Setting up a coding agent
Installing and configuring OpenCode, Qwen Code, Cursor, Codex, or Claude Code on the team's own repositories, whichever fits the team's work and your organisation's rules.
AI and coding-agent training for company teams
Led by the people who write ZharfAI's tutorials, your team sets up a coding agent or AI tool on its own real work and learns to judge what it produces.
How this differs from a general course
General courses teach a tool on a demo project. Here your team works on its own repositories, documents, and processes, and what is built in the sessions stays in place afterwards: a configured agent, a project rules file, permission rules, and a usage policy the team wrote itself. Each team's programme is planned separately after a discovery call.
What we teach
Rules files and permissions
Writing a project rules file such as AGENTS.md or CLAUDE.md and setting permission rules so the agent cannot read secrets or run risky commands without approval.
Internal systems through MCP
Connecting the agent to a database, issue tracker, or internal documentation with the least access needed, and understanding the risk of servers that read outside content.
Reviewing agent output
Writing clear requests, planning before large changes, reviewing diffs in git, and the rule that no code is merged without human review.
AI for non-developer teams
Working with Persian documents, summarising, drafting, and using AI in Excel, for finance, operations, and content teams.
n8n automations and output checks
Building simple n8n workflows and learning where model output must be checked against a source and where it needs a person's sign-off.
How we work
Each team gets its own programme. We usually start with one team and decide on the next ones after seeing the result.
- 01
Discovery call
With the team lead we review what the team does, which tools and languages it uses, which tools and models the organisation is permitted to use, and where data must stay.
- 02
Tool and model choice
Tools and models are chosen within those constraints. Where needed, an open-source agent such as OpenCode or Qwen Code is connected to a model on the organisation's own server.
- 03
Training on real work
Each person sets the tool up on the team's own repository or documents and completes a few real, recurring team tasks with it.
- 04
Shared team setup
Rules files, permission rules, MCP connections, and workflows are committed to the team's shared repository so everyone works from one configuration.
- 05
Written usage policy
Together with the team we write down which data must never reach a model, which tasks require review, the spending cap, and who keeps the rules up to date.
Principles we hold to
- Within the provider's terms
- We train only on tools and models your organisation may use under their providers' terms.
- Secrets and data protection
- We work on real repositories and documents, but secrets, keys, and personal data are put out of the agent's reach before we start.
- Human review
- An agent's report is not proof that the work is correct. The team learns to check output with tests and diff review, especially in finance, security, and personal-data code.
- Ownership stays with you
- Configuration, rules files, workflows, and the written policy belong to your organisation and keep working without us after the training.
- No promised numbers
- We do not quote time savings before the training. If you want, we measure the time and quality of a few recurring tasks before and after.
Zharf
Try the ZharfAI assistant right now
Ask about Persian assistants, document automation, or how to start a pilot.
Frequently asked questions about AI team training
How much does AI training for a team cost?
Cost depends on the number of people, the track, the tools and models to be set up, on-site or online delivery, and how much preparation your repositories or documents need. After the discovery call you receive a written plan and estimate. We usually start with one team so you can see the result before deciding on the rest of the organisation.
Which tools do you teach?
OpenCode, Qwen Code, Cursor, Codex, and Claude Code. The choice depends on the team's work, its programming languages, and your organisation's rules, and we make it together in the discovery call. If code and data must not leave the organisation, we connect an open-source agent such as OpenCode or Qwen Code to a model running on your own server.
Is the training on site or online?
Either: on site at your organisation in Tehran, or online. The format is chosen in the discovery call. If the agent has to connect to a model or system inside your network, on-site setup is usually simpler. In both cases each person works on their own machine and repository, not on a demo.
What does our team need to prepare?
For software teams: a computer on which each person may install software, repository access with a branch for experiments, and the list of tools and models the organisation permits. For other teams: sample documents and everyday Excel files, without personal or confidential data that must not reach a model. We send the exact list after the discovery call.
What does the team have at the end?
The team keeps its own work: an agent or tool configured on your own repository or documents, a rules file and permission rules committed to git, the workflows built in the sessions, and the usage and security policy the team wrote. The training does not issue a certificate.
Start with one team and a few real tasks
In the first conversation we review what your team does, which tools and models the organisation permits, which data must stay inside, and whether the training should be on site or online. If training is not useful to you at this stage, we say so plainly.