Case studiesBusiness operations

AI Employee: agents for connected business workflows

Built for Cognum, AI Employee brought repeated business tasks into an agent platform with organizational memory, connected tools and visible execution history. Ego Eimi’s work covered research, product design and engineering.

Business
Cognum
Sector
Business operations
Scope
Agents, memory and integrations
Contributor
Ego Eimi
Editorial illustration: a coordinated camshaft assembly with one orange cam
AI-generated editorial illustration of the project’s domain.

Repeated work crossed tools and needed business context.

A team automating prospecting, service or back-office work needs more than a chat response. The task can involve several accounts, information from previous work and actions that change a business record. People need to understand what happened and where to intervene.

The platform therefore needed workflow controls, memory, permissions, connectors and execution status. These requirements shaped the product around carrying out work across tools, rather than treating every request as an isolated conversation.

Separate orchestration, memory, permissions and connectors.

The application separated orchestration, memory, integrations and authentication into Dockerized services. A React interface exposed workflow execution and live status, while Node.js services coordinated the tasks.

LangChain supported orchestration across model providers and internal tools. MongoDB Atlas Vector Search held organizational knowledge and task history. OAuth connectors connected services including Gmail, Slack, HubSpot, Salesforce, Notion and PostgreSQL.

What the workflow includes

  • Give workflows reusable skills and access to task history.
  • Connect business accounts through OAuth authorization.
  • Apply role-based access and record execution events.
  • Package services for deployment through Google Cloud Run.

An application for running and inspecting connected tasks.

The product combined the pieces needed to run connected administrative workflows: a user interface, orchestration, memory, integrations and a history of execution. Teams could follow a task through the application instead of treating its final answer as the only output.

What this means for the technology owner.

When agents act inside business tools, their permissions and failure paths become part of operating the system. A technology owner should know which accounts each task can reach, how execution is recorded and how to stop or recover an unsuccessful run. Model changes need the same review discipline as other production changes.

How we operate AI workflows →

Node.jsReactTypeScriptLangChainMongoDBPostgreSQLDockerGoogle Cloud Run

Project contribution: Ego Eimi. The scope described above identifies the work behind this case.

Open-source codebase — project link

Editorial review: Ego Eimi · Updated September 6, 2026 · About the team

Related case studies

Editorial illustration: a sculptural indexing instrument with an orange tab
SEO software

SearchAtlas

Agent orchestration and persistent execution

Read the case →
Editorial illustration: a modular knowledge organizer with one orange fitted module
Business AI software

UtopikAI

Models, retrieval and team workspaces

Read the case →
All case studies46 projects

Your technology

Start with the business

Your business depends on it.

Tell us which system matters, what has changed and who controls it. We will define a reviewable next step. Read Foundation and the commitments and evidence.

Start a conversation More case studies