Chapter 6
Build a shared workspace for people and AI
Help multiple people and AI agents work toward the same result
By Chapter 5, you could run your own work with AI. This chapter applies the Goal Loop across an organization.
An organization's Goal Loop contains many smaller Goals
One person's Goal Tree may be modest in size. At company scale, it becomes more detailed.
Company objectives contain department Goals, projects, individual opportunities and each person's responsibilities. Every level has a Goal Loop with a desired state, current state and difference to resolve. The nested structure from Chapter 2 extends to the whole company.
Goal: Win the proposal
Understand the customer
Identify the problem
Prepare the proposal
Choose the structure
Complete review
Each Goal contains smaller Goals. Every loop works to bring the current state closer to the desired state.
More people also participate. The CEO, department heads, individual contributors and their AI agents all work within the same Goal Tree. Work becomes inconsistent if one participant relies on a different structure.
The Goal Loop itself is unchanged. The number of participants increases. Step 4, updating the current state with results, connects their work: Step 4 in a child Goal changes the information used in Step 1 of the parent Goal.
Existing coordination practices are insufficient for AI
Organizations have developed ways for people to stay aligned: chat messages, meetings and quick questions about progress. People have made those practices work.
AI cannot automatically participate in that informal coordination.
- It may be unable to determine which message contains the current decision.
- It does not know a meeting's decisions unless it has access to the record.
- It does not hear a conversation between desks or in the corridor.
As a result, each AI can treat a different point in time as the current truth. Everyone works diligently, but the results disagree. Adding people and AI agents increases the opportunities for inconsistency.
Four colleagues agree on a direction in a meeting. Back at their desks, each AI produces different work because the agreement has not reached every agent.
AI needs explicit access to the shared current state to participate in the same Goal Loop.
Consider what happens in a training operations team.
When the instructor and date changed
The team takes information sharing seriously. Everyone belongs to the same chat group and can access the shared drive. Each member has completed the first five chapters and works with their own AI. The arrangement appears sound.
Then the instructor and date change. The team posts the update in chat: “The September 10 session will move to September 12, with Sato replacing Tanaka as instructor.” Read receipts appear.
The team lead posts the new instructor and date in the shared chat. Read receipts make the lead confident that everyone has received the change.
For several days, work appears to proceed normally. Three people delegate their responsibilities to AI.
The invitation, materials and venue work proceed from different records. Each person and AI acts confidently using the information available to them.
- The invitation AI uses the previous invitation in the project folder. It still names Tanaka as the instructor.
- The materials AI uses an old template in the shared drive. The cover still says September 10.
- The venue AI checks the calendar and books the venue for September 10.
Each AI has used its available information correctly. None has received the chat message containing the change.
As the event approaches, the team compares its work.
The team brings together the invitations, materials and booking record and finds inconsistent details. All three people read the change, but their AI agents did not.
The invitation has the old instructor, the materials have the old date, and the venue is booked for the wrong day. The people had read the update. They delegated work to agents that had not.
The information was shared. The current state of the work was not.
Why a shared chat is not enough
Chat records the sequence of a conversation: an original date, a proposed change, a decision and further corrections. Old information, new information, discussion and confirmed decisions appear together in time order. Messages also omit context. “Apparently we are using instructor B.” “Understood.” People present at the time know which training session is meant. A later reader or AI may not.
“What is the correct date now?”
The answer is not recorded in one authoritative location. A reader must reconstruct it from the conversation. Human teams often correct misunderstandings informally: someone asks why a colleague is not preparing for a deadline, or checks whether they heard about the date change. AI lacks those unrecorded corrections.
The result is one job with different accounts of its current state. The problem grows as more people and AI agents participate.
Separate records give people and AI different accounts of the same work. Returning results to one Goal Tree lets everyone act from the same current state.
Three changes create a shared current state
If the problem is inconsistent current states, the solution is one shared current state. Three changes provide it.
- Share one Goal Tree. Put the team's work in the same structure. “Deliver the September 12 training successfully” can contain “Recruit participants,” “Prepare materials” and “Secure the venue.” Keep them as separate Goals because they have different constraints. A recruitment problem can continue within its Goal without involving the materials owner in every action.
- Assign each Goal to the people and AI it needs. The invitation owner and their AI handle invitations; the materials owner and their AI handle materials. Each runs its Goal using the approach from the first five chapters.
- Return results to the parent Goal. The update connects the separate work.
Work information in separate places
One Goal Tree with the current state
Use your usual tools and return results to one Goal. People and AI can then choose the next action.
“The materials are done” alone does not advance coordination. Unless the parent Goal's current state changes, the difference remains the same and the team's next action remains unclear.
Returning the result changes the next decisions.
Return the child Goal's result to the parent Goal
The parent Goal's current state changes
The remaining difference and next action change
Other people and AI act from the updated state
A completion message alone does not update the team's next action.
In the training example, record the instructor and date change once in the parent Goal's current state. The invitation AI, materials AI and venue owner can all reassess their next action from the same updated state. Their reference no longer depends on a message reaching each participant separately.
Results can also reveal a limit. The event owner might report, “The venue can hold no more than 500 people.” The parent Goal may then change to “Run the training on two dates.” Invitations change and another event operations Goal is created. A child Goal's result can change the parent Goal itself. That is the Goal update loop from Chapter 2 at organizational scale. Evidence from the people doing the work keeps the Goal Tree consistent with reality.
Everyone does not need a copy of every source. They need to use the same state of the work as their reference. Collaborative work means several people and AI agents running the same Goal Loop.
A shared system for Goals and current state
Most companies have informal authoritative sources: a board for management decisions, a spreadsheet for sales, and a list for ToDos. Those sources must become explicit enough for everyone's AI to use.
Software development provides an example of coordinated work. Hundreds of engineers can work on the same software from different computers with different tools. Git helps them combine their changes into a shared version. Each participant returns completed changes to the shared project.
Collaborative business work needs a shared system for two things:
- One Goal Tree: the team's work as a whole
- The latest current state of each Goal
Conversations and files can stay in external tools. Keep using email for invitations, calendars for schedules and the appropriate software for documents. After acting in those tools, record the result in the Goal.
Addness provides that shared system.
Addness lets an organization run the Goal Loop with people and AI.
Keep the Goals and current state of collaborative work in Addness. Continue executing the work in the tools you already use.
The process is brief: share the Goal Tree and current state in Addness, act in external tools, and return results to the Goal. Each person's or AI's work adds context to the Goal for the next participant. We call this a shared workspace designed for people and AI.
To connect your AI, follow the Addness MCP guide. To extend the approach to a team, read Operations lesson 1.
Compare the two arrangements:
- Before: People share chat and documents, but use different versions of the current state and produce inconsistent results.
- After: People and AI act from the same Goal Tree and current state, then return results that update the team's next actions.
Three changes become especially noticeable in daily work.
1. AI can find work someone else has already done.
Similar research, proposals and problems recur within a company. Previously, you had to find someone who remembered the earlier work. Your AI can read context stored in other relevant Goals before acting, removing repeated requests for past examples.
2. A handoff can proceed without a separate meeting.
Passing work used to require explaining the history and document locations. With the shared record, you pass the Goal. The recipient's AI reads its context and continues.
At Addness, the approach reduced a full-day handoff to 30 seconds. The example on note describes the work.
3. Feedback reaches the next AI working on the Goal.
Previously, a correction relied on a person remembering and applying it later. That could require the same feedback repeatedly.
Feedback remains in the Goal. The next AI reads it before acting, so the correction can affect future work. It is the same practice of recording corrections from Chapter 5, now shared across the organization.
There is another consequence: documents and meetings used only to report progress become unnecessary when the current state is available in the Goal Tree. Some work can be removed entirely. That is why the course continues beyond speeding up individual tasks.
Leadership determines whether the approach becomes normal practice
Company-wide adoption requires leadership.
An enthusiastic employee can improve personal work, but changing where work is recorded also changes reporting and meetings. Company leaders must decide those working practices.
People keep the work current in the place leadership actually uses.
Leaders should use the system themselves first. Then put the company's work into the Goal Tree and assign responsibilities. If leaders stay elsewhere while requiring employees to enter updates, the system becomes an additional reporting task.
The structure works across departments
The training operations example applies to other kinds of work too.
- Sales: meeting results update the current state of the sales Goal.
- Hiring: interview results update the current state of the hiring Goal.
- Marketing: customer responses update the current state of the content Goal.
Execute the work where it belongs; record the result in the Goal. The Goal Loop from Chapter 2 becomes a shared way of organizing the company's work.
Read an example
We publish records of the work as well as the underlying approach.
- A full-day task became a 30-second handoff — One AI handed work to another using the context stored in a Goal.
Further examples are added to the Addness account on note.
Review
- Chat alone can leave different accounts of the same work's current state.
- Use one Goal Tree, the appropriate people and AI for each Goal, and results returned to the parent Goal.
- A returned result changes the current state, the remaining difference and the next action for another person or AI.
- The shared system holds the Goal Tree and current states. Execute in external tools and return the results.
- The Goal Loop gains more participants. A child Goal's Step 4 changes the parent Goal's Step 1.
- Information sharing, handoffs and feedback can proceed without repeated meetings or explanations.
- Addness keeps the Goals and current state of collaborative work together.
Set up the working environment
The core course is complete. Next, put it into practice: prepare a personal workspace and extend it to shared work with people and AI. One setup guide and one operations lesson cover that sequence. For recurring automation, return to Chapter 5 and Supplementary lesson 3.
- AI workspace setup guide — Install an AI app and connect it to Addness, with setup instructions in your language.
- Operations lesson 1: Make it a daily working practice — Use the environment every day and extend it to the team.
Learn to achieve results with people and AI.