AI Agents in Practice – Automation of Working with Code, Text, and Browser

1. Introduction to AI agents

AI agents differ from classic chatbots because they can plan, execute, verify, and improve work with greater autonomy. This section explains their operating model, shows how to define goals, context, constraints, and expected outputs, and covers instruction files such as AGENTS.md and approval of plans before execution.

• what an AI agent is and how it differs from a classic chatbot,
• independent execution of tasks by agents AI,
• agent operation scheme: planning, execution, control and improvement of the result,
• delegating complex tasks and multi-step work,
• creating effective instructions for agents,
• defining the goal, context, constraints and the expected result,
• approval of the action plan before executing the task,
• project instruction files, e.g. AGENTS.md,
• creating permanent rules concerning the project, the company and the client.

2. AI Agents in working with text and files

Text and file operations are handled as structured agent tasks rather than isolated prompts. The section includes creating and editing content, preparing briefs and SEO materials, generating metadata, processing many documents at once, organizing and converting files, and producing summaries or recommendations.

• creation and editing of content with the use of AI agents AI,
• preparation of briefs, articles, descriptions and SEO content,
• generation of meta title, meta description and content structures,
• analysis and modification of many documents simultaneously,
• bulk operations on files,
• organizing, renaming and classification of documents,
• conversion of files between different formats,
• automatic editing and updating of many documents,
• analysis of data located in files,
• generation of reports, summaries and recommendations.

3. AI Agents at work with code

Code-related agent work starts from natural-language requests and extends to broader programming support. The scope includes generating scripts, analyzing existing code, finding defects, proposing fixes, refactoring, producing technical documentation, and checking results through testing.

• code generation based on a description in natural language,
• creating simple scriptós automating daily work,
• analysis of existing code,
• finding errors and proposing fixes,
• expansion and refactoring of code,
• creating technical documentation,
• performing multi-step programming tasks by an agent,
• control of resultós and testing of the generated solution.

4. Task and Process Automation

Repetitive operational work can be moved into automated flows that run with limited manual involvement. This section addresses recurring reports, monitoring of data, files, and processes, automatic handling of new information, multi-step workflow design, and points requiring user approval.

• automatic execution of repetitive tasks,
• creation of cyclical reportós,
• monitoring of data, fileós and selected processóes,
• automatic processing of new information,
• designing workflow with the use of agentós AI,
• combining several stepós into one automatic process,
• defining momentós requiring user acceptance.

5. AI agent integrations with external tools – MCP

MCP connects AI agents with external applications, files, databases, and other data sources needed to perform real operations. Participants examine the role of Model Context Protocol, how tools are exposed to agents, and how permissions and access control should be managed securely.

• introduction to the Model Context Protocol,
• the role of MCP in the work of AI agents,
• connecting agents to external applications and data sources,
• providing the agent with tools to perform specific operations,
• examples of integrations with systems, files, databases and applications,
• security and control of the agent's permissions.

6. Playwright MCP – browser work automation

Playwright MCP enables an AI agent to operate directly in the browser across pages, forms, and online panels. The training covers data entry and extraction, link and form testing, desktop and mobile checks, screenshots, deployment audits, and reporting on completed tests.

• browser control by an AI agent,
• automatic opening of pages and navigation between subpages,
• clicking elementsów and handling forms,
• entering and retrieving data from pages,
• automation of repetitive activities performed in online panels,
• testing forms and linksów,
• verification of website operation andnternet,
• testing the view desktopowy and mobile,
• automatic taking screenshots,
• audit of deployments and control of the correctness of changes,
• creating a report of the tests carried outów.

7. Skills – creating repeatable procedures for agents

Skills provide a way to convert recurring company procedures into instructions that agents can execute consistently. Attention is given to standardizing task delivery, designing reusable routines for departments and processes, and combining several skills into a broader workflow.

• what skills are and when it is worth using them,
• conversion of company procedures into instructions performed by AI,
• building repeatedly used procedures,
• standardization of the way of performing tasks,
• creating skillós for specific departmentsós and processesós,
• combining several skillós into a more elaborate workflow.

8. Subagents – division of complex tasks

Complex work becomes easier to manage when responsibilities are split across specialized agents. This part explains how to delegate research, content, coding, testing, and quality control, then coordinate outputs into one consistent result within a multi-agent process.

• what subagents are,
• delegating part of the task to specialized agents,
• division of work according to competencies,
• an agent responsible for research, content, code, tests and quality control,
• coordination of the results of the work of many agents,
• building multi-agent processes.

9. OpenSpec – specification before task execution

OpenSpec structures work by separating planning from execution before the task begins. The section covers writing a solution specification, defining requirements and acceptance criteria, verifying scope, and using user review to reduce misunderstandings before moving into implementation.

• why it is worth separating planning from execution,
• creating the solution specification before starting work,
• defining requirements and acceptance criteria,
• verification of the scope before implementation,
• accepting and modifying the plan by the user,
• reducing errors and misunderstandings in larger projects,
• moving from the approved specification to execution.

10. Practical workshops – building your own workflow with an AI agent

Hands-on work focuses on turning a selected process into an agent-driven workflow. Participants define instructions and rules, map execution stages, add tools and integrations, then test outcomes, diagnose errors, and refine a solution ready for daily use with text, files, or a browser.

• selection of the process for automation,
• preparation of instructions and rules for the agent,
• planning the subsequent stagés of operation,
• adding tools and integrations,
• automation of work with text, files or a browser,
• testing the agent's operation,
• analysis of errors and improvement of the workflow,
• preparation of a solution possible to be used in everyday work.