A good AI training for employees uses the team’s own weekly tasks: prospect research, proposals, weekly reports, content briefs and draft replies to customers. The team learns Claude, Claude Cowork, Claude Code and agents on that work. The course also sets clear data rules, supports the EU AI Act duty on AI literacy and measures three tasks before and after. The team keeps every prompt.
Key takeaways
- Own work: train each person on 3 to 5 tasks they do every week, with the team’s own files and accounts.
- Legal duty: Article 4 of the EU AI Act has applied since 2 February 2025. Its supervision and enforcement rules apply from 2 August 2026.
- No certificate: the European Commission says no certificate is needed. A company can keep an internal record of its trainings.
- Data: use company accounts. Anthropic says it does not train its models on chats from commercial plans such as Team and Enterprise, unless the customer chooses to take part.
- Measure: time 3 tasks before the course and again after it, and count the time spent checking the AI’s work.
- Bottom line: train on real tasks, write the data rules first, and measure 3 tasks before and after the course.
AI training topics for employees by role
The best AI training topics are the tasks each role already does every week. A sales rep trains on prospect research and first messages. A sales lead trains on the pipeline report and proposals. A marketing lead trains on content briefs. Each person leaves with a prompt or a brief for their own task, so the training is in use the next working day.
Generic courses explain what AI is with examples from other companies. They rarely change how a person writes a proposal on Tuesday. The table lists the tasks we train on for each role, and what each person keeps.
| Role | Tasks to train on | What the person keeps |
|---|---|---|
| Sales rep | Research on companies from the target list, and a first message draft for each one | A research prompt and a message brief for each customer segment |
| Sales lead | The weekly pipeline report from a CRM export, and first drafts of proposals | A report prompt and a proposal brief with the company’s own sections |
| Marketing lead | Content briefs from customer questions, and the monthly campaign report | A content brief template and a report prompt |
| Account manager or customer service | Draft replies to customer emails, checked by a person before anyone sends them | Reply prompts with tone rules, and a list of what never goes in a draft |
| Owner or manager | The 3 tasks to give to AI first, and the company’s data rules for AI | The task list, the data rules and the internal record of the training |
EU AI Act AI literacy duty under Article 4
Article 4 of the EU AI Act asks companies that use AI systems to take measures to support the AI literacy of their staff. It has applied since 2 February 2025. The supervision and enforcement rules apply from 2 August 2026, and national authorities are in charge of them. The duty covers companies whose teams use AI tools at work.
The Digital Omnibus on AI changed Article 4 in July 2026. The amended text asks providers and deployers of AI systems to take measures to support the development of AI literacy of their staff. It also says the duty does not require a specific level of AI literacy for any person.
The European Commission’s AI literacy Q&A gives an example. A company whose staff use ChatGPT to write ad copy or translate texts must meet the duty. Those staff should know the specific risks, such as hallucination, where the AI states something false as a fact. The Commission also says there is no need for a certificate, and a company can keep an internal record of its trainings.
The Commission lists the points a company should consider at least. This is our reading for a commercial team, and it is not legal advice.
- A general understanding of AI inside the company: what it does well and where it fails.
- The company’s role: a provider that builds AI systems, or a deployer that uses them. Most commercial teams are deployers.
- The risks of the AI systems the staff use, such as false facts in a proposal or personal data in a prompt.
- Actions that match the staff’s knowledge and experience, and the context where they use AI.
AI training on your team’s own tasks
Train on 3 to 5 tasks that each person repeats every week and that end in a draft, a list or a report. These tasks are a safe start, because a person can check the result in a few minutes. Ask each participant to bring one real task and the files it needs, without personal data that the task does not use.
Consultancies and professional services firms often start with proposals and client reports. Our page on AI for consulting and professional services shows how those tasks fit next to LinkedIn and content work.
We work the same way in our own company. Vytas, our co-founder, runs CreatorFlow, his own SaaS company, with AI every day. Each change there passes a copy check, a language check and a test suite (114 tests passed on 13 November 2026). The rule in a course is the same: AI makes the draft, and a fixed check or a person approves it.
- Prospect research. AI reads a company’s website and public news and writes a short profile. The rep checks the facts.
- Proposals. AI fills the company’s proposal template from call notes. The sales lead edits the scope and the terms.
- Weekly reports. AI reads the CRM or ads export and writes the summary. The manager reads it and decides.
- Content briefs. AI turns customer questions and sales calls into briefs. The marketing lead picks the angle.
- Draft replies to customers. AI drafts each reply from the email and the order status. A person checks and sends it.
AI tools for sales and marketing teams: Claude, Claude Cowork and Claude Code
A commercial team needs three levels of AI tool: a chat for writing and research, an agent that works on the team’s files, and agents that repeat a task the same way every time. With Anthropic’s tools, these are Claude, Claude Cowork and Claude Code. Start with the chat, then add the other two for weekly tasks.
Anthropic describes Cowork as a tool that brings Claude Code’s agentic capabilities to knowledge work beyond coding. On the desktop app, Claude reads and writes files only in the folders you connect. Its own examples include scheduled reports and the analysis of every objection heard in sales calls this quarter. Cowork is on paid plans: Pro, Max, Ultra, Team and Enterprise.
Claude Code is an agentic coding tool that reads files, edits them and runs commands. Skills package a repeatable workflow that a team can share. With MCP, the open standard that connects AI tools to data sources, Claude Code can read files in Google Drive or data from Slack. Vytas spoke at MCP Connect Day in Paris in 2026.
When a task needs to run every week without a person starting it, it becomes a workflow. That is the work of our AI workflows and agents service, in your own accounts and with your approval before anything is sent.
- Claude: writing, research and analysis in a chat. The first tool for every role.
- Claude Cowork: tasks with several steps on the team’s own files, such as a report from three exports.
- Claude Code: agents and skills that do one task the same way every time. Nobody in the team needs to code.
- Agents: a Claude Code agent for one weekly task, such as prospect research for the target list.
Hands-on AI workshop format
A hands-on AI workshop works best in a small group, on the team’s own data, in half a day. Our format is about 4 hours, for up to 10 people from one company, online or in Barcelona. Each person works on their own task at every step, and a trainer helps them at their screen.
Our half-day agenda
Updated November 2026aimo
Every step of the half day ends with work the team keeps.
Each person brings one real task and the files it needs.
- 4 hours About half a day
- 10 people at most
- Claude and Claude Cowork One real task in Claude, then a task with several steps on the team’s own files.
- AI for prospecting Research companies on the target list and draft a first message for each one.
- Agents with Claude Code Build a first agent for one weekly task, described in plain words.
- Tools and first tasks Choose one tool for each task and list the 3 tasks to give to AI first.
The online and in-person sessions have the same agenda and the same length.
We keep the group small so that every person finishes a real task in the session. A course for 50 people becomes a talk, and nobody writes a prompt in a talk.
Half a day is enough for a first result, but not for every tool. So the last step chooses the 3 tasks to give to AI first. Our AI training for commercial teams follows this agenda.
Data rules for using AI at work: company accounts and GDPR
Before the training, the company writes one page of data rules: which accounts to use, what never goes into a prompt, and who checks the output. Use company accounts on a business plan, never personal free accounts. One person in the team owns the rules and answers questions about them.
The plan type changes how your data is used. Anthropic says it does not train its models on chats or coding sessions from its commercial products, such as Team and Enterprise, unless the customer chooses to take part. On the consumer plans (Free, Pro and Max), each user chooses whether their data helps improve Claude. If they allow it, Anthropic keeps the data for five years. If not, it keeps it for 30 days.
The GDPR limits personal data to what is necessary for the purpose. If an AI provider processes personal data for you, the GDPR requires a contract with that provider. In Cowork, Anthropic offers three approval modes: Manual, Auto and Skip. It warns that the automatic check in Auto can miss risky actions. New users should start in Manual.
- Passwords, API keys and access codes. They never go into a prompt or a shared file.
- Customer personal data that the task does not need. Remove names, emails and phone numbers first.
- Data that contracts or confidentiality agreements protect, unless the client allows it.
- Health, financial or other sensitive data about people. Keep it out of AI tools unless your legal team approves it.
- Final facts in a proposal or report. A person checks every number and name before anyone sends it.
Measuring the results of AI training
Measure 3 tasks: time them before the training, then time the same tasks two and four weeks after it. Count the time a person spends checking and fixing the AI’s work, because that time is part of the task. If the total time and the number of fixes do not fall, change the prompt, the tool or the task.
The AI Act does not ask you to test what your staff know. The Commission says Article 4 does not entail an obligation to measure the AI knowledge of employees. The measure is for you: it shows whether the course changed daily work.
- Choose 3 tasks from the list made in the training, one per role if you can.
- Note the time each task takes today, and how often it needs changes.
- Time the same tasks two weeks after the training, with the AI and with the checking time.
- Repeat at four weeks. Keep the tasks that saved time and drop the others.
Prompts, briefs and notes the team keeps
After a good course, the team keeps everything it made: the prompts, the briefs for a first agent or workflow, notes from each step and the list of the 3 tasks to give to AI first. These files make the training repeatable. A new person in the team uses the same prompts on their first day.
Keep the files in one shared folder and name an owner for each prompt. Add the date of the training, the participants and the topics. That folder becomes your internal record for Article 4.
If the first workflow is for an online store, our guide to AI agents for ecommerce lists five first tasks and the read-only settings for each store platform.
Common questions
Is AI training mandatory for companies in the EU?
Companies that use AI systems must take measures to support the AI literacy of their staff. Article 4 of the EU AI Act has applied since 2 February 2025, and its enforcement rules apply from 2 August 2026. Since July 2026, the text says the duty does not require a specific level of AI literacy for any person.
Do employees need an AI certificate?
No. The European Commission says there is no need for a certificate under Article 4. A company can keep an internal record of its trainings and other actions. A simple record works: the date, the participants, the topics and the tools covered.
How long should AI training for employees last?
Our format is half a day, about 4 hours, for up to 10 people. That is enough for each person to finish one real task with AI and to choose the 3 tasks for the following weeks. A longer course makes sense when the team builds several workflows.
Does the team need to know how to code?
No. The team types requests in plain words, in Claude, Claude Cowork and Claude Code. Claude Code reads and edits files for you when you describe the task. Cowork and Claude Code in the desktop app need a paid Claude plan, so agree the plan before the session.
Can a Spanish company pay for AI training with its FUNDAE credit?
FUNDAE says companies that pay the vocational training contribution have a yearly credit, applied as bonuses on Social Security contributions after the training. Courses must last at least 2 teaching hours, and seminars and conferences do not count. The company notifies the start of the training in the FUNDAE application. Ask your advisor whether a specific course qualifies.
Which Claude plan should a company use for AI training?
Use a business plan, such as Team or Enterprise, for work with company data. Anthropic says it does not train its models on chats from these commercial plans unless the customer chooses to take part. Claude Cowork is available on the Pro, Max, Ultra, Team and Enterprise plans.
Sources and checks (9)
- European Commission, AI Act Service Desk: Article 4, AI literacy (as amended, checked November 2026)
- European Commission: AI literacy, questions and answers (updated October 2026)
- FUNDAE: frequently asked questions on training bonuses (in Spanish, checked November 2026)
- Claude Help Center: get started with Cowork (checked November 2026)
- Claude: Claude Cowork product page
- Claude Code docs: overview
- Anthropic Privacy Center: how Anthropic uses personal data in model training (commercial products)
- Anthropic: updates to our consumer terms and privacy policy
- GDPR, Regulation (EU) 2016/679, Articles 5 and 28