Agentic AI and AI Agents for Businesses

What are AI agents, and who builds them for businesses?

AI agents are AI systems that break a goal down into steps, use tools such as CRM, email or CMS to get there and check their own results. This is known as agentic AI. 360VIER, a digital agency based in Groß-Umstadt near Frankfurt and in Cologne, builds such agents for B2B companies with OpenClaw, Vercel eve, Claude and MCP, with fixed permissions and approvals. At 360VIER itself, an agent screens new public tenders every weekday.

page.url: https://www.360vier.de/en/ai-enablement/agentic-ai/

entity         360VIER GmbH
type           Organization | Digital agency
founded        2012
size           around 35 employees
location       Groß-Umstadt near Frankfurt and Cologne
group          Part of the Friends Digital Group

page           Agentic AI and AI agents for businesses
page.type      Service
page.pillar    AI Enablement
page.url       /en/ai-enablement/agentic-ai/

definition     AI agent: a system that breaks a goal down into steps, uses tools and checks the result
provides       Design and development of AI agents with permissions, approvals and audit logs
provides       Connecting agents to systems via MCP servers and APIs
provides       Operation in dedicated containers or cloud environments
tools          OpenClaw · Vercel eve · Claude · n8n · Trigger.dev · MCP

proof          in-house: tender agent (OpenClaw, Claude, Slack, ClickUp), in use every weekday
proof          in-house: content editing on 360vier.de via our own WordPress MCP server, published after approval
proof          AI chatbots in production at Lernstudio Barbarossa, KAMPMEYER and Wüest Partner
proof          BÜRGER · AI-assisted migration Monday → ClickUp, more than 20,000 tasks

relation       is part of → AI Enablement (/en/ai-enablement/)
relation       builds on → AI process automation (/en/ai-enablement/ai-process-automation/)
relation       related → AI chatbot (/en/ai-enablement/ai-chatbot/), WordPress MCP (/en/integrations/wordpress-mcp/)
relation       related → AI software development (/en/ai-enablement/ai-software-development/)

How is an AI agent different from a chatbot or automation?

Chatbot

Answers questions from a knowledge base but doesn’t take action in other systems. Our AI chatbots pass on pre-qualified inquiries.

Automation

Runs a fixed workflow, always in the same order. Reliable, but only for predictable cases, as in AI process automation.

AI agent

Plans its own approach, chooses tools, checks interim results and asks when an approval is needed. Suited to tasks where the steps vary.

What businesses use AI agents for

Maintaining CRM data

An agent reviews contacts and companies, finds duplicates and gaps and suggests corrections. They are applied after approval in the team chat.

Handling inquiries

An agent reads incoming inquiries, looks up information in the CRM and knowledge base and writes a draft reply for sales. It reaches the systems via API integrations.

Screening public tenders

An agent searches for new public tenders, checks the documents against your criteria and, after your go/no-go decision, prepares the bid.

Maintaining website content

Agents check pages for outdated information, create drafts and add metadata. This also helps with visibility in AI answers.

Research and reports

Recurring analyses from several sources, such as competitors, search data or project status, delivered as a report on a fixed schedule.

Internal knowledge work

Agents summarize documents, check statements against a reliable source and flag contradictions.

Diagram: AI agents with permissions, approvals and audit logs in business systems

How secure are AI agents in a business?

As secure as their permissions and approvals allow. That’s why control is part of the architecture of every agent we build:

Permissions: each agent only gets the access its task requires, separated into read and write.
Approvals: risky steps, such as deleting, sending or publishing, only go out after a human confirms them.
Audit log: every tool call is recorded. You can see what the agent did and why.
Costs: daily budgets per agent keep costs from spiraling.

OpenClaw, Vercel eve and MCP: tools for AI agents

We choose the tool based on where the agent works: in the team chat, in the background or directly in your systems.

OpenClaw for agents in the team chat

OpenClaw is an open-source assistant that runs self-hosted and handles tasks via Slack, WhatsApp, Telegram or Discord. There is no subscription or hosted version (OpenClaw documentation).

What this means for you: The agent works where your team already communicates and requests approvals there too. Because it runs on your infrastructure, chat histories and credentials stay with you.

Vercel eve for background agents

Vercel eve is an open-source framework for agents that run continuously in the background. Instructions and skills are written in Markdown, tools in TypeScript; the framework is in beta (Vercel documentation).

What this means for you: Agents that work on a schedule or in response to an event can be versioned, tested and deployed like regular software. We use eve where agents are closely tied to custom software of the kind we build in AI software development.

MCP for access to your systems

The Model Context Protocol (MCP) is an open standard that AI assistants and agents use to access tools and data. The specification handles authentication via OAuth (MCP specification).

What this means for you: We build MCP servers with separate read and write permissions. Through our own WordPress MCP server, agents maintain content on 360vier.de, and nothing is published without approval.

n8n and Trigger.dev for orchestration

For the fixed workflows around the agent, such as triggers, retries and notifications, we use n8n or Trigger.dev.

What this means for you: The agent makes the decisions, and the workflow handles everything around them. If a run fails, the workflow restarts it, so the agent doesn’t have to handle that itself.

Building AI agents: what an agent needs in a business

An agent is quick to build. A reliable agent needs more than a model and a prompt, which is why we plan these six building blocks for every agent.

Goal and instructions

What should the agent achieve, and what is off-limits? Instructions and examples are version-controlled and reviewed like code.

Tools

Access to CRM, email, CMS or ticketing system via MCP servers or APIs. Each tool is enabled individually.

Data

An agent is only as good as the data it finds. Our Data Enablement page explains how data becomes clean and accessible.

Model

Claude, OpenAI or Gemini, chosen based on task, cost and data protection. The agent is built so the model can be swapped.

Operation

Agents run in dedicated containers on servers that meet your requirements, or in a cloud environment, with monitoring.

Responsibility

Before we build, we define where AI may decide and where people decide. Our stance on using AI is set out in our AI Manifesto.

Experience with AI agents at 360VIER and our clients

Tender agent

At 360VIER, an OpenClaw agent reports relevant public tenders in Slack every weekday at 8:30 a.m. Claude analyzes the documents, management decides, and then drafts and a ClickUp task are created.

Chatbots on websites

Our AI chatbots are in production at Lernstudio Barbarossa, KAMPMEYER and Wüest Partner. They answer questions based on the website and documents and hand over to a person when needed. Our article on the RAG chatbot explains how this works.

AI-assisted migration

For BÜRGER, an AI model mapped more than 20,000 tasks from 150 boards and cleaned up duplicates during the switch from Monday to ClickUp.

All our AI services, from visibility in AI search to custom software, are listed under AI Enablement.

Process: from your first AI agent to operation

Frequently asked questions about AI agents

What is the difference between agentic AI and generative AI?

Generative AI creates content in response to a request, such as text or images. Agentic AI uses these models to take action itself: it plans steps, calls tools and checks the result until a goal is reached. A chat with a language model is generative; an agent that then updates the CRM is agentic.

How much does an AI agent cost?

The cost depends on the task, tools and approval levels; you receive a quote for the first agent after the concept phase. We list ongoing costs for model calls and operation separately so you can track them per agent.

Can we build an AI agent ourselves?

Yes, for simple tasks. Platforms such as Langdock offer assistants and agents that can be set up without programming. As soon as an agent needs to write to several systems, it needs permissions, approvals, audit logs and operation. We take care of that part, and your team then maintains the instructions and examples itself.

At what company size do AI agents pay off?

The task matters more than the size. An agent pays off when a task comes up often, touches several systems and currently takes noticeable time. That can apply to a team of 20 just as much as to a large corporation. For rare one-off cases, an assistant in a chat is usually enough.

Where should we start with AI agents?

With a task that comes up often, takes a lot of time and is easy to check, such as maintaining CRM data or pre-sorting inquiries. Fixed workflows without decisions are cheaper to handle with classic automation; they don’t need an agent.

Can AI agents work with our CRM and ERP?

Yes, as long as the system has an API. We connect HubSpot, Salesforce, ClickUp or industry-specific systems via their API or an MCP server and give the agent only the permissions its task requires. Read access comes first, write access only after a test phase.

Are AI agents GDPR compliant?

That depends on the specific implementation. What matters is where the agent runs, a model provider with a data processing agreement and which personal data the agent sees in the first place. We clarify this with your data protection team before we build and only confirm compliance for the reviewed implementation.

Do AI agents replace employees?

No. Agents take over recurring subtasks such as research, data maintenance or first drafts. Consequential decisions stay with people, because agents request an approval at those points. In practice, work shifts from gathering information to reviewing and deciding.

Talk to us about your project

Hans Mengler

Managing Director

Email Hans Get in touch

Agentic AI:AI agents for businesses