I came across this post on LinkedIn and I’m curious to hear the hive-mind’s perspective on this.
Microsoft’s (and many other traditional software vendors’) bet that people will want each program or application suite infused with its own focused AI is looking like a risky one in the face of Codex, Antigravity, and Claude Code. It is becoming clear that, with good enough models, people will want to delegate to an agent that uses tools and works across apps (or builds its own) to do tasks. Why open individual applications and use disconnected AI tools with limited context? The applications were always a means to an end, not a goal in themselves. — Ethan Mollick, Associate Professor at The Wharton School. Author of Co-Intelligence
If you are designing nuclear weapons you need to keep your workspace isolated.
If you are handling confidential medical records you need to pay extreme attention to HIPAA compliance issues - which likely make data integration impractical or unsafe.
If you are balancing your checkbook or planning a family reunion integration is wonderful.
I think in general many users cut their nose off to spite their face by being paranoid about security issues to the point where it limits their progress. In reality it all depends what type of data you have.
It appears that Professor Mollick is saying, in light of the direction that AI appears to be heading, individual apps may not be a long term solution for everyone.
I have very little knowledge of AI. But I studied business at university and had 17 years of experience in multiple industries before moving into I.T. And I think it’s not too soon to start planning how our tools and buckets of data will be able to work together without us exporting from one and importing into another, etc.
+1
I’m probably more careful with my personal data than I was with the company data I once protected. Executives are not normally fans of long, complex, passwords .
But I don’t worry about email because I can’t protect all the copies scattered around the Internet. And I keep my online data on Google Drive because Google is a business platform which is compliant with hundreds of organizations and regulations, including HIPAA.
And all my Google Workspace apps work together, including Gemini. So I think I’m as well prepared as I can be at this point. Especially because everything already runs pretty well on my Apple devices.
If you spent years mastering a particular app like Excel, is it really a good idea to give it up in favor of typing in some vague commands into a chat bot? I’m sure the Excel experts can run laps around people typing in paragraphs of text to get Excel to do something they can do in a few steps because they’re Excel power users.
No. But we are already seeing AI creating new spreadsheets and doing some limited analysis on existing spreadsheets. Is it a possibility that in a few years most spreadsheet users, with the help of AI, could be power users?
Claude in Excel is only available for Max, Team, or Enterprise plans, so I haven’t been able to take it for a test-drive, but it might be a case where integrating AI into an app is better than working through an all-purpose agent. I wouldn’t ask Antigravity or Claude Code to spin me up an intricate Excel workbook from scratch (yet). I have asked Claude Code to digest a batch of account statements and return a CSV file detailing changes in portfolio balances and composition over time, which I proceeded to load into an already existing Excel workbook. Interestingly enough, without being prompted to do so, Claude Code returned a message along with the CSV file that pointed out the exact points in time where either the portfolio balances or their composition deviated materially from the norm. So, I can see using an agent like Claude Code to gather and organize data and present it in a straightforward spreadsheet. I’m not sure it could replace Excel entirely if you’re doing anything complex.
But there are some use cases where I think it would be simpler to use an agent than to use an app with built-in AI. Batch file renaming is an example. I don’t need a file renaming app + AI—I just need the AI. After renaming batches of files to conform them to a complex naming protocol with a single natural language prompt using Claude Code, I’m never going back.
Another example: I’ve been using AI in DEVONthink because that’s where the files in question live, but I could have done most if not all of that work by pointing Claude Code at a selection of Finder folders. I didn’t need DT + AI for those particular tasks —I could have done it with the AI Agent alone.
I have seen similar reports in various places on the Internet.
I understand Claude Code makes it easy to point Claude Desktop at a particular folder and to set up standard instrucdtions in a Claude.MD file.
But you can also point the regular Claude Desktop chat at a particular folder via prompting if you activate the filesystem MCP integration. And you can have a standing general prompt to always follow instructions in a folder’s Claude.MD file. If you do this then you get the added benfit of MCP servers and other integrations not available in Claude Code.
Unless you are setting up a folder where you plan to execute code, what is the benefit of Claude Code?
Uhhhh … The Claude Code tab is there in the desktop app and all I have to do is select a folder from my directory tree and type a prompt into a box?
And … I didn’t realize how dead simple it has become to install the filesystem MCP integration into the desktop app. No fussing around in the terminal—just click the darn button right there in settings. So, I gave Claude a file-renaming task. After working on it for a few minutes, it suggested that the best way to proceed would be via a python script I could run myself. (Which certainly sounds better than boiling a lake just to rename files.) It wrote the script, told me the best way to use it, and off I go,
PS - If I’m reading the Claude Code Docs page correctly, MCP servers are available in Claude Code.
.