Every Monday, someone opens the list of sent quotes, spots the ones that have had no reply and prepares the follow-ups. Useful, repetitive, and always put off until later. Odoo 20 launches AI agents: assistants that you configure by talking to them and that work on their own, at the scheduled time. Here is how they work, where to find them, and what they cost.
1. The problem: the tasks nobody does on time
The SMEs we work with all have their list of chores: following up quotes, sorting received invoices, summarising new leads, answering recurring questions. Until now, automating them in Odoo meant writing an automated action, with a domain, fields and sometimes code. Effective, but reserved for those who know how to configure it.
The AI agent changes the way in: you describe the task in plain language, and it asks for whatever it is missing.
💡 Key takeaway
An agent does not invent new capabilities in Odoo: it makes accessible, in everyday language, what used to require configuration.
2. Creating an agent: it asks its own questions
It all happens in the AI app, Agents menu. On a new database, the screen shows "Create your first Agent!". An agent's form has four tabs: "Instructions", "Skills", "Sources" and "Triggers". Two of them, "Instructions" and "Triggers", are not translated in V20: you will see them in English.
The most interesting novelty is a behaviour: an agent without instructions does not set off blindly. If your request describes a lasting role ("every Monday, follow up quotes with no reply"), it first asks you its questions, with two to four clickable suggested answers: after how many days to follow up, which quotes to target. It then saves its own instructions. And a recurring request ("every day at...", "when a quote is created...") is treated as an automation to build, thanks to the "Build Automation" skill (untranslated).
The "Sources" tab gives it material: documents from the Documents app (PDF, Word, Excel) or articles from Knowledge, if those apps are installed.
💡 Key takeaway
You don't fill in a form: you answer a colleague who wants to understand properly before acting.
3. Running it on its own: triggers
In the "Triggers" tab, the add button offers four cards, all in English in V20:
- "On a Schedule": at a fixed frequency, for example every Monday at 8 am;
- "Based on a Date Field": once per record, when a date is reached (quote expiry, contract end);
- "After Creation": on every new record;
- "After Update": when a specific field is modified.
An important point, found in the code: when the agent runs on its own, its actions are confirmed automatically. In a conversation, Odoo shows you a confirmation card before creating or modifying anything; in a scheduled run, there is nobody to click. In return, every run is kept in a conversation that you can reread.
💡 Key takeaway
For a first automation, prefer an agent that prepares (a "Follow up" activity) rather than an agent that sends.
⚔️ The Match: manual follow-up vs an Odoo 20 agent
| Criterion | By hand | Odoo 20 AI agent |
|---|---|---|
| Consistency | Depends on everyone's diary | At the scheduled time, every time |
| Setup | None | A conversation, then a trigger |
| Traceability | Rarely noted | Every run kept |
| Direct cost | Someone's time | Odoo AI credits on every run |
| Human judgement | Intact | To be kept at the end of the chain |
Behaviour observed in the Odoo 20 Enterprise source code (ai_agentic module), October 2026. We do not publish a figure for time saved: it depends on your volume.
4. What it costs, and who it is for
Let's be clear: agents go through the Odoo AI service and consume IAP credits. In Odoo 20, you can no longer plug in your own OpenAI key. An agent scheduled every Monday therefore consumes credits on every run. Before multiplying them, measure the consumption of a first agent over a few weeks.
The other limits to know about:
- it is an Odoo 20 Enterprise feature;
- the interface is partly in English (the triggers, some tabs);
- an agent is still a language model: vague instructions give vague results. Test it on a copy of your database before letting it run on your real customers.
It is made for SMEs that have identified one well-defined repetitive task, with a simple rule: quote follow-ups, sorting incoming requests, a weekly summary of leads. Less so for decisions that require tact or detailed knowledge of the customer.
💡 Key takeaway
A first agent, one task, one measurement of consumption: that is the right scale to start with.
The Verdict: an AI colleague is only as good as the task you entrust to it
Odoo 20 agents are simple to create. Choosing the right task, writing clear instructions and keeping a human at the end of the
chain is where the result is decided. An
audit of your Odoo processes
helps spot the chores that suit this, and an
Odoo training
makes your team self-sufficient for the next ones.
Which task would you entrust to an agent first? We will build the first one with you.
Every Monday, someone opens the list of sent quotes, spots the ones that have had no reply and prepares the follow-ups. Useful, repetitive, and always put off until later. Odoo 20 launches AI agents: assistants that you configure by talking to them and that work on their own, at the scheduled time. Here is how they work, where to find them, and what they cost.
1. The problem: the tasks nobody does on time
The SMEs we work with all have their list of chores: following up quotes, sorting received invoices, summarising new leads, answering recurring questions. Until now, automating them in Odoo meant writing an automated action, with a domain, fields and sometimes code. Effective, but reserved for those who know how to configure it.
The AI agent changes the way in: you describe the task in plain language, and it asks for whatever it is missing.
💡 Key takeaway
An agent does not invent new capabilities in Odoo: it makes accessible, in everyday language, what used to require configuration.
2. Creating an agent: it asks its own questions
It all happens in the AI app, Agents menu. On a new database, the screen shows "Create your first Agent!". An agent's form has four tabs: "Instructions", "Skills", "Sources" and "Triggers". Two of them, "Instructions" and "Triggers", are not translated in V20: you will see them in English.
The most interesting novelty is a behaviour: an agent without instructions does not set off blindly. If your request describes a lasting role ("every Monday, follow up quotes with no reply"), it first asks you its questions, with two to four clickable suggested answers: after how many days to follow up, which quotes to target. It then saves its own instructions. And a recurring request ("every day at...", "when a quote is created...") is treated as an automation to build, thanks to the "Build Automation" skill (untranslated).
The "Sources" tab gives it material: documents from the Documents app (PDF, Word, Excel) or articles from Knowledge, if those apps are installed.
💡 Key takeaway
You don't fill in a form: you answer a colleague who wants to understand properly before acting.
3. Running it on its own: triggers
In the "Triggers" tab, the add button offers four cards, all in English in V20:
- "On a Schedule": at a fixed frequency, for example every Monday at 8 am;
- "Based on a Date Field": once per record, when a date is reached (quote expiry, contract end);
- "After Creation": on every new record;
- "After Update": when a specific field is modified.
An important point, found in the code: when the agent runs on its own, its actions are confirmed automatically. In a conversation, Odoo shows you a confirmation card before creating or modifying anything; in a scheduled run, there is nobody to click. In return, every run is kept in a conversation that you can reread.
💡 Key takeaway
For a first automation, prefer an agent that prepares (a "Follow up" activity) rather than an agent that sends.
⚔️ The Match: manual follow-up vs an Odoo 20 agent
| Criterion | By hand | Odoo 20 AI agent |
|---|---|---|
| Consistency | Depends on everyone's diary | At the scheduled time, every time |
| Setup | None | A conversation, then a trigger |
| Traceability | Rarely noted | Every run kept |
| Direct cost | Someone's time | Odoo AI credits on every run |
| Human judgement | Intact | To be kept at the end of the chain |
Behaviour observed in the Odoo 20 Enterprise source code (ai_agentic module), October 2026. We do not publish a figure for time saved: it depends on your volume.
4. What it costs, and who it is for
Let's be clear: agents go through the Odoo AI service and consume IAP credits. In Odoo 20, you can no longer plug in your own OpenAI key. An agent scheduled every Monday therefore consumes credits on every run. Before multiplying them, measure the consumption of a first agent over a few weeks.
The other limits to know about:
- it is an Odoo 20 Enterprise feature;
- the interface is partly in English (the triggers, some tabs);
- an agent is still a language model: vague instructions give vague results. Test it on a copy of your database before letting it run on your real customers.
It is made for SMEs that have identified one well-defined repetitive task, with a simple rule: quote follow-ups, sorting incoming requests, a weekly summary of leads. Less so for decisions that require tact or detailed knowledge of the customer.
💡 Key takeaway
A first agent, one task, one measurement of consumption: that is the right scale to start with.
The Verdict: an AI colleague is only as good as the task you entrust to it
Odoo 20 agents are simple to create. Choosing the right task, writing clear instructions and keeping a human at the end of the
chain is where the result is decided. An
audit of your Odoo processes
helps spot the chores that suit this, and an
Odoo training
makes your team self-sufficient for the next ones.
Which task would you entrust to an agent first? We will build the first one with you.