ET RU EN

Blog

Spreadsheets, slides and calls with AI: which office routine you can already hand to an assistant

Published · Prepared with AI

Anthropic's Claude is close to having a full office suite. Slides, documents and design, already released, are now joined by spreadsheets and whiteboards in beta, and an assistant for video calls has been announced. This autumn the loudest debate is about which model is smartest. But for a small business owner or SMM specialist in Estonia, a different question is more useful: which daily routine an AI assistant already takes off their hands. Below is a short list of such tasks: spreadsheets, presentations, documents, call notes and simple yes/no decisions on enquiries. Next to each is a plain note on what works today, what is in beta and what has only been announced so far.

A new model tops the rankings: what does that change in your working day?

The news is hard to miss. On 9 November Anthropic released Claude Fable 5.2, and the model went straight to first place in the Intelligence Index, a ranking by Artificial Analysis. It beat Google's Gemini 4 Carbon by 5 points, reports @rowancheung.

From this it is easy to conclude that the main choice is to find the smartest model and stick with it. But for a spreadsheet of a month's posts or a draft report for a client, ranking position matters less. What matters far more is where the model is built in and how close to a finished result it takes the work.

That said, the same news contains figures you feel every day. According to the same source, Fable 5.2 is 20% cheaper and up to 30% faster than the previous version, 5.1. So it is worth looking at the tools around the model: which of your everyday tasks already fit into them.

What is already in Claude's office suite, and what is still in beta?

According to @rowancheung, Anthropic has released Claude Sheets and Whiteboards in beta, opened Claude Data to everyone and announced Claude Meetings. Together with Slides, Design and Docs, released earlier, that makes an almost complete set of office tools.

A beta is a trial version that is still being refined. You can already use it, but its features and appearance may change. The status of each tool today:

  • Slides - presentations: released
  • Docs - documents: released
  • Design - design: released
  • Claude Data - open to everyone
  • Sheets - spreadsheets: beta
  • Whiteboards - boards for diagrams and plans: beta
  • Meetings - an assistant that joins video calls: announced only

Which office routine can you hand to an assistant right now?

The rule of thumb is simple: the task comes up every week, has a clear result, and you can check it in a couple of minutes. That description covers much of what eats up the evenings of small business owners and SMM specialists:

  • Spreadsheets: pull a month's posts, enquiries or expenses from scattered notes into one table. In Claude this is Sheets, still in beta.
  • Presentations: put together a draft client report or meeting slides from your key points. Slides is already out.
  • Documents: a first version of a brief, a commercial proposal or instructions for a contractor. Docs is already out.
  • Plans on a board: lay out a content plan or a campaign launch step by step. Whiteboards is in beta.
  • Call notes: Claude Meetings has only been announced. Until it launches, a meeting summary is easier to build from your own notes in Docs.
  • Simple yes/no decisions: sort incoming enquiries into suitable and unsuitable by your own rules. Google now has a dedicated tool for this, covered below.

How does AI make yes/no decisions on enquiries and leads?

Google has introduced Gemini Decision 1, a model for decision-making, and with it the Gemini Decisions API, writes @OfficialLoganK. An API is a programmatic entry point through which services connect to each other. This one answers three kinds of question: yes or no, a choice between several options, and scoring against a rubric, that is, against criteria set in advance.

The model accepts text, images, audio, video and PDF, and attaches a confidence score to every answer. According to @_philschmid, a single request can hold up to 1000 questions and a context of up to 256 thousand tokens (tokens are the chunks of text used to measure volume). The answer arrives in 150-250 milliseconds, a fraction of a second.

For a small business, it works like sorting the post. Website enquiries, social media messages and client questionnaires are sorted by your rubric: “fits the service”, “needs a call”, “too early for now”. You can try it in the API for free. In production, input costs $0.10 per 1 million tokens, and the model's answers are free.

But this is a tool for developers: it lives inside the Gemini API. It is connected to an enquiry form or a CRM (customer management software) through an automation service or with a developer's help. That is why the confidence score is especially useful here: enquiries with a confident answer move on by themselves, while doubtful ones go to a person.

Where will the assistant fall short for now, and what can you do about it?

There are three limitations, and each has a workaround.

First: Sheets and Whiteboards are in beta, and their features may change. What to do: keep drafts and working tables there, but store the final version for the client wherever you store it now.

Second: Claude Meetings has only been announced, and the news gives no release date. What to do: until it launches, write three lines after each call - decisions, tasks, deadlines - and give them to the assistant in Docs to turn into a follow-up email.

Third: yes/no decisions through the Gemini Decisions API require a connection to be set up, and a mistake in the rubric will repeat across every enquiry. What to do: start with a free trial on past enquiries where you already know the right answer, and compare the result with your own decision.

In all three cases, the final say rests with the person running the assistant. The assistant prepares a draft; a person checks the figures, names and tone before anything goes to the client.

The bottom line: for daily work, what counts is which tasks the assistant takes to a finished result. Presentations and documents - today; spreadsheets and boards - in beta; calls - once Meetings launches; sorting enquiries - through the Gemini Decisions API. One step for this week: list last week's recurring tasks and mark where each one falls - a released tool, beta, or still waiting. Start with one task from the first group. If your question goes beyond the list and concerns how to restructure your team's work or sales, you can talk it through with MOST experts.

Sources

Questions and answers

Which office tasks can I hand to AI right now?
Ones that repeat every week and are easy to check: draft presentations and documents, summary spreadsheets, plans on a board and sorting enquiries by your own rules. Claude's presentations and documents are already out; spreadsheets and boards are still in beta.
What is in Claude's office suite?
Slides for presentations, Docs for documents and Design are already out. Sheets (spreadsheets) and Whiteboards (boards) are in beta, and Claude Data is open to everyone. Claude Meetings, an assistant for video calls, has only been announced so far.
Can I already hand call notes to AI in Claude Meetings?
Not yet: Meetings has only been announced, and the news gives no release date. Until then, a simple trick helps: after a call, write down decisions, tasks and deadlines in three lines, and the assistant in Docs will turn them into a follow-up email.
Why would a small business need the Gemini Decisions API?
To sort enquiries and leads by your own rules: yes/no, a choice between options, scoring against a rubric. The model reads text, images, audio, video and PDF and adds a confidence score to every answer, so doubtful cases go straight to a person. It is connected to an enquiry form or CRM.
Do you need the smartest model for office routine?
For a draft report or a summary spreadsheet, ranking position matters less than the tool the model is built into. Speed and price are more noticeable: the new Claude Fable 5.2 is 20% cheaper and up to 30% faster than version 5.1.
Can I try the Gemini Decisions API for free?
Yes, you can try it in the API for free. In production, input costs $0.10 per 1 million tokens, and the model's answers are free. Start with past enquiries where the right answer is already known: that way you see straight away whether the rubric matches your own decision.

← All blog posts