Working with CloudIA
This is the second part of the onboarding. The first one, the Complete guide or the Assisted installation, ends when CloudIA answers you. This is where the work begins: first the idea that shapes everything, then your working day and, at the end, the concepts and regulation worth knowing.
What changes
Section titled “What changes”Almost everyone has already used AI personally: you ask it for a summary, a text or a spreadsheet, and the result stays on your computer. It is useful, but it is not a way of working for the company: nobody else sees what was done, the knowledge leaves with the person and the organisation does not know which data has gone through the AI.
CloudIA changes three things:
- Enterprise AI, not personal AI. You work with the licence your organisation gives you and connected to your organisation’s platform, not to a personal account. AI providers’ business licences include contractual commitments on confidentiality and data use that individual licences do not offer in the same way; your organisation decides which one to use.
- Everything goes through a central system. Each person identifies themselves with their user and second factor, and every action of the AI on company information is recorded: who, with which model, when and on what. What is done stays on the platform, not on a computer.
- From doing tasks to supervising them. The company defines positions: roles with their mission, responsibilities and objectives, performed by AI. People do not join to carry out every task by hand, but to direct and supervise the work of those positions, bringing their judgement and making the decisions.
The practical consequence: knowledge no longer depends on who holds each role. If someone changes role or leaves, the position keeps what it has learned and whoever arrives carries on from there.
Your working day with CloudIA
Section titled “Your working day with CloudIA”-
Open your AI client in your working folder, as explained in Start working. Always work from that folder: that is where your organisation’s connectors and instructions are.
-
Verify your identity. Your first message is
2fafollowed by the 6-digit code from your authenticator app. Without that step, CloudIA’s connectors do not answer. See Identity and access. -
Choose which position you will work with using
/impersonarand its name. From then on, the conversation is with that role: it greets you, tells you it is an AI and works with its mandate and memory in front of it. See Your journey with a position. -
Work as you would with a colleague. Ask, review and decide. If the work has several steps, the position opens an Activity with its Checks and documents it as you go.
-
Close or prepare the handover. What was done stays in the Activity and in the position’s memory, so tomorrow you or someone else picks it up from the same point, even from another computer.
One session per position. If you work with several positions, open a session for each one and name it after the position: that way you know who you are talking to in each window. Sessions can work in parallel: while one prepares something that takes time, you move on with another. When you change subject within the same position, start a new conversation: the assistant works better with a clean context (see Context and tokens).
Identity and access
Section titled “Identity and access”The second factor
Section titled “The second factor”CloudIA never asks for your password. It identifies you by your token (installed on your computer) and also by the 6-digit code from your authenticator app, which proves it is you in front of the screen. You type it as 2fa 123456 when you start; if the code expires while you type it, use the next one.
The verification lasts 24 hours from when you type the code, and is not extended by use: in practice, one code when you start your working day. When it expires (or, occasionally, earlier), the assistant will tell you it needs a new code: type 2fa again with the code your app shows at that moment and carry on where you were. One code is valid for all CloudIA connectors.
This double check is what allows everything the AI does to be tied to a specific person, which is the first thing any audit asks for.
Your token, the key that identifies you to the connectors, expires and is renewed: see Your tokens and how to renew them at the end.
What an AI position is
Section titled “What an AI position is”An AI position is a role in the company defined by what the organisation needs, not by the abilities of whoever is available, and performed by AI under human supervision. Each position has a record with:
| Element | What it holds |
|---|---|
| Mission | Why the role exists and what it answers for. |
| Responsibilities | What it does, each one backed by the company documentation it relies on. |
| Exclusions and limits | What it will never do, for example sign documents or make decisions about people. |
| Objectives | How you measure that the role works, with its metric. |
| Supervisors | The people who answer for what the position does and who can direct it. |
| Journal | Its memory: decisions, context and instructions it has received, with date and author. Entries are only added, never rewritten. |
Two traits worth keeping in mind:
- It always says it is an AI. When it starts talking to you, the position states that it is an AI working under human supervision, and it never denies it if asked.
- It has a boundary. What it can do is written in its record. What falls outside it does not do, even if you ask: it tells you who can.
Your journey with a position
Section titled “Your journey with a position”This is what working with a position looks like from start to finish: how you begin, how you find out what is going on, how you work, how you open new work and how you make the position improve. Each step comes with example phrases you can copy and adapt.
1. How do I start
Section titled “1. How do I start”Type /impersonar "Position name", using the position’s name, not its role: on the card at the top of this guide, the name is IsaIAs and the role, Product Direction. If you do not know which one, type just /impersonar: it will show you the available positions, with who supervises each, so you can choose.
The position greets you, tells you it is an AI and loads its mission, objectives, memory and recent reports. From that moment you are talking to the role, not to a generic assistant.
What you can ask of it depends on who you are, and the platform decides it from your identity, not from what you say in the conversation:
| If… | You can |
|---|---|
| You are assigned to the position | Work with it with all its capabilities. You are assigned when you are listed among the people who supervise it and have working permission in CloudIA; your organisation decides. You carry out its Activities and Checks with it, refine and claim the Requirements in its scope, leave instructions in its memory (the Journal), adjust how an objective is measured, document the Processes it coordinates and carry out the acts its record allows. |
| You are not assigned | Only consult: its mission, objectives, status and reports; ask it to explain one or for its opinion within its mandate; and ask it to propose a Requirement signed by it, which stays as a draft until a person reviews it. |
| You administer positions | Administer it, as a person and outside the conversation with it, from Gobernanza IA (AI Governance) in the CloudIA module: create it, change its mission, responsibilities and objectives, assign who supervises it and activate, pause or retire it. This requires the position administration permission your organisation grants. Approving, discarding or distributing its reports also requires being one of its supervisors. |
/impersonar IsaIAsAnd then, for example:
Introduce yourself: what your mission is, what objectives you have and what you cannot do.2. How do I find out what is going on
Section titled “2. How do I find out what is going on”In CloudIA everything starts from documentation and everything stays in the system’s knowledge. Work does not live in your head, your inbox or your notes: it lives in the Activities of the documentation system (CLOUD Documentum), with their goal, plan and Checks. That is why the day always starts the same way: by looking at which Activities you have pending.
Good morning. Which Activities do I have pending today? Sort them by urgency and tell me which ones are overdue.Before asking for anything new, see what is going on. Every Activity always has a person who does it and may also have a position that coordinates it:
- Who does it: the assigned person. It is the name on the Activity, yours or a colleague’s. That person carries it out, usually working with the relevant position.
- Who looks after it: the coordinating position. It is the position in charge of making sure the Activity does not stall: it follows it up and reports on how it is going. Which position coordinates each Activity is not decided by the AI: it is always decided by a person.
That is why you can ask from both sides: about what is assigned to you and about what a position coordinates.
The Activities a position coordinates. Ask it while you impersonate it:
Which Activities do you coordinate and where does each one stand? Tell me which ones are overdue or stalled.The Activities assigned to you. Ask about yourself, with or without a position impersonated:
Which Activities are assigned to me? Tell me which ones are overdue or due this week.And to catch up:
What has happened in your area since Monday? Decisions, reports and anything waiting for me.Which Requirements have you proposed and which are still waiting for approval?You can also see it without a conversation, on the platform, CloudIA module, Mi dashboard (My dashboard):
- Mi Resumen (My summary): your pending and overdue Activities, the ones assigned to you.
- Mis puestos (My positions): the positions you supervise. Each one has its record with tabs: Resumen (summary), Mandato (mandate), Trabajo (work: proposed Requirements and the Activities it coordinates, with their Checks), Informes (reports), Objetivos y OKRs and Journal.
This is what the full journey looks like in a conversation, from opening the session to completing the first Check of an Activity:
2fa 482913/impersonar TecnIA1. Renewal of the SSL certificates of the public domains: 0 of 3 Checks complete; the first one expires on Friday.
2. Equipment for the two October hires: blocked, waiting for the laptops to arrive.
I suggest starting with the certificates, which expire sooner.
• 5 expire in the next 30 days and need renewing.
• 2 renew automatically.
• The intranet one has no owner assigned.
Shall we assign the intranet one to the IT team, like the rest?
3. How do I work on an Activity
Section titled “3. How do I work on an Activity”An Activity is a unit of work with its goal, its plan, what is out of scope and a list of Checks.
What a Check is, and what it is not
Section titled “What a Check is, and what it is not”A Check is not an order for you to carry out, nor a to-do list. It is an acceptance criterion: something that must be true for the Activity to be considered complete, together with the evidence that proves it. Four questions to understand them:
-
When is an Activity finished? When all its Checks are met. Not before, and not “just because”.
-
How do we know? Each Check keeps the evidence: what was done, how it was verified and with what result.
-
What happened? The set of Checks is the Activity’s documentation. Whoever comes later understands it without asking anyone.
-
Are the Checks the position proposes final? No. The list is dynamic: the position proposes a starting point and whoever carries out the Activity decides. As you progress you can add, remove or rephrase Checks, always according to your criteria, and the position adjusts them. For example:
Remove the training Check: this supplier does not need it. And add one: "Supplier bank details verified before the first payment".
That is why a good Check is written as a verifiable result and not as an action:
| Not like this (an order) | Like this (a verifiable result) |
|---|---|
| Talk to the supplier | Financial terms agreed and documented |
| Review the contract | Contract reviewed by legal advisers with no open comments |
| Train the team | Team trained: session held and guide published |
Who does what. You do not fill in the Checks by hand. You work on them with the position:
- The position documents them. As you progress, it writes in each Check what was done, how it was verified and with what result, following your criteria.
- You contribute and validate. You give it the information and decisions it needs and, when it presents a Check to you, you check that the evidence proves what it says. It is only considered complete with your validation.
A Check goes through states such as pending, in progress, blocked (when something external is missing), in review (waiting for your validation) and complete. When all of them are complete, the Activity is closed with a summary of the solution.
To work on an Activity, ask the position to pick it up:
Let's pick up the monthly expense report Activity. What is the next pending Check and what do you need from me?The position reads the Activity, tells you where it stopped and proposes the next step. You provide the information and decide; it does the work and documents each step in its Check. When a Check is met, it marks it as complete with its evidence; you review it.
Here are this month's invoices. Reconcile them with the bank movements, record the result in the Check and tell me what does not match.Handover. Because everything stays in the Activity, the work does not depend on who started it. If you have to leave it half done, the Activity is reassigned to another person, who picks it up from the last Check, helped by the coordinating position if it has one:
Continue the Activity my colleague left half done: summarise what was done and let's carry on with what is pending.4. How do I create an Activity
Section titled “4. How do I create an Activity”When a new piece of work with several steps comes up, do not start it “from memory”: ask the position to open an Activity. Tell it what you want to achieve and why; it proposes the plan and the Checks, and does not create it until you confirm.
Open an Activity to onboard a new supplier: we need to agree the terms, sign the contract, connect their system and train the team. Propose the plan and the Checks, written as verifiable results, before creating it.Add a Check to the Activity: "Contract reviewed by our legal advisers before signing".Who it is assigned to depends on how your organisation has decided to work. In some, anyone can assign Activities to any colleague. In others, you only assign them to yourself, or they go to a person in charge of receiving them and sharing them out among the team. The same applies to which position coordinates it. If you do not know how it works in your organisation, ask the position or your manager before assigning work to others.
Every Activity is linked to a Process
Section titled “Every Activity is linked to a Process”An Activity never stands alone: it always serves a Process or SubProcess of the company, directly or through the Requirement it comes from. That is what lets you know later why it exists and which part of the company it changes. It also belongs to a Project, which groups related work.
That is why, when creating it, the position looks for where it fits:
- If it is clear, because it comes from a Requirement or the work belongs to a known Process, it links it and tells you.
- If there is no clear link, it asks you where you want to document it and suggests the Processes that fit best.
- If that Process does not exist yet, it has to be created and documented first. The position assists you: it proposes the Process, drafts it with you and, once you confirm, someone with permission creates it. Then it links the Activity.
Open an Activity to prepare the annual financial audit. Which Process does it belong to? If none fits, propose how to document it.5. How do I improve the position
Section titled “5. How do I improve the position”A position improves with use, and the fastest improvement is you. It always starts the same way: correct it and explain why. If a draft does not work for you, do not fix it silently: tell it what is wrong and why.
In the contract you proposed I changed the notice period from six to three months, because that is our policy with suppliers.What happens next with that instruction depends on your role.
If you are assigned to the position
Section titled “If you are assigned to the position”You can leave the instruction in its memory, the Journal. What is there is taken into account in every conversation, with anyone, and in its reports:
Write this decision in your Journal: expense reports are closed on the 5th of each month.You do not need to remember the word “Journal”: just ask it to keep it in mind from now on.
From now on, remember that expense reports are closed on the 5th of each month.If you are not assigned
Section titled “If you are not assigned”Your correction serves the work you are doing and is documented in that Activity, but it does not go into the position’s memory: only someone assigned to it can do that. To make an improvement permanent you have two paths:
-
Ask the position to propose a Requirement. It is the formal path: it is signed by the position, linked to the Process it improves, and reviewed by a person (step 6 explains it).
This notice period should always apply. Propose a Requirement to make it part of how you work, linked to the supplier contracting Process. -
Pass it on to someone assigned to the position. The position knows who supervises it and helps you explain it.
Who supervises you? Prepare a summary of this correction for me to send them.
In any case
Section titled “In any case”What changes its design (its mission, responsibilities or objectives) is not changed in the conversation, whether or not you are assigned: it is proposed as a Requirement and a person decides. That is the next step.
6. How do I propose a Requirement
Section titled “6. How do I propose a Requirement”A Requirement is a written change request: what is missing or should work differently, why, and how you will know it is solved. The position can propose it, signed by itself, when it detects a gap or when you ask it to.
Every month we redo the expense reconciliation by hand. Propose a Requirement to automate it: link it to the monthly closing Process, explain the problem, the benefit and how we will check that it works.I want you to also take care of reviewing contracts with partners. Propose a Requirement on the Process you coordinate to extend your responsibilities, with what you would need to do it well.The Requirement starts as a draft. A person reviews it and approves or discards it; if approved, it is resolved through one or more Activities, and you are back at step 4.
Several positions at once
Section titled “Several positions at once”You can work with several positions in separate sessions and in parallel, and even ask one to consult another. It is powerful, but you direct it: two positions talking without control can loop or fail to agree. Ask for something concrete, review the answer and decide.
What no position does
Section titled “What no position does”There are things nobody can ask from within the role, not even someone assigned to it: activate, pause or retire positions, change their supervisors, approve or distribute their reports, create or delete objectives, or change their mandate. A person does that as a person, outside the conversation with the position. It is not a missing permission: it is the guarantee that the AI does not govern itself.
How positions stay in sync
Section titled “How positions stay in sync”There are two things to keep in sync, and they work differently.
Work and memory: always up to date
Section titled “Work and memory: always up to date”Knowledge lives on the platform, not on your computer. Each position’s record, its Journal, its Activities and its reports are on your organisation’s platform. That is why it does not matter which computer you work from, and why two people impersonating the same position see the same thing. You do not have to do anything to keep this in sync.
A good habit at the end of the day, or when you come back after a few days away: ask the position for the news in its area. It will summarise what has changed, what was decided and what is pending, with no need for catch-up meetings.
Your computer’s capabilities: when your organisation updates them
Section titled “Your computer’s capabilities: when your organisation updates them”The skills, agents and commands your assistant uses (including /impersonar) are installed in your working folder and maintained by your organisation from the platform. To keep them up to date:
-
/ai-sync, on its own. It brings the latest version of your repository, compares your capabilities with the platform’s, updates the shared ones with a single confirmation and asks you what to do with your own ones that differ. If your project uses git, it saves the result in the repository. -
/ai-check-drift, if you only want to know what is out of date without changing anything. -
If you want to review the whole CloudIA setup, ask the assistant:
Verify the CloudIA setup of this project.
How positions evolve
Section titled “How positions evolve”A position is not designed once and forgotten: it improves with use, always with a person approving each change.
- It learns from corrections. When someone assigned to it leaves it an instruction, it stays in its memory and it takes it into account next time (see How do I improve the position).
- It proposes improvements as Requirements. When it detects something missing or that should change, it proposes it as a Requirement signed by the position, always linked to the Process it improves. It starts as a draft; a person reviews it and approves or discards it (see How do I propose a Requirement).
- Its record is versioned. Changes to its mission, responsibilities or objectives are decided by a person, recorded and can be looked up.
- From describing to automating. At first the position works with what you tell it and the documentation it has. When a piece of work repeats and is well defined, it is connected to the systems needed so it can read, execute, save or send on its own, always within its boundary.
Documentation strategy: CLOUD Documentum
Section titled “Documentation strategy: CLOUD Documentum”Why document
Section titled “Why document”AI can only work well with what is written down and organised. What lives in someone’s head, in an email or in a file on their computer does not exist for it, nor for the person who arrives tomorrow. That is why in CloudIA text is the new programming language: describing well how the company works is what lets positions carry it out, improve it and automate it.
CLOUD Documentum is where that knowledge lives. It gives you three things:
- A single source of truth, the same for people and for AI. Nobody works with a different version.
- Knowledge that stays in the company: if someone changes role or leaves, what they knew is still there.
- Traceability: who decided what, when and why, and how it was verified that it was done. That is what any audit asks for, including the AI Act.
How it works
Section titled “How it works”Documentum organises information into three questions, within Business Groupings (a department, a product or a functional area):
| Question | Where it is answered |
|---|---|
| What the company does | Processes and SubProcesses, Functionalities and Requirements: the functional knowledge, written for the business. |
| How it does it | Business Operations, also called development units, and their modules: how each process is built. |
| When it is done | Projects and Activities with their Checks: execution and follow-up. |
Where each thing goes
Section titled “Where each thing goes”Documentation always lives in these places, and nowhere else:
| Where | What is documented |
|---|---|
| Process | What the company does in an area, end to end, written so the business understands it. |
| SubProcess | Each part of a Process that stands on its own: its steps, rules and exceptions. |
| Functionality | A specific capability the company offers, which several Processes can share. |
| Requirement | Each change or improvement: what is missing, why, and how you will know it is solved. Always linked to a Process. |
| Activity | The work that carries it out: goal, plan and what is out of scope. Always linked to a Process, directly or through its Requirement. |
| Check | The acceptance criteria of an Activity and the evidence that they are met (see What a Check is). |
Chained together, they tell the full story of any change: a Process describes how work is done; a Requirement on that Process asks to improve it; an Activity carries it out; its Checks prove it is done; and the Process is updated with the result. A decision that stays in a conversation or an email is lost: the rule is that it ends up in one of these places.
Positions in a project
Section titled “Positions in a project”A fictional example. A company of three partners launches a new business. Instead of splitting the work according to what each partner knows how to do, it first designs the positions the company needs: General Management, Finance, Risk, Operations, Customer Service… For each one it defines its mission, its responsibilities and what it will never do.
Then it shares out the supervision: each partner supervises the positions closest to their experience. When a partner needs to close the monthly expense report, they do not open a spreadsheet on their computer: they impersonate the Finance position and ask it for the report, and the position opens an Activity, gathers the invoices, prepares the draft and documents each step. If the partner has to leave, the Activity is reassigned to another team member, who picks it up from where it stopped.
This way the company grows through well-defined positions, not through the sum of what each person carries in their head, and when it brings someone in, that person finds the role already described and with its history.
What you can ask and what to expect
Section titled “What you can ask and what to expect”You can ask, among other things:
- “What do I have pending this week and what is overdue?”
- “Prepare the draft of the monthly report in last month’s format.”
- “Summarise what has changed in this process since Monday.”
- “Open an Activity for this integration with these four steps.”
- “Review this document and tell me what is missing according to our procedure.”
- “Propose a Requirement so that this step is no longer done by hand.”
You can expect speed, consistency and that it will not forget what is written down. Do not expect it to guess what is not documented: if a fact is not on the platform, it will tell you it does not know instead of making it up, and it will ask you for the information.
Regulation in day-to-day work
Section titled “Regulation in day-to-day work”The European Artificial Intelligence Regulation (the AI Act) applies to organisations that use AI, not only to those that build it. Much of what you have read in this guide exists because of it: positions say they are an AI and sign what they produce, a person decides and validates, and everything is recorded under your name. Remember too that risk depends on the use, not on the technology: using AI to screen CVs, for example, is a high-risk use.
Training on the use of AI in your organisation is in the Academy courses, which record who has taken them. This is guidance, not legal advice.
- Speak instead of typing. Dictating with a voice tool is faster and you give more context than when typing.
- Name each session after the position, and do not mix positions in the same conversation.
- Give the why, not just the what: with the reason, the position gets it right more often and remembers it next time.
- Ask for the plan before the work when the task is large: review the steps and then let it carry them out.
- Ask it to write it down. A decision that is not in Documentum is lost with the conversation.
Your tokens and how to renew them
Section titled “Your tokens and how to renew them”The token is your personal key to the connectors. It can have an expiry date:
- If you create it with the installer, you choose 30, 90 or 365 days, or no expiry; the default is 90 days.
- If you create it on the platform, the expiry is optional: with no date, it does not expire.
You manage them on the platform, in the CloudIA module, section Mi dashboard → Conectar CloudIA: there you see their status (active, revoked, expired or scheduled), create new ones, retrieve one to install it on another computer and revoke the ones you no longer use.
A token is not extended: it is replaced. When it expires, the connectors stop answering and the assistant shows an error saying the token has expired and that you should generate a new one. To renew it, run the installer again in your working folder and create a new one, or create it in Conectar CloudIA and install it again. The platform’s Regenerar (regenerate) option revokes the previous token and creates another with a different value, which you will have to install again. The easiest way is to note when yours expires and renew it beforehand.
