What changed
Claude can now build a dashboard from your company data and turn a report into a short animated video
On October 8th, Anthropic added two tools to Claude, both in beta: Dashboards and Motion.
Dashboards connects Claude to where your company data lives, such as BigQuery, Snowflake or Databricks (systems that store large amounts of company data), or to a sales tool like Salesforce. You ask a question in plain words and Claude builds a dashboard that updates as the data changes.
Click any number and Claude shows the query behind it (the request it sent to your data), and each chart lists when its data was last refreshed. Anthropic says dashboards suit quick questions, like how this week's signups compare with last month's. For deeper analysis, you can send a dashboard on to analytics tools such as Sigma or Mixpanel.
Anthropic's announcement names data platforms and tools like Salesforce, and it doesn't say spreadsheets or accounting software work. If your numbers live in Google Sheets or Excel, ask your admin or check Anthropic's help pages before you plan around Dashboards.
Motion turns a report into a short animated explainer, such as a 30-second summary of a quarterly report for an all-hands meeting, which you can download as a video file. Claude builds it from your own text, charts and numbers, with no generated footage and no AI-generated people.
Dashboards is in beta on paid plans, and Motion only on Team and Enterprise plans. Anthropic also says both are off by default and are switched on in your organization's settings, so someone with admin access has to do it.
Try: if your sales data is in Salesforce or a data platform, ask your admin to switch Dashboards on, then ask Claude for this month's sales by owner. Click one number and look at the query behind it.
Claude can now work inside your Google Docs, Sheets and Slides
On October 6th, Anthropic released Claude for Google Workspace in public beta on all paid plans. It's an add-on that opens Claude in a sidebar next to the file you have open. Claude can read the document, sheet or deck, see what you've selected, and change it in place.
In a document, it can fix a sentence or restyle a heading without touching the rest of the formatting. In a spreadsheet, it writes formulas and builds pivot tables and charts. In a deck, it builds slides in the deck's own layouts and checks for text that runs off the slide.
You choose how much it does on its own. In the default "Ask before edits" setting, every change waits for your approval. In "Accept all edits", Claude works through the task and applies changes without stopping.
You can also start from Claude and ask it to create or edit a Google file from the chat. That uses connectors, the links between Claude and your Google files, and Claude's access matches your existing Google sharing. On Team and Enterprise plans, an owner has to switch the connectors on first.
Try: install Claude for Google Workspace from the Google Workspace Marketplace, open a budget sheet, leave "Ask before edits" on, and ask it to flag anything over budget.
Claude Code's advisor setting now works with Haiku 5.5, Anthropic's small, fast model
Claude Code, the version of Claude built for writing software, has an advisor setting: your main model does the work and, at key moments, asks a stronger model for advice. Those moments are before it commits to an approach, when it's stuck, and before it says a task is done.
The advisor has to be at least as capable as your main model. On October 7th, Haiku 5.5 became usable as the main model or the advisor, and it's the model you get when you choose Haiku. The helper assistants Claude Code sends off for side jobs use the same advisor setting. Anthropic says the advisor fits long, multi-step tasks where most turns are routine but the quality of the plan decides the outcome.
We run Sonnet (a mid-size Claude model) for the main work and Opus (a larger one) as the advisor, with Haiku doing the helper jobs. Anthropic labels the advisor experimental. It works with a Claude subscription or an Anthropic API account (the pay-as-you-go kind), but not through cloud providers such as Amazon Bedrock.
The advisor reads the whole conversation each time it's asked, so it can use up more of your plan's usage.
Try: if your team uses Claude Code, open it, type /advisor and pick a model stronger than the one you work in. Claude Code remembers your choice the next time you open it.
Source: Claude Code docs, "Escalate hard decisions with the advisor tool" · Claude Code changelog, version 2.1.293, October 7th
One thing worth thinking about
Fix the path before you speed it up
Vivian Lee and Linda Mantia, writing in Harvard Business Review with Jon McNeill, argue that if you put an AI agent (an assistant that carries out a step on its own) on each step of the process you have today, you can end up with a faster version of a process nobody would design now. Lee's phrase for this is paving the cow path: paving the winding route cows happened to wear into the ground.
Their alternative, which they laid out in a Harvard Business Review webinar on September 2nd: write down the outcome first, map the steps you have now, then decide for each step whether an AI agent or a person should own it.
Mantia gave one example in the webinar: NTT Data's process for answering requests for proposals. She said its AI agent writes a first draft in 20 minutes, work that used to take tens, if not hundreds, of hours across the company, and that people still decide what to promise the client. She also said NTT Data now bids on more work and wins more of what it bids on.
Her warning is for companies with decades of products, acquisitions and systems underneath them. She calls that build-up organizational plaque, and a company can run hundreds of successful AI projects while most of what its customers see stays the same.
Try: take one process you're thinking of handing to an AI agent, write down the outcome you want in one sentence, then list today's steps and mark which ones a person should still own.