Colombani.ai's Claude training for companies teaches your teams to work with Claude Team or Enterprise, Cowork, now built into Claude, and Claude Code on their own files, and settles with you the rollout decisions: the plan, the data, the permissions, the connectors.
Claude Certified Architect Professional
Certification issued by Anthropic to Ulysse Trin, the trainer, on 19 September 2026: the most advanced exam of the Claude Certified Architect certification, publicly verifiable.
Verify the certification ↗Qualiopi-certified provider
Certificate QUA009472 for training actions: your OPCO or the FIF-PL can be asked to fund the course.
See the certificate ↗Satisfaction: 4 out of 4
Average score of 4 on a 4-point scale, across all courses, as at 23 September 2026: six sessions, fourteen participants, 96% of objectives achieved, 100% recommendation, no dropouts.
See how it is measured →On-site, anywhere in France
At your premises, on your accounts and documents, or remotely, for one to ten participants per group.
Training across France →Colombani.ai's Claude training for companies, from a Qualiopi-certified provider (certificate QUA009472), comprises three programmes: a half-day briefing for leadership, one day for business teams on Claude, Cowork and Claude in Chrome, and two days of "Claude Code in production" for technical teams. Each session runs on the organisation's Team or Enterprise accounts, using its own files, and settles the rollout decisions: plan, data, permissions, connectors. Each participant leaves with their own skill, a written procedure Claude applies on request, built for a task that recurs every week. The trainer, Ulysse Trin, holds the Claude Certified Architect Professional certification, issued by Anthropic. From €1,000 excl. VAT per half-day and €1,800 per day for an in-company group; final price depends on group size and length, on quote. OPCO funding is possible.
Rolling out across a large organisation
For several hundred users, we roll Claude out in waves, on quote: scoping and usage rules, a pilot with two or three teams, then a department-by-department rollout from their own cases, internal leads trained to write and review their team's skills, and usage measured after each wave in the admin console analytics.
Each programme is part of Colombani.ai's Qualiopi catalogue and runs on your organisation's Team or Enterprise accounts, using your teams' documents, repositories and processes. After the training, a thirty-day activation lets you apply the method to your real files and measure the result.
What Claude changes in each of your departments, demonstrated live on your cases; choosing between Team and Enterprise, the data question, the AI Act and GDPR framework.
What you receive
Your teams hand whole tasks to Claude without writing code: analysing a complete folder with Cowork, browser research with Claude in Chrome, connectors to their tools, skills for recurring tasks.
What you receive
Your team sets up Claude Code on its own repositories: project context, quality and security hooks, MCP servers, skills, subagents, rules shared by the whole team.
What you receive
Claude Certified Architect exam preparation
For teams aiming at the Foundations or Professional exam: assessment on each tested domain, corrected exercises, individual revision plan. Price on quote, after the assessment.
See the preparation →The follow-on to Claude Code in production, to design your own agents and MCP servers and connect them to your systems.
See the programme →Anthropic describes Cowork adoption as a five-level ladder in its public guide "Deploying Claude across your organization", dated 29 April 2026 and linked from Claude Academy. According to the guide, a rollout should move every user up one level, and nobody needs to reach the top in the first month. Our courses use this ladder to set where each participant ends up.
| Level | What the person does | Example | In our courses |
|---|---|---|---|
| Level 0 | Asks Claude questions, as in a chat. | Summarise an email, rephrase a paragraph. | Where most participants start, including those already copy-pasting into another assistant. |
| Level 1 | Hands over a real folder and gets a finished deliverable back. | From a client folder, a summary memo or a table of deadlines. | Reached during the morning, on the participant's own documents. |
| Level 2 | Turns a task that worked into a skill that runs the same way every time. | Reviewing a contract against the in-house negotiation playbook. | Where the day ends: the skill each participant takes away. |
| Level 3 | Combines several skills and schedules them. | A legal or competitive watch delivered every Monday morning. | Started in the session with a first scheduled task, then consolidated during the thirty-day activation. |
| Level 4 | Receives from the administrator a set of skills and connectors built for their department. | The legal department's plugin, installed on every workstation. | A matter for the rollout: internal leads write and review the skills, the administrator distributes them. |
The same guide notes that the progression also applies to Claude Code for technical teams: the Claude Code in production course reaches level 2 on the team's repository, with a shared CLAUDE.md, hooks and skills.
Anthropic's guide names four families of tasks for a first pilot. We add one condition: the measure is taken before the training, otherwise no gain can be demonstrated.
Volume
A task that recurs dozens of times a week along a known pattern: meeting preparation, data extraction, standard replies.
Dense synthesis
Wherever a person acts as the link between several systems: portfolio reviews, regulatory monitoring, quarterly reporting.
Work that creates a queue
A review that holds up other people's work, such as legal sign-off on a campaign: speeding it up unblocks the whole chain.
Procedural expertise
What only your best people do well, because they follow a method nobody wrote down. Once that method is written into a skill, the whole team benefits: according to Anthropic's guide, this family pays back the most.
When the department does not know where to start, the opening exercise is to give Claude its queue of recent requests and ask where the time goes. Each use case by role below states the measure to take.
A training day and a measured result are not the same kind of evidence. The choice is made before starting, because a baseline cannot be taken after the fact.
Adoption signals
By the end of the day, each participant has delegated a real task, written a skill and knows whom to ask when stuck. A single day cannot, however, demonstrate a gain in time or quality: it lacks a point of comparison.
What is measured: satisfaction, objectives achieved and usage reported thirty days later.
A measured result
The baseline is taken before the training, on the chosen task: turnaround, volume handled, number of rework rounds. Participants then apply the method to their real files, with drop-in sessions, a check-in with internal leads on day seven and a leadership review on day thirty that ends with a decision.
What is measured: the change from the baseline, on a task and a deadline set before starting.
See the thirty-day activation, from €3,000 excl. VAT →List prices read on Anthropic's pricing page on 28 October 2026 from France: prices in US dollars, per user per month, excluding applicable taxes. Subscriptions are paid by your organisation.
| Plan | List price | For whom | What it allows | What it does not allow |
|---|---|---|---|---|
| Team, Standard seat | $20 billed annually, $25 monthly | Teams of 2 to 150 people | Sign-in with company credentials (SSO) after domain verification, project sharing, organisation-wide skills, connector controls, usage analytics, Claude Code | No automatic directory sync, no log of admin and member actions, no custom roles, no adjustable retention, no automatic capture of accounts created with the company address |
| Team, Premium seat | $100 billed annually, $125 monthly | Heavy users, often technical | Five times the usage of a Standard seat, in the same team | Same limits as the Standard seat |
| Enterprise | $20 per seat billed annually, plus usage at API rates; 20 seats minimum | Organisations with governance requirements | Directory sync (SCIM), custom roles, audit logs, compliance API, adjustable retention, IP allowlisting, domain capture, spend limits by group | No European hosting when bought directly from Anthropic; total cost depends on usage |
For individual use, the Pro plan ($20 a month, or $17 billed annually) and Max (from $100 a month) offer no central administration, company sign-in or project sharing: they suit personal trials. The guide to Claude for business covers each plan in detail and costs out two scenarios, at 20 and 50 seats.
Your conversations are not used for training, but they are kept indefinitely and, when bought directly from Anthropic, processed outside Europe: the question is which data your teams are allowed to give Claude.
On Team and Enterprise, Anthropic does not train its models on your conversations or coding sessions, unless you opt into its Development Partner Program (source). Without any setting, conversations are kept indefinitely; only Enterprise lets you set a period, thirty days minimum, which covers neither local Cowork sessions nor features that run Claude Code in the cloud (source).
When bought directly from Anthropic, no setting guarantees hosting in Europe: the only geographic restriction in the apps limits inference to the United States, on usage-based Enterprise (source). Through Amazon Bedrock or Google Cloud, however, Claude models have European endpoints that keep processing within the European Union, at a 10% premium (Bedrock, Google Cloud). For customers in the European Economic Area, the contracting entity is Anthropic Ireland Limited (source), and the data processing addendum is part of the contract (source).
For documents covered by professional secrecy or files entrusted by your clients, there are three routes: anonymising before upload, Enterprise with a set retention period, or Claude Desktop connected to your organisation's own cloud provider, for instance on those European endpoints, in which case prompts go only to that provider and conversations stay on the device (source). When no data may leave, the Local AI training covers models installed on your own servers; for lawyers, notaries and accountants, the AI for regulated professions training deals with professional secrecy.
What we settle with you: the list of permitted, to-be-anonymised and prohibited data, and the rule handed to every user.
Who signs in, with which account, and who may change every setting: decide this before opening the first account.
Single sign-on (SSO) is available on Team and Enterprise after domain verification; once SSO is required, accounts that were not added become inaccessible (source). SCIM, which syncs groups with your directory, is Enterprise only (source).
There are four base roles: Primary Owner, of which there is only one, Owner, Admin and User, and Enterprise adds custom roles (source). Since 23 October 2026, Enterprise also sets which models each role can use and spend limits per organisation, group and member (source).
What is agreed before accounts open: who administers, who writes skills and who approves them.
Several powerful features are on by default: letting Claude act in the browser or write in the mailbox should follow a decision rather than a default.
Cowork is on by default; the Owner can turn it off in Organization settings > Cowork. Cloud execution is on by default on Team and off on Enterprise (source). Claude in Chrome is on by default on Team; the Owner keeps lists of allowed and blocked sites, which also apply to Cowork's browser, and zero data retention is not supported (source).
On Team, only Owners add custom connectors, meaning remote MCP servers (source). For Microsoft 365, an Entra Global Administrator grants a one-time consent, then each write tool, including sending Teams messages, is enabled separately; access cannot be limited to a list of SharePoint sites (source). For Google Docs, Sheets and Slides, in beta since 6 October 2026, an Owner enables the connectors first (source).
What we decide together: what opens on day one and what waits until the pilot ends.
A skill turns an in-house procedure into an instruction the whole team can reuse: decide who writes it, who reviews it and who publishes it.
On Team and Enterprise, only an Owner adds or removes an organisation-wide skill, in Organization settings > Plugins & skills; the skill is then on for everyone, and each member can turn it off for themselves. Targeting by group is Enterprise only (source), and Enterprise can also have third-party skills and plugins scanned when imported, in beta since 6 August 2026 (source). Project sharing is on by default (source).
Since 22 October 2026, the skills format has been entrusted to a Linux Foundation body: a skill written for Claude does not lock you into one vendor (source). The Claude Skills guide covers how to write and secure them.
What we prepare: your teams' first two or three skills and the review rule before publication.
An agent that changes code and runs commands needs shared rules, enforced on every workstation rather than left to each developer.
A Team or Enterprise Owner enforces managed settings from the console, in Organization settings > Claude Code, or through a managed-settings.json file deployed to workstations; no local setting overrides them (source). Permission rules are evaluated in the order deny, ask, allow, and auto mode can be disabled by policy (source). Reserved keys allow only the hooks, MCP servers and rules defined by the organisation.
Command isolation, for files and network, is off by default and can be made mandatory, except on native Windows, where it does not exist (source). The documentation describes these settings as a client-side control rather than a security boundary: they complement repository permissions and do not replace them. See also Claude Code hooks.
What you receive: the policy file, the hooks and the team CLAUDE.md, delivered during the Claude Code in production course.
What must never happen is not merely written into an instruction, which the model can lose sight of: it is set in the tool, where the action is blocked or requires human approval.
In Claude Code, a hook is a command Claude Code runs itself at a set point in its lifecycle, for example before each tool call. Anthropic's documentation describes this as deterministic control: certain actions always happen rather than depending on the model's choice (source). A hook placed before a command can refuse it, and permission rules deny, ask for confirmation or allow each tool; the documentation states that they are enforced by Claude Code, not by the model, whereas an instruction in CLAUDE.md only shapes what Claude tries to do (source).
In unified Claude on the Enterprise plan, the equivalent is approval: the administrator can disable the mode that skips all approvals, the one that lets Claude approve its own actions, and standing approval for connector tools, so that Claude asks for permission every time (source). An instruction such as "never send an email" still helps guide Claude; it does not replace a setting that makes sending impossible without human validation.
What we settle with you: the list of actions that require human validation, translated into hooks and permissions for Claude Code and approval settings for Claude, then demonstrated in the session.
What Claude does not yet allow, worth knowing before you sign.
What we recommend: choosing the plan with these limits in mind, rather than discovering them after purchase.
Prepare these settings before the course
One first conversation is enough to agree the plan, the data rule and the list of features to open; the programme and quote follow within two business days.
The Microsoft 365 connector opens mail, calendar and SharePoint only after consent from an Entra Global Administrator (source). Without it, the email workshop stops in the first minute. We have it checked the week before and set aside time for connector setup.
In our sessions on Windows, Cowork only opened once virtualisation was enabled on the device. Cowork can install correctly and still refuse to open: the guide sent three days before asks everyone to open Cowork and give it a first task.
Cloud execution is on by default on Team (source), and we have seen on Windows that a folder's files went to the cloud until local work was set up. For a firm that wants its documents to stay on the device, this is set before the session, together with the data rule.
Files often live in management software that has no connector. There are three routes: a regular export to a folder Cowork reads, Claude in Chrome on the software's web interface if the administrator allows it, or a connector built on the vendor's API, which we assess with you.
The question comes up in every group and is not answered orally. Before the first session, we hand over a written note: what happens to conversations, what the chosen plan lets you set, what your teams are allowed to upload.
For lawyers, the note relies on the guide on professional ethics and artificial intelligence adopted by the French Conseil national des barreaux on 17 March 2026, which recalls in particular the protection of professional secrecy, GDPR compliance and the duty to inform. It tests each requirement against the actual settings of the plan subscribed: model training, retention period, where processing takes place. After a course run on this basis, a business law firm decided to revise its AI usage charter and to inform its clients of how it uses AI. The AI for regulated professions training goes further into these questions.
In existing Enterprise organisations, unified Claude is off by default and will turn on automatically on 16 November 2026, unless an Owner chooses to keep it off or enable it earlier; a choice already made is kept. Access can be opened role by role, hence in waves through SCIM sync, and Anthropic advises two to four weeks of monitoring before expanding. Unified Claude gets its own line in analytics: chat and Cowork active users will appear to drop as members move over, so read the overall active user count rather than adding products together. Anthropic admin guide
Chat and Cowork are now a single Claude, generally available on Enterprise, with model access by role and spend limits per organisation, group and member. Pro and Max received it from 16 September; on Team, the setting is still in beta. Anthropic announcement, release notes
Anthropic hands the Agent Skills standard to the Agentic AI Foundation; according to Anthropic, more than seventy agent platforms have already adopted it. Skills your teams write remain reusable elsewhere. Anthropic announcement
All three are open on every plan and on by default on Enterprise from 15 October. Dashboards and Motion enter beta and stay off until the administrator turns them on. Anthropic article
Team and Max plans include monthly credits to run your own apps and agents: $20 per Standard seat and $100 per Premium seat, capped at $500 a month, no rollover, Claude Code excluded. Claude Haiku 5.5 ships the same day. Release notes, credit terms
Public beta on all paid plans. On Team and Enterprise, an Owner enables the connectors, then the Google administrator deploys the add-on to the domain or selected groups. Anthropic announcement
According to Anthropic, Opus 5.5 costs 40% less than Opus 5 on typical workloads, and five-hour usage limits are raised on Pro, Max, Team and seat-based Enterprise. Sonnet 5.5, faster and cheaper, follows on 28 September. Anthropic announcement, release notes
In beta on Enterprise, smart reports analyse a team's usage: work done, cost, friction points, tasks worth turning into shared skills. Release notes
Enterprise organisations can have third-party skills and plugins scanned automatically for malicious content when they are imported or changed (beta). Release notes
Six tasks worked on in training, each with the Claude feature used, a sample prompt, the measure to take before starting and the final check.
A shared project holding the playbook and the approved fallback clauses.
« Here is the supplier contract received this morning. Compare each clause with our playbook, cite the relevant article where it departs from it, propose the planned fallback clause and flag anything you could not decide. »
Measure before and after: contract review turnaround and the number of contracts waiting.
Check: the lawyer reviews every deviation, and no reference is reused without checking it at the source.
Cowork, on the month-end close folder.
« In the September close folder, reconcile the bank export with the general ledger, list differences above €500 in an Excel table with a column for the likely explanation, and do not create any entries. »
Measure before and after: time from close to variance commentary, and analyst hours per cycle.
Check: the controller approves each explanation before any entry is made.
Claude, with the job description and the collective agreement attached.
« Using the attached job description and collective agreement, write the job ad, then an interview grid with six observable criteria, and flag any wording that could be discriminatory. »
Measure before and after: time from job description to published ad, and the number of review rounds.
Check: HR reviews the ad and the grid, and no candidate data is uploaded without a prior rule.
Claude in Chrome, on the target company's public sites, within the site list set by the administrator.
« Look at this company's website and latest public news, summarise its business in ten lines, suggest three questions for the first meeting and draft an introductory email, without sending it. »
Measure before and after: preparation time per meeting.
Check: the salesperson checks the facts cited and sends the email personally.
One shared project per client, holding the brief, the editorial guidelines and past campaigns.
« Using the attached brief and our editorial guidelines, suggest three campaign angles, each with a title, a hook and the target audience, then flag the information missing from the brief to price the project. »
Measure before and after: time from brief to first presentable proposal, and the number of revision rounds.
Check: the project lead picks the angle and rewrites the hook, and the brief is uploaded only if the agency's data rule allows it.
Claude Code, on the team repository, in plan mode.
« Read the repository and explain the architecture of the billing module, then propose a plan to add tests for the VAT calculation. Wait for my approval before changing any file. »
Measure before and after: time to a first reviewable plan, and the share of the module covered by tests.
Check: the developer reviews the plan and then each change before merging, and the team's hooks run the tests.
These prompts are starting points: in training, each is rewritten on a real document from your teams, then turned into a skill when the task recurs every week. Full method in the guide to Claude Cowork for business.
From €1,000 excl. VAT per half-day and €1,800 per day for an in-company group; final price depends on group size and length, on quote. Claude subscriptions and API usage are paid by the client.
Colombani.ai is a Qualiopi-certified training provider (certificate QUA009472, training provider registration number 11757549975, this registration does not constitute State approval). Claude programmes can therefore be submitted for OPCO funding, or FIF-PL funding for self-employed professionals, subject to the funder's decision. They are not eligible for CPF funding. Colombani.ai prepares the application with you; allow at least two weeks between enrolment and the session.
Ulysse Trin
Founder of Colombani.ai, holder of the Claude Certified Architect Professional certification
Engineer (Epitech, MSc AI), holder of the Claude Certified Architect Professional certification, issued by Anthropic on 19 September 2026, and of the Claude Partner Badge: Claude Code, issued by Anthropic on 16 October 2026. A consultant, trainer and developer, he uses Claude daily on client engagements, and his programmes build on cases developed there.
Claude Certified ArchitectProfessional · Official Anthropic certification. Verify the credentialSee also the nine practitioner certifications issued by OpenAI → From €1,000 excl. VAT per half-day and €1,800 per day for an in-company group; final price depends on group size and length, on quote. Claude subscriptions are paid by the organisation.
Yes, an application can be made: Colombani.ai is Qualiopi-certified (certificate QUA009472). Funding is requested from the company's OPCO, or from the FIF-PL for self-employed professionals, and the decision rests with the funder. The programmes are not eligible for CPF. Colombani.ai prepares the application with you.
Half a day for the executive briefing, one day for business teams, two days for Claude Code in production. The three programmes can be combined within one organisation, on separate dates. Allow at least two weeks between enrolment and the session.
Your choice: on-site anywhere in France, or remote. It is delivered in-house, for one to ten participants, so that the work happens on your accounts and real documents.
On Team and Enterprise, Anthropic does not train its models on your conversations unless you opt in. Only Enterprise lets you set a retention period. When bought directly from Anthropic, no setting guarantees hosting in Europe; European endpoints exist through Amazon Bedrock and Google Cloud. The course sets permitted and prohibited data in the first hour.
A Data Processing Addendum, published on Anthropic's website, forms part of its commercial terms. It includes the standard contractual clauses for transfers outside the European Union and the list of subprocessors. For customers in the European Economic Area, the contracting entity is Anthropic Ireland Limited. The executive briefing reviews this agreement with you against your GDPR records.
Team suits most teams of 2 to 150 people: Claude, Cowork and Claude Code, SSO, shared skills, connector controls. Enterprise, from 20 seats, becomes necessary when the organisation requires SCIM, audit logs, the compliance API, a set retention period, custom roles or spend limits by group.
As of 28 October 2026, the Standard seat costs $20 per user per month billed annually, or $25 monthly; the Premium seat $100 annually or $125 monthly, excluding applicable taxes. Enterprise costs $20 per seat per month billed annually, plus usage at API rates, with a 20-seat minimum. List prices to recheck on Anthropic's page before any purchase.
Both cover conversation, work on files and code: Cowork and Claude Code at Anthropic, ChatGPT's agent and Codex at OpenAI. For an organisation, the differences lie mainly in the plans and administration: seats, retention, connectors, data location. Colombani.ai trains on both tools, and the executive briefing compares them on your cases before any purchase. If your teams already use ChatGPT or Microsoft 365 Copilot, there is no need to switch: the training is also delivered on those tools.
It teaches participants to hand whole tasks to Claude on their own documents, then to turn recurring ones into skills, written procedures Claude applies on request. The Method section of this page places their starting and finishing points on the five-level maturity ladder published by Anthropic. Technical teams follow the same progression in Claude Code.
Yes: the one-day "Working with AI agents" course covers Cowork, Claude in Chrome, connectors and skills, on the organisation's Team or Enterprise accounts. Each participant leaves with a Cowork agent configured for their role and a skill written on a real task. The settings that most often block the session, Microsoft 365 consent and virtualisation on Windows, are checked beforehand.
Anthropic publishes free guides and courses on Claude Academy, designed for self-paced learning, and issues the Claude Certified Architect certification. Colombani.ai is an independent, Qualiopi-certified training provider: its courses run on your teams' accounts and files and can be submitted for OPCO funding. The trainer, Ulysse Trin, holds the Claude Certified Architect Professional certification, issued by Anthropic.
One day is enough for each participant to delegate a real task and write a first skill: it produces adoption signals, measured by the end-of-session evaluation and by usage reported thirty days later. To demonstrate a measured gain, a baseline must be taken before the training and then compared: that is the purpose of the thirty-day activation, from €3,000 excl. VAT.
By deciding in writing, before the first session, which documents may be uploaded, which must be anonymised and which stay out of Claude, then configuring the plan accordingly: retention period on Enterprise, disabling certain features, or a locally installed model when no data may leave. For lawyers, the note tests the profession's ethical requirements against the settings of the plan subscribed.
None for the executive briefing, as the trainer runs the demonstrations. For business teams, each participant works on the organisation's Team or Enterprise account, never a personal one, and receives an installation guide three days before. For Claude Code in production, participants need access to Claude Code and a working repository.
Yes, built into the conversation: since 16 September 2026 on Pro and Max, since 23 October on Enterprise, and in beta on Team. Anthropic's pricing page states that Claude Cowork is now just Claude. The course keeps the name for what Claude does when you hand it a whole folder rather than a question.
Yes, from the Standard seat, and Claude and Claude Code share the same usage quota. The API credits included in Team do not cover Claude Code: heavy users take Premium seats.
Claude Code in production covers several assessed domains: Claude Code configuration, MCP integration, context management and reliability. For the exam itself, dedicated preparation is available; the exam is reserved for people working at an organisation that belongs to the Claude Partner Network, and the access route is reviewed with you before any enrolment. Only Anthropic issues the certification.
Tailored programme and quote within two business days.
Describe your teams, the plan chosen or under consideration and your use cases: programme and quote within two business days, or a call at a time that suits you.
