Chapter 2 · Advanced Systems — Turn One-Time Wins Into a System

2.1 Workspaces and Automation: Let the Work Continue

There are two kinds of work you can bring to AOS Work.

One is a one-off task — "summarize this document" — raised, finished, done.

The other is ongoing work — a season, a buying round, a product launch — which runs across several stages, each one depending on the conclusions of the last.

Handling the second like the first means a new session at every step and the same background explained again: which season, which competitors, where the data sits, what format to output. By the third repetition the cost is obvious.

What a Workspace Is

A workspace is a long-lived space for one business topic: conversations, files, and outputs in one place, with later sessions able to build on earlier conclusions.

The product has three names across two levels:

NameWhat it isWhere it sits in the sidebar
TaskOne conversational executionThe session you have open
Standalone conversationA task outside any workspaceThe "Conversations" group
WorkspaceA container of tasks for one business topicThe workspace list

In short: a task is the unit of execution, a workspace is the unit of organisation. With no workspace selected, the task goes to a standalone conversation; with one selected, it belongs to that workspace.

Deciding Whether You Need One

The test is not "how big is this". It is three conditions:

ConditionThe questionIf it holds
Context reuseWill the next task refer back to this one's material or conclusionsYes → create a workspace
Output consolidationDo the outputs need to sit in one place for review or handoverYes → create a workspace
Shared assetsWill the same files, tables, or knowledge be called on repeatedlyYes → create a workspace

When none of the three holds, a standalone conversation is lighter: the output is delivered and the work ends.

Two cases do not need one: the goal is still being explored and the boundary keeps moving, or you simply need somewhere to store files — that is Library, not a workspace.

Three Typical Scenarios

Scenario A: Seasonal range selection and the buying meeting (about 6 weeks, 5 stages)

StageWhat happensHandover
1Pull last season's sales and sell-through data; produce a sell-through and slow-mover listSales and inventory tables in Base
2Competitor and trend research; produce a trend briefTrend brief in Library
3Produce the range plan and SKU structureThe "Range Review" table in Base
4Build the buying meeting deckBrand templates and deck in Library
5Write back orders after the meeting and review selection accuracyOrder table in Base

With and without: tasks scattered across conversations require separate searches for material and decisions. Grouping them in one workspace and identifying the Base records and Library files makes all five stages easier to review.

Scenario B: Supplier onboarding and contract review (2–3 weeks once, re-reviewed annually)

StageWhat happensHandover
1Collect qualification documents and contract versionsSupplier folder in Library
2Compare against standard clauses and flag differencesStandard clauses in Library
3Produce the risk list and remediation requirementsSupplier register in Base
4Follow up remediation over several emailsThe inbox read and written through a Connector
5Annual re-review, citing the previous round's conclusionsThe previous workspace

With and without: clause versions, risk judgements, and remediation records scattered across conversations take time to find during re-review. Grouping them in one workspace makes review and consolidation easier.

Scenario C: New product launch content (about 2 weeks, 4 channels)

StageWhat happensHandover
1Extract selling points from reviews and after-sales recordsReview analysis table in Base
2Generate multilingual copyWriting standards in the Skill
3Produce hero images and detail pagesBrand assets in Library
4Adapt size and length to each channel's rulesChannel specification table in Base
5Review conversion and return reasons after launchContent performance table in Base

With and without: the channel adaptation stage keeps referring back to stage 1's selling points and stage 3's assets. Grouping these tasks in one workspace makes the material easier to find and reference in the next task.

When You Do Not Need One

Translate a contract, tidy up meeting notes, capture a few competitor screenshots — a standalone conversation is enough. The test: the output is delivered and the work ends, it will not be cited by a later task, and no one else needs to take it over.

The Boundary Rule: By Business Cycle, Not by Department

A workspace boundary should line up with a deliverable: a season, a buying round, a launch, an onboarding review. Three reasons:

  1. Memory already holds the person. Role and preferences apply across workspaces, so there is no need for one per person.
  2. Shared assets live in Library and Base. Workspace tasks can reference those materials and save files generated during the task.
  3. Hand over through specific outputs. In the new conversation, name the workspace file, Library file path, or Base table and records to continue the work.

The third is the most practical: turning "what we said last time" into "which table we wrote, which file we saved" is what lets someone else take the work over.

Creating and Using One

  1. Start a conversation. Click 【New Chat】 on the left.
  2. Select or create a workspace. Use the workspace area above the input box to pick an existing one or create one, for example "Spring 2027 launch".
  3. Work inside it. Every message and output belongs to that workspace.
  4. Review outputs. Task Outputs holds every deliverable in one place and can trace back to the source message.
  5. Continue the work. In the next conversation, specify the files, records, and earlier conclusions to reference.
Inside the "Spring 2027 launch" workspace: use a competitor analysis Skill for research, a pricing calculator Skill for cost work, and a lookbook generator Skill for the deck.

For creating a workspace and continuing a conversation, see New Tasks and Conversations. For files and source-message navigation, see Task Outputs.

Exercise: Continue from a Report to an Owner Checklist

Complete two conversations in the demo workspace named Product Team. First, select the added Weekly Report Work package and turn the work notes into a report named Weekly Report.md.

Click 【New Chat】 and select the same workspace and Work again. Ask the agent to read that report, identify the owner of each task, and save the list as Owner Checklist.md.

How Workspace and Memory Divide the Job

ComparisonMemoryWorkspace
ContentYour preferences, habits, roleThe files, conversations, and outputs of one topic
ScopeGlobal, across all sessionsLocal to that workspace
Question it answers"Who am I and how do I work""Where has this piece of work got to"

They work together: Memory tells the agent who you are; the workspace tells it the background of this work.

How Workspace and Skill Divide the Job

ComparisonWorkspaceSkill
ProvidesContext (files, history, outputs)Method (how this class of task is done)
ScopeOne business topicOne class of task
Rate of changeEvolves as the work advancesRelatively stable, revised occasionally

One workspace can invoke several Skills. The workspace supplies the background, the Skill supplies the method — together they make the collaboration repeatable.

Workspaces and Automation: Let the Flow Run

A workspace solves context persistence, but if you still start every run by hand, you have removed the explanation and kept the action.

To repeat the work at fixed times, combine the workspace with a scheduled task: create a task in 【Scheduled Tasks】 and select its workspace, the agent runs on its own timetable, and the result waits for you there.

ScenarioPainTriggerAutomated flow
A. Daily industry brief30 minutes browsing, still missing thingsDaily 09:00Search the news → filter for business relevance → build a structured brief (3–5 items with summary and source) → email it
B. Sales monitoringTwo hours every Monday on week-on-week comparisonMondays 09:00Pull the data table → compare with last week → flag growth, decline, anomalies → generate an HTML report → send to management
C. Competitor social monitoringManual checks, easy to missDaily 08:00Collect competitor posts → classify and analyse → generate a weekly report → write to the table and send
D. Contract expiry alertsRenewal windows slip pastDaily 08:00Scan the contract register → find items expiring within 60 days → build an alert list → email the owner

For creating one, changing the schedule, and reviewing runs, see Scheduled tasks.

Summary

Use a standalone conversation for a one-off task and a workspace for ongoing work. Three tests: will it be cited later, do the outputs need consolidating, and will the assets be shared.

One level up, the chain runs itself — with a scheduled task attached, your role moves from initiator to reviewer. The next lesson turns to capability building, starting with how a Skill is structured inside.