Skip to content
e-ntegration
Contact
Liquid business code

Agentic engineering of business processes

We redesign a business process so that several agents run it together as a self-learning liquid workflow. On top of that we build the human-in-the-loop interfaces and flows through which your people stay in charge.

Watch a workflow runLet's talkGen‑Z: this way!
  • MCP
  • Agent Skills
  • Microsoft Azure
  • Amazon Web Services
  • EU
Liquid business code

One offering, two layers

II · Building on that

Human-in-the-loop user interfaces and flows

The interfaces and flows through which your people steer the agent: approvals, structured edits and live views of what it is doing. They work because the process underneath was engineered for an agent.

I

Agentic engineering of business processes

We take one business process and rebuild it for an agent: connections to your data through MCP, your experts' process knowledge as skills, and a check after every step.

That is all we do. Every other page of this site shows how and why.

AI bolted onto an existing workflow is not a good idea

A workflow built for people handing work to each other keeps its waiting times, handovers and blind spots when an AI step is attached to it. The better question is how the process would look if it were designed today, with an agent doing the work and people making the decisions.

Bolted on

  • The old process stays, with one more step
  • The model guesses at knowledge nobody wrote down
  • People check everything again by hand

Engineered

  • The process is redesigned around the agent
  • Expert knowledge is captured as skills
  • People decide at defined points

Watch the agent work

Pick a workflow and start it. One check fails on purpose, one step waits for you, and at the end the agent learns from your decision.

TRIGGERRequestarrivesSKILLPricingMCPERPMCPSupplierdatabaseCHECKMarginPERSONSalesapprovesACTIONSend quoteMEMORYDecision
Ready

Connections: every data source through MCP

The Model Context Protocol gives the agent one standard socket for every system. A connection is set up once per tool group, with its own credentials and permissions. All MCP servers sit behind an MCP Security Proxy: the agent reaches no data source directly.

MCP SECURITY PROXYERPCRMFilesE-mailDatabasesMachinesAgent
MCP · ERP

Reads prices, stock and orders.

The call goes through the MCP Security Proxy.

Skills: your experts' process knowledge

A skill is a folder: instructions written by the business expert and code written by an engineer. The agent loads it when a step calls for it, and follows it.

Written by the expert
# Pricing

## When to use
A customer asks for a quote with more than
one line item.

## Steps
1. Read list price and stock from the ERP.
2. Run check_margin.py for every line.
3. Below minimum margin: look up the current
   purchase price in the supplier database.
4. Never send a quote. Hand the draft to sales.
  • Reviewed by a second person before release
  • Kept in your company's skills store
  • Portable across model providers

Checks: no step without proof

Between the connections sit the checks. The agent only moves on when a check passes, and it goes back when one fails.

Rules

Deterministic program code your engineers signed off.

Data

Is the answer from the source complete and in the expected form?

Budget

Does the run stay within its token and cost limit?

Approval

Decisions with consequences wait for a person.

Ontology: the agent's model of your business

An ontology is a knowledge graph of the objects that matter in your company and how they are linked. A question such as “which customers are affected if this supplier is late?” becomes a walk along the links instead of a guess.

Select an object to see what depends on it
deliversbuilt intorunsforconcernsfound onSupplierPartMachineProduction orderCustomerQuality finding

Supplier

Who delivers what, with which lead time and quality rating.

Affected downstream

Part → Machine → Production order → Customer

When experts leave, their knowledge stays

Agents are the way to conserve what an expert knows before they retire or move on. The agent can train the successors and, where needed, stand in. That only works with an ontology as its memory, developed and verified together with the expert beforehand: knowledge extraction.

ExpertSuccessorDaily workAgent with ontology
01 / 05

Extract

The agent interviews the expert along real cases and drafts the ontology: objects, rules and the reasons behind decisions.

Harness engineering: everything around the model

The model is rented and the agent loop comes with an SDK. What makes it a system your business can run is the harness around both, and that is what we engineer for your process.

AGENT LOOPModel
01 / 08

Skills

Process knowledge the agent loads for a step.

Where agents earn their keep

Furniture manufacturing

Five agents, one factory

Manufacturing

Quality on the production line

Engineering and research

Engineering decisions over months

Energy infrastructure

From requirements to a binding specification

Shipping and long-lived assets

Projection versus reality

Systems engineering

Requirements that argue back

Knowledge transfer

Onboarding that stays current

Industrial distribution

Quotes without the manual lookup

Open source at the core, run in your cloud

Our systems run on Etienne, our MIT-licensed agent harness. You get the source, the skills and the knowledge your experts taught it, deployed to your Azure or AWS account.

Explore the platformBulloRosso/etienne
Ralph Navasardyan

Who builds it

Ralph Navasardyan began commercial software development at the age of 15 and has spent over 20 years helping international enterprises implement innovation projects using AI, cloud, and web technologies.

About Ralph

Let's talk

Tell us which decision or process your experts need help with.
E-mail
ralph.navasardyan@e-ntegration.de
Address
Weinbergstr. 29, 90607 Rückersdorf, Germany