Back to Services

Consulting

AI Agent Layer for CRM and ERP Systems

An agent layer on top of the CRM or ERP you already run: people ask in plain language, the system answers from its own data and acts through its own rules, with a human approving what matters.

AI agents CRM ERP MCP Human in the loop

Business software has always had two things: people and systems. The only bridge between them was the UI, and people adapted to the software's logic. Now a third layer is possible: people say what they need, and the system understands, answers and acts. Not a chatbot bolted onto the side, but an agent layer that works through the same rules, permissions and data as everything else.

I build that layer on top of existing CRM and ERP systems, without replacing them.

What the agent layer does

  • Answers from your data. "Which customers in Sweden haven't ordered since March?" "What is blocking the Q4 deliveries?" The agent turns the question into calls to your own services and queries, and shows where every number came from.
  • Acts through your rules. Drafting an offer, updating a deal, opening a ticket, scheduling a follow-up: every action goes through the same validation and policies as the UI, and anything consequential waits for a person to approve it.
  • Brings in the knowledge your records don't hold. Records answer what and who. Procedures, pricing rules and the reasons behind decisions answer why and how. That knowledge is kept versioned and served to the agent alongside the data.
  • Works where people already are. Inside the system, in Claude or ChatGPT through MCP, or in Slack, Telegram, Linear and Trello through the integrations your team already uses.

Built for systems that have to be right

The agent layer is designed for environments where a wrong answer has a cost, including regulated industries. That means:

  • Permissions per person. The agent sees exactly what the person asking is allowed to see.
  • Evidence with every answer. Answers cite the records and documents they came from.
  • Human approval for anything that changes data outside a narrow, agreed set of safe actions.
  • A full audit trail of questions, tool calls and outcomes.

How it is built

The core is a tool layer over your existing application, usually exposed through the Model Context Protocol, with retrieval over your documents and, where relationships matter, a knowledge graph (Graph RAG). Your CRM or ERP stays the source of truth. If it is a Laravel or PHP system, the layer lives inside it; if it is a third-party product with an API, the layer sits beside it.

What you get

  • A map of the workflows worth automating first, ranked by value and risk.
  • A working agent for one or two of those workflows, in production, with permissions, approvals and logging.
  • The tool layer and conventions to add the next workflows without starting over.

When it fits

Your team spends hours a week pulling information out of the system, re-keying it somewhere else, or asking the one person who knows how a report is built. The data is there; the bridge to it is missing.

Knowledge as Code: Connecting Business Knowledge to AI with Git and MCP Records answer WHAT and WHO. Knowledge answers WHY and HOW. Keep that knowledge in Git as one source of truth, edit it in a simple web editor, and serve it to every AI tool through MCP. AI 6 min MCP in Laravel Applications: Adding a Natural Language Layer to Your Existing System How to add MCP to a Laravel app as a controlled natural-language interface over existing services, with permissions, auditability, and production-safe tool boundaries. Laravel 11 min Why RAG Is Not Only Retrieval-Augmentation-Generation RAG sounds like three steps, but production pipelines need classification, intent, graphs, reranking, summarization, and more. Here is how real systems differ from the tutorial version. AI 12 min