Daynis OlmanAI, Cloud & Digital Platform Leader

Agentic OS: Governed Agentic Software Delivery

A governed operating system connecting client requests, AI-assisted triage, work tracking, GitHub delivery, review, deployment and QA.

Every request governed from intake to release.

  1. 1. Client request
  2. 2. AI-assisted triage
  3. 3. GitHub delivery workflow
  4. 4. Deployment and QA
Evidence: Public productAgentic AI and cloud platform
Role
Product architecture and engineering
Serves
Delivery teams and their clients.
Core technologies
Vertex AI · Gemini · GitHub · MCP

Executive overview

The problem. AI agents can write code, but autonomous delivery without visibility is an opaque experiment. Organisations need control, review and audit around every step.

Who it serves. Delivery teams and their clients.

What Daynis did. Product architecture and engineering. A governed operating system connecting client requests, AI-assisted triage, work tracking, GitHub delivery, review, deployment and QA.

Why it matters. Turns autonomous delivery into a controlled and visible operating model rather than an opaque AI experiment.

Business architecture and impact

Capabilities created

  • Client requests captured and triaged with AI assistance
  • Work tracked from request through deployment and QA
  • GitHub delivery with review steps
  • Command-line, agent and Slack interfaces

Turns autonomous delivery into a controlled and visible operating model rather than an opaque AI experiment.

System architecture

  1. Clients & team members

    Where requests and source data originate.

  2. Slack, CLI, agent & MCP interfaces

    The interface people use day to day.

  3. Workflow state machine

    Orchestrates requests and enforces business rules.

  4. Vertex AI / Gemini triage

    Models applied under defined, reviewable constraints.

  5. Firebase work tracking

    The governed system of record.

Governance: RBAC, audit trail, short-lived credentials
Delivery: Review, deploy, QA
Requests enter through Slack, the command line or agent interfaces, including an MCP connector. Vertex AI and Gemini assist with triage. Workflow state lives in Firebase with role-based access and an audit trail. Delivery runs through a GitHub App that uses short-lived credentials, followed by review, deployment and QA stages.

Key flows

  • Clients & team members to Slack, CLI, agent & MCP interfaces
  • Slack, CLI, agent & MCP interfaces to Workflow state machine
  • Workflow state machine to Vertex AI / Gemini triage
  • Workflow state machine to Firebase work tracking
  • Workflow state machine to GitHub App

Technical depth

Highlights

  • Firebase with Vertex AI and Gemini
  • GitHub App integration using short-lived credentials
  • MCP connector plus command-line and agent interfaces
  • Role-based access, workflow state and auditability
  • Slack integration

Architecture decisions

Explicit workflow state

Every request has a visible state, so automation never happens off the record.

Short-lived credentials

GitHub access uses short-lived tokens to limit blast radius.

Meet people where they work

Slack, CLI, agent and MCP interfaces all drive the same governed workflow.

Security and governance

  • Role-based access control
  • Auditable workflow history
  • Short-lived GitHub credentials

What I led

  • Product architecture and operating-model design
  • Engineering standards for governed AI delivery

Evidence and links

Evidence: Public productPublicly available product.

Related speaking topic

Building something similar?

Talk to Daynis about agentic ai and cloud platform architecture and delivery.

Start a conversation