Development of AI agents and AI solutions for business
RESTART develops AI agents and custom AI solutions for large companies: agents for contracts and procurement, tenders and sales, finance, HR, support services and development. The agent is integrated into the customer’s processes and systems - 1C, SAP, EDMS, ITSM - and acts within the limits of the issued authority: with human confirmation, a log and rollback of operations.
AI agents for business
RESTART Group — development and implementation of AI agents for business processes of large companies: from scenario selection to production operations within the customer’s environment.
An AI agent differs from an assistant in that it not only responds, but also acts: it creates requests, changes records, prepares documents, and calls external services. Therefore, when developing an agent, it is important not only the quality of the response, but also the scope of authority: which actions are allowed, which require confirmation, and how to cancel the performed operation.
What AI agents are we developing?
Contracts and procurement
Extraction of essential conditions, comparison of editions, risk checklist according to internal rules, integration with EDMS, purchasing systems, ERP, Diadoc and Kontur. AI agent for contracts and procurement
Tenders and sales
Monitoring of purchases by interest profiles, relevance scoring, analysis of technical specifications and compliance matrix, draft commercial proposal and selection of cases. AI agent for tenders and sales
Finance
Explanation of plan-actual deviations, analysis of cash-flow and accounts receivable, draft explanations for management based on data from 1C:ERP, 1C:UH, SAP, BI and DWH. AI agent for finance
HR
Competency base, resume analysis and interview preparation, answers to newcomers on documents and processes, training recommendations - with work with personal data in accordance with Federal Law No. 152-FZ. AI agent for HR
Help Desk
Classification of requests, operator tips, knowledge base and SLA control on top of Jira, Naumen, ServiceNow, 1C ITIL and internal ticket systems. AI agent for service desk
Development
Search and explanation of code, tests, documentation and working with legacy systems in a closed loop: GitLab, Jira, Confluence. AI assistant for development
If there is no ready-made module for the process, the agent is designed within the framework custom development on a common core Restart AI Enterprise Platform.
How is the development of an AI agent going?
Scenario selection
We record the business task, the owner of the result, users, sources, information security limitations and success criteria. We cut off scenarios where AI will bring more risk than benefit. Short start - AI-discovery within 10 working days.
Data and access
We define data classes, source systems, roles, access matrix, sensitive fields and log requirements.
Agent Architecture and Authority
We describe RAGs, models, agent actions, integrations and APIs, threat model and human confirmation points.
MVP and pilot
We launch the working module in a limited loop on real data and users, set up rules and integrations.
Security check
We test prompt injection, leaks, agent rights, logs and error behavior.
production launch
We transmit roadmap, HLD/LLD, regulations and support requirements; we expand the agent to new units and scenarios without reworking the kernel.
How the actions of an AI agent are controlled
| Mechanism | How it works | What prevents |
|---|---|---|
| Limitation of authority | A separate agent service account with the minimum necessary rights, a white list of operations and systems, limits on the volume and frequency of actions | Actions outside the agreed scenario |
| Human verification | Operations are divided into those performed immediately and requiring confirmation; it is recorded who confirmed and on what basis | Irreversible operations — payments, changes to contractual data, customer communications — without involvement of the responsible person |
| Rollback of operations | For each operation, a cancellation method is defined: a reverse transaction, a version of the record before the change, or a manual procedure | A situation where the error has been noticed, but there is no way to return the system to its previous state |
A separate log is kept: what action was performed, on what request, with what parameters, by whom it was confirmed and what the result was. More details - safe automation of processes by AI agents.
environment and platform
Agents are deployed as modules Restart AI Enterprise Platform - with common roles, journals and integrations. Hosting – on-prem, private cloud or hybrid; production data is not transferred to external AI services without the written permission of the customer. environment architecture - on page corporate AI environment, computing power - AI Compute.
Systems around agents — portals, integrations, APIs — are developed within this area of focus. corporate software development.
Experience
In a banking project under NDA, RESTART implemented an AI platform and RAG agents for a bank from the top 5 of Uzbekistan: the solution has been piloted and is being maintained and developed. Case: AI agents for a bank. The entire implementation path is on the page implementation of AI in business.
Frequently asked questions
What is an AI agent for business?
An AI service that not only answers questions, but also performs actions in corporate systems: creates applications, changes records, prepares documents and drafts - according to an agreed scenario and within the limits of the issued authority.
How is an AI agent different from a chatbot?
The chatbot responds, the agent acts: creates requests, changes records, calls external services. Therefore, the agent has a separate account with limited rights, transactions that require human confirmation, a method for canceling actions, and a log.
Which processes should agents automate first?
One process with an understandable pain, an owner and a measurable result: analysis of contracts, classification of support requests, explanation of plan-facts, monitoring of tenders. Decisions with legal consequences, access to money, changes in rights and personal data are not given to the agent without human confirmation.
How do AI agents work with internal systems and 1C?
Through integrations and APIs with a separate agent account: 1C:ERP, 1C:UH, 1C:ZUP, 1C:Accounting, SAP, EDMS, purchasing systems, Jira, Naumen, ServiceNow, 1C ITIL, GitLab, Confluence. As a rule, the agent is built on top of existing systems; their replacement is not required.
How to control the actions of an AI agent?
Three mechanisms that are designed before launch: restriction of authority, human confirmation for irreversible operations, and rollback of completed actions. Plus a log: what action, for what request, confirmed by whom and with what result.
Does the agent make decisions himself?
In responsible processes - no. The agent prepares the analysis, drafts and tips, and the decision is made by the lawyer, buyer, operator or process owner.
Is it possible to develop an AI agent for our process?
Yes. Rules, checklists, sources and roles are configured according to the customer’s regulations; if there is no ready-made module, the agent is designed as part of custom development on the common core of the platform.
A separate direction - AI agents for 1C.
Let's discuss your environment
Describe the task, current systems, constraints, and expected results. We will offer a practical first step: diagnostics, pilot, audit, roadmap or project team.
