Skip to content
Important: We're now taking requests for PHP2Shell emergency response —Learn more →
SolanaLink LogoSolanaLink
After Excel, the Agent: What Anthropic's New Office Tools Say About AI Agents in Japanese Workplaces

After Excel, the Agent: What Anthropic's New Office Tools Say About AI Agents in Japanese Workplaces

Image generated by xAI Grok
Development
17 min read
0
0 comments
544 views

On November 10, Anthropic launched Sheets, Whiteboards and Data, and Meetings reportedly went into testing. The tools are still basic, but they reverse the old order of adding AI to existing tools. Drawing on how Japanese companies prepare their data, where they keep it and how they govern AI, this article forecasts how fast AI agents will enter Japanese workflows and offers five practical steps for small and mid-sized firms.

In Brief

On November 10, 2026, Anthropic added a spreadsheet (Claude Sheets), a whiteboard (Claude Whiteboards) and a data dashboard (Claude Data) to Claude. According to reports, its meeting assistant, Claude Meetings, also went into testing. The company was nine days away from its planned Nasdaq listing.

My view: the tools themselves are still basic, and they will not replace Excel any time soon. But they reverse the order of office software. Until now, AI was added to existing tools. Now the AI already holds your data and context, and the tools grow out of it.

For Japanese companies, how fast AI agents enter real workflows will not depend on how capable the models are. It will depend on three things: whether the data is in a shape AI can read, where that data is kept, and who presses the final "approve" button.

On those three conditions, here is my forecast. Paperwork such as meeting minutes and email goes to agents first, and becomes routine around 2027. Spreadsheet roll-ups and monthly reports come next. Cross-system processes such as ordering, expenses and invoicing follow from 2028 onward. Decisions that carry external accountability stay with people.

1. What Anthropic Launched on November 10

ProductWhat it doesStatus
Claude SheetsBuilds a spreadsheet with multiple tabs, formulas, formatting and drop-down menus from a single request; the team and Claude edit it together, and it exports as an Excel or CSV fileBeta, all paid plans (off by default on Enterprise until an admin turns it on)
Claude WhiteboardsA shared board for team discussion; Claude turns the discussion into a timeline with owners and dates, which can later become a document or slide deckBeta, all paid plans (off by default on Enterprise until an admin turns it on)
Claude DataFormerly Claude Dashboards; connects to sources such as Snowflake and Salesforce to build dashboards and data storiesGenerally available
Claude MeetingsJoins video calls as a participant, records the discussion and updates the documents the team is working onAnthropic's blog says "coming soon"; several outlets report a beta on Team and Enterprise plans, currently on Google Meet only
Interactive connectorsOpens partner apps such as Adobe, Atlassian and Box directly inside ClaudeAvailable

Anthropic positions these tools modestly. Nicholas Lin, one of Anthropic's product leaders, told Bloomberg that Sheets is "fairly basic" for now and is not meant for work such as financial modeling; for more advanced enterprise use, Microsoft's and Google's products still have their place. Anthropic's own announcement says the spreadsheets have simple functionality for now, with more to come.

The market reacted anyway. On the day of the launch, Microsoft's shares fell about 3% at one point during trading and closed down 2.2%.

The timing is worth noting too. The same day, Anthropic filed an amended prospectus with a price range of $70 to $78 per share. According to reports, it plans to price the offering on November 18 and begin trading on Nasdaq on November 19 under the ticker ANT. Japan is one of the markets in the offering: shares sold to investors in Japan are capped at about ¥552.9 billion, and Mitsubishi UFJ Morgan Stanley, Mizuho, Rakuten and SBI are the Japanese distributors.

Please note: this article mentions Anthropic's IPO, share prices and valuation only to explain the context of the launch. Nothing here is investment advice or a recommendation to buy or sell any security. Make your own investment decisions.

2. The Underlying Order Has Flipped: From "AI in the Tool" to "Tools Out of the AI"

What matters most in this launch is not the spreadsheet. It is the order.

Over the past decade or so, Google Sheets, online documents and Airtable all challenged Office without shaking Microsoft's position. The reason, as I see it, is simple. They changed where files are stored and how many people can edit at once, but not the logic of the work itself. People still find the data, clean it up, build the table and report on it, in that order.

Microsoft's own AI strategy adds a layer on top of that order: Copilot inside Excel and Teams. That approach is working. In the fourth quarter of fiscal 2026, Microsoft's Productivity and Business Processes segment, which includes Office, brought in $37.8 billion in revenue, up 14% year over year, and Microsoft 365 Copilot passed 30 million paid seats.

Anthropic is doing the opposite. Claude becomes the entry point for work first: employees already write code and documents and do research there, and they have connected it to the company's data warehouse and CRM. Sheets, whiteboards and meetings grow out of that accumulated context. When you open a sheet, the AI does not start from zero. Based on the data and conversations you have already given it, it first lays out what you want to see and what is worth seeing. It is not making things up; it works from data you have entered or connected.

This reminds me of the story of drones and action cameras. In January 2013, DJI launched the Phantom 1 with no camera of its own. It came with a GoPro mount instead, and users attached their own GoPro to shoot aerial footage. That October, the Phantom 2 Vision became the first Phantom with a built-in camera, and in the generations that followed, DJI designed the camera and the aircraft as one product. GoPro did not launch its own drone, the Karma, until 2016, and it left the drone business in 2018. Once aerial footage went from a toy to a necessity, an integrated design beat a bolted-on device by a wide margin.

Office software may be reaching the same point. When AI is doing much of the work, growing spreadsheets and meetings directly out of the AI is more natural than bolting AI onto old tools.

None of this means Microsoft will disappear. Excel's depth in complex modeling, and the fact that it works offline, remain important for work that should not touch the internet or the cloud. Anthropic itself says that when you need Excel's full functionality, you can export to Excel or use an add-in such as Claude for Excel, and Lin said add-ins for existing apps work better for jobs like financial modeling. In the short term, I expect the bigger impact to fall on spreadsheet and document tools whose main selling point is online collaboration, because Claude can now do that too.

3. Where Japanese Offices Really Stand

How fast will this play out in Japan? Start with the baseline.

The Ministry of Internal Affairs and Communications' 2026 White Paper on Information and Communications, published on July 24, 2026, found that 86.4% of Japanese companies now use generative AI in at least one area of work. Some 68.9% have a policy of using it "actively" or "in limited areas," up from 49.7% a year earlier. On whether they use it at all, the gap with the US, Germany and China has narrowed sharply.

On how they use it, the gap remains:

IndicatorJapanUSChina
No organization-level initiatives27.0%1.4%2.6%
An environment for learning generative AI skills39.9%62.7%71.7%
An internal database that AI can draw on24.5%57.2%73.0%

The white paper also shows that the most common use in Japanese companies is the category of help with meeting minutes and email. Around 70% of companies use AI there, and around 70% feel a real effect, both the highest of any category.

Other surveys paint a similar picture. In Teikoku Databank's March 2026 survey, with valid responses from more than 10,000 companies, 34.5% use generative AI in their work, including 46.5% of large companies, 32.4% of SMEs and 28.0% of small businesses. According to Kyodo News, in a PwC survey of large companies in six countries, only 9% of Japanese respondents said AI had greatly exceeded expectations, the lowest of the six (PwC's own report, counting a different group of respondents, gives 10% for Japan and 42% for the US). In Confluent's survey of 14 countries, just 17% of Japanese companies were running AI agents in production, the lowest of the 14.

Large companies are already moving. Rakuten used Claude Code to cut the average time to bring a new feature to market from 24 working days to 5. NEC provides Claude to about 30,000 employees across its group. Hitachi formed a strategic partnership with Anthropic in May 2026 to apply Claude across the business processes of a group of about 290,000 people.

Put these numbers together and my reading is this: Japanese companies are not avoiding AI. They are using it at the edges, polishing minutes and emails, without yet letting it into the core of their work. What holds them back is the context described in the previous section. Only about one company in four has an internal database that AI can search. Without data, even a very capable agent can only do writing work.

Time is not on the side of waiting. Recruit Works Institute projects that Japan's labor supply will fall short by about 3.41 million workers in 2030 and about 11 million in 2040. A shortage on that scale turns "should we use agents?" into "which processes have to go to agents?"

4. Where the Data Lives: Japan's Preference for Keeping It Local

The Japanese companies I have worked with feel very strongly that their data must stay in Japan. That is not a bad thing, but it will shape how agents are adopted.

The good news is that the model layer already offers options. Amazon Bedrock offers Japan cross-region inference, which routes Claude inference requests only to its Tokyo and Osaka regions. Anthropic's regional compliance page also says that regional data residency is available across Asia-Pacific, including Japan, through cloud platforms such as AWS Bedrock and Google Cloud Vertex AI.

The other side needs to be clear as well. Sheets, Whiteboards and Meetings all run in the claude.ai layer. Anthropic's privacy center says that data for commercial products such as Claude Enterprise is stored in the United States. As of this writing, I have not found anything in the public documentation that lets these new tools keep data in Japan.

So I expect Japanese companies to move toward a two-layer setup:

  • A confidential data layer: customer, financial and HR data go through APIs or Bedrock with inference inside Japan, handled by agents built into the company's own systems or by partners.
  • A collaboration layer: meetings, whiteboards and general material use SaaS such as claude.ai, as far as the company's data classification rules allow.

The signals from vendors are clear too. In 2024, Oracle announced it would invest more than $8 billion in Japan over ten years and expand its Japan-based operations and support teams to help customers meet digital sovereignty requirements. Mitsubishi UFJ Financial Group has a multi-year global agreement with AWS, and moving to the cloud has cut the group's IT infrastructure costs by 20% compared with on-premises systems. Anthropic opened its Tokyo office in October 2025. According to Impress Watch, Hidetoshi Tojo, president of its Japanese entity, said demand from Japanese customers for partnerships is especially high, and many want to contract through resellers or technology partners. In early 2026, Nomura Research Institute expanded its partnership with Anthropic Japan and began building Claude implementation services for the Japanese market.

In my experience, Japanese B2B customers pay reliably and like systems tailored to their own processes. The part of the work that model vendors cannot do on their own is exactly where partners and system integrators come in.

5. Who Presses "Approve": Governance and the Ringi Culture

The third condition is governance. On March 31, 2026, the Ministry of Internal Affairs and Communications and the Ministry of Economy, Trade and Industry published version 1.2 of the AI Guidelines for Business. For the first time it explicitly covers AI agents. It stresses building ways for people to step in on important decisions and limiting the tools and systems an agent can reach. The guidelines are not legally binding, but in Japan they have become an important reference for corporate governance.

That fits well with ringi, the approval process Japanese companies know well: agents draft, compile and cross-check; people approve and remain accountable to the outside world. Claude Meetings reflects the same caution. According to reports, Claude will not join a meeting that includes external guests unless an admin allows it, and it leaves automatically if an external guest joins partway through.

So I believe the agents that will spread fastest in Japan are not fully autonomous ones. They are agents that write a good ringi document and then wait for a person's stamp.

6. Forecast: Four Stages of AI Agents in Japanese Workflows

The table below is my own reading of the data above, not a forecast from any institution.

StageTiming (author's forecast)Work that moves to agentsPreconditions
1. PaperworkNow to 2027Meeting minutes, email, internal Q&A, assigning follow-up tasks after meetingsLargely in place; meeting agents turn minutes directly into tasks
2. Spreadsheets and reports2027–2028Excel roll-ups, monthly reports, sales and inventory dashboardsAn internal database AI can read; data classification rules
3. Cross-system processes2028–2029Ordering, expenses, invoicing, HR administrationSystem connectors (such as the NetSuite and Workday plugins) and human approval points
4. Department-level agent teamsAround 2030A few people lead several groups of agents to run a departmentA widening labor shortage and mature governance rules

Three signals are worth watching next: when Meetings adds support for Teams and Zoom; when the new claude.ai tools offer data residency in Japan; and when the share of Japanese companies with an internal database that AI can draw on passes half.

7. Five Steps for Japanese SMEs

  1. Start with processes that already show results. Going from meeting minutes to task assignment is the best starting point; it is the category where the white paper found the strongest felt effect.
  2. Get your data in order before you talk about agents. Gather the material scattered across Excel files, email attachments and shared drives into a knowledge base AI can search.
  3. Classify your data. Send confidential data through inference inside Japan, use SaaS for general collaboration, and write the rules down.
  4. Spell out where people must approve. Decide in advance which actions an agent may take on its own and which need human approval, using the AI Guidelines for Business (version 1.2) as a reference.
  5. Know the costs and use subsidies. Claude Pro is $20 a month. Meetings requires a Team or Enterprise plan; a Team Standard seat is $25 a month, or $20 a month billed annually (check the official site for current prices). SMEs bringing in IT tools with AI features should look at the Digitalization and AI Adoption Subsidy 2026 (デジタル化・AI導入補助金2026), the successor to the IT Introduction Subsidy.

Closing

The demand for AI, and the efficiency it brings, no longer needs debating. The real questions are how much faster it can go and how many industries it can change. Judging by this launch, it is already moving fast enough.

For Japanese companies, rather than waiting for the perfect tool, the better move is to get your own data and processes ready first. When the agents arrive, how much work they can take on will depend on what you have prepared for them.

Disclosure: The author's company uses several AI tools, including Claude, in its day-to-day development work.

Disclaimer: This article is for information only and is not investment advice. Details about the IPO, share prices and valuation come from public reports and may change. Make any investment decision yourself, and consult a professional where needed.

About SolanaLink

SolanaLink Co., Ltd. is a Tokyo-based IT services company offering cloud solutions, web and mobile development, IT operations, and AI agent implementation. If your team wants help working out which processes to hand to AI agents, building an internal knowledge base AI can search, or setting up agent systems with inference inside Japan, contact us at info@solanalink.jp.

References

Launch and market reaction

IPO

Generative AI in Japanese companies

Data location and vendors

Governance, subsidies and pricing

The drone example

Comments

Comments (0)

#Productivity#AI#Japan Market
Back to News & Blog