top of page

AI Team Collaboration: Don't Let AI Be Just a Personal Plug-in

5 hours ago
4 min read
Six colleagues sitting around a meeting table with laptops open, discussing together
Bring AI to the meeting table, not just keep it on everyone's laptop. (Photo: Ngital / Unsplash)

Does this sound like your company? The owner uses AI to write proposals, sales uses AI to reply to customers, accounting uses AI to reconcile the books. Everyone says they're faster, yet meetings are just as long and decisions are still lost somewhere in LINE groups. This is exactly where AI team collaboration most often gets stuck: AI is treated as a personal plug-in and never makes it into the way the team works together.


A few recent pieces of news are worth reading together. One is a Harvard Business Review (HBR) study of more than 300 managers on how AI fits before, during and after meetings. The other is Anthropic bringing Claude directly into Google Docs, Sheets and Slides. One is about method, the other about tools.


Today's Highlights: AI team collaboration starts before and after meetings


Research: managers mostly use AI on their own


Rosani and Farri of Capgemini Invent's Management Lab wrote in HBR, based on research covering 35 organizations and more than 300 managers, that most managers are used to using AI privately to prepare, but rarely bring it into meetings, workshops and other settings where people think together. Organizations pushing AI also focus mostly on individual tasks, even though much of the real work gets done in teams.


They split teamwork into three stages: before the meeting, use AI to prepare the agenda and materials and stress-test ideas; during the meeting, let AI challenge assumptions and add perspectives no one thought of; after the meeting, use AI to help review and turn lessons into practices you can reuse next time.


The authors stress two conditions. First, intentionality: only team leaders can decide how the team meets and works, so if leaders don't actively change things, AI never gets in. Second, craft: you have to give AI the right context and well-designed prompts for its output to be useful.


Tools: Claude moves into shared documents


On October 6, Anthropic launched Claude for Google Workspace as a public beta, available on all paid plans. Claude appears as a sidebar next to Google Docs, Sheets and Slides: major rewrites in Docs become suggestion cards that people choose to apply or skip; in Sheets it can write formulas and build pivot tables and charts; in Slides it keeps the existing layouts and themes.


For teams, three details matter most: by default it asks before every edit; which files Claude can see follows your existing Google sharing permissions; and on Team and Enterprise plans, the owner has to enable the connector first.Anthropic also gives the example of saving a deck format as a skill so the whole team can reuse the same format.


Trend: from personal tricks to shared skills


The Claude Help Center release notes point in a consistent direction: in mid-September, Claude merged regular chat and Cowork into a single entry point and introduced Claude Docs, "living documents" that several people can work on together (a beta feature on Enterprise, off by default). The usage reports launched for Enterprise on September 10 highlight a team's Claude costs, sticking points, and repeated practices worth turning into shared skills.


Jesse's Take: Start by getting one document truly shared


What I see most often when advising small businesses is this: everyone uses AI, but each in their own silo. Good prompts are saved on individual computers, and customer data is scattered across LINE, Excel and the owner's head. Individuals do get faster, but the team doesn't, because AI plays no part in the handoffs, meetings and alignment that eat up the most time.


Suggestion 1: Start with one regular meeting, not with a whole system


HBR's three stages fit small companies well. Before the meeting, ask AI to turn last week's to-dos and this week's topics into a one-page agenda. During the meeting, when there's disagreement, ask it to list the arguments on both sides. After the meeting, turn the notes into a three-column table: decision, owner, deadline. You don't need a new system; one shared folder is enough.


Suggestion 2: Save what works as the team's templates


A common scenario: at a trading company of about 20 people, one sales rep writes especially clear quotation notes because she has her own finely tuned prompt, but nobody else knows about it. Instead of everyone figuring it out from scratch, turn her approach into a shared team skill or template that new hires can use on day one. Practices that keep repeating are the ones most worth turning into shared assets.


Suggestion 3: Agree on permissions and reviews before you start


Once AI gets into shared files, the question becomes "who can edit, and who is responsible for the edits?" I suggest starting with "ask before editing" everywhere, with important files applied only after the owner confirms. Since Claude's access follows your Google sharing settings, spend an afternoon tidying up shared folders first. Otherwise AI might see the payroll sheet it shouldn't, but not the quotation it should.


An honest caveat: vendor feature announcements show the most ideal use cases. In practice you'll run into broken formatting and permission snags. And HBR's research mostly covered managers in large organizations, so a 20- or 30-person company will need to adjust the pace.


A first step you can take today


Pick one regular meeting this week. Afterwards, spend ten minutes pasting the meeting notes into Claude and ask it to turn them into a "decision, owner, deadline" table, then save it in a shared folder everyone can see. Start the next meeting from that table.AI team collaboration doesn't have to happen all at once. Getting one document truly shared by everyone is the best place to start.


Further Reading


Related posts on JCXO


Original sources

 
 
 

Comments


©2025 | Omni Optimize Your Biz | JCXO Consultancy | JC Capital

Thanks for subscribing!

bottom of page