small business operations with its new Claude connector – Publication unknown, 2026-08-20
- Calendly throws its hat into meeting note-taker circus – Publication unknown, 2026-08-19
- Slack is launching collaborative vibe-coding channels – Publication unknown, 2026-08-20
