AI agents and automation
27 use cases, 9 specialists
Prompt engineer
How to write an AI prompt: an AI assistant for phrasing
Writes an AI prompt from a description of the task, with a role, a response format, explicit constraints and examples.
Testing AI prompts: checking against a set of cases
Creates test cases for the prompt — a typical one, an edge case and a refusal case — runs them and shows where the answers differ from what was expected.
AI gives different answers each time: how to make them predictable
Rewrites the prompt so that the model consistently returns the answer in the specified structure, and keeps versions of the prompt with a change log.
System prompt architect
System prompt for a company's AI assistant
Designs a system prompt for the bot — role, boundaries, tone, refusal rules and sample dialogues — and tests it on trial questions.
Instructions for an AI agent with tools: rules and prohibitions
Describes when and with which parameters the agent should call tools and in what form it should return the result, and sets prohibitions on risky actions.
How to reduce AI hallucinations: analyzing errors and fixing the prompt
Collects examples of failed answers, sorts them by cause and edits the prompt, re-testing it on the same examples.
AI agent coordinator
An AI agent runs development from spec to finished code
Breaks the technical specification into stages — planning, architecture, development, verification — and takes the work through them in order, giving each participant the context they need.
Supervising an AI agent in development: checking every task
Runs a check after each task, sends the task back for rework with specific comments if there is an error, and hands the issue over to a person after three failed attempts.
Large project with an AI agent: plan, statuses and resuming after a pause
Keeps a file with the project's task list, the status of each task, the number of attempts and the decisions made, so that the work can be stopped and resumed without losing progress.
Multi-agent systems architect
Designing a multi-agent system with AI
Chooses how the agents are connected — a chain, parallel work or a hierarchy — and describes the input, output and area of responsibility for each agent.
Failures in an AI agent chain: fallback paths and recovery
Lists where the agent chain can break and, for each point, defines a fallback path, a retry limit, an end-to-end log and a handoff to a person.
Where an AI agent needs human approval: checkpoints
Determines which agent actions require human approval, which signs call for escalation and how to avoid swamping the person with unnecessary requests.
MCP server developer
Building an MCP server with AI
Writes an MCP server in TypeScript or Python with clear tool names and descriptions and validation of input parameters.
Connecting 1C or a CRM to an AI agent via MCP
Wraps the API of 1C, a CRM or your own database into tools for the agent, takes keys from environment variables and returns clear error messages.
Testing an MCP server on a live AI agent and publishing it
Runs the full cycle — the agent reads the description, chooses a tool, gets the result — checks the responses when an external service fails and prepares the package for publication.
Process architect
AI for describing business processes
Lays out the process step by step with all branches, failures, timeouts and recovery, and delivers a finished description for development.
An AI agent finds undocumented processes in the code
Reads the project's routes, background jobs, migrations and configuration files and makes a list of processes that have no description.
Business process register with AI
Maintains a register of processes linked to code components and user scenarios, with the status of each description: approved, draft or missing.
Automation governance architect
Is a process worth automating: AI assesses the benefits and risks
Assesses the time savings, data criticality, dependence on external services and load growth, and delivers a verdict: implement, pilot, automate partially or reject.
Reviewing n8n workflows with AI
Checks an n8n workflow against a standard — input validation, an error branch, logging, manual recovery — and lists what is missing.
Rules for automations in a company: an AI assistant
Draws up rules for naming, versioning, ownership and documentation of automations so that they can be handed over to another person.
AI agent trust architect
Identity verification for AI agents: access design
Designs keys for agents and a procedure for issuing, rotating and revoking credentials so that agents can verify each other without human involvement.
Tamper-evident action log for an AI agent
Designs an append-only log in which each entry is linked to the previous one and any retroactive edit is detected by verification.
Delegating authority between AI agents: chains of trust
Describes how one agent passes limited rights for a specific action to another, how to verify this without contacting the issuer and how to revoke them.
Identity graph operator
Duplicate counterparties in 1C and CRM: AI finds the matches
Matches records by email, phone, INN (taxpayer ID) and name, allowing for abbreviations and typos, and shows which ones belong to the same company or person, with a confidence score.
Merging duplicates with justification and rollback: an AI operator
Prepares proposals to merge or split records with a field-by-field comparison, keeps a history of decisions and makes it possible to roll back a mistaken merge.
Single customer record from different systems with AI
Maintains a shared directory that returns one and the same customer or company identifier for a record from any system or any agent.