Designing Genie: Pioneering Agentic AI for Delivery

Genie is an agentic orchestration layer for delivery operations. It connects to your live project ecosystem—Jira, Slack, Mavenlink, project databases—and continuously synthesizes context, surfaces insights, and automates workflows without manual intervention.

At its core, Genie does four things:

1. Consolidates Context: Pulls information from 10+ systems and maintains persistent awareness of project state, giving you a reliable unified view instead of fragmented data.

2. Generates Requirements: Auto-generates backlogs, requirements, and scope documents using historical patterns and project data, and improves as it learns.

3. Surfaces Risks: Flags emerging problems, sending daily alerts and weekly summaries that highlight signals worth paying attention to.

4. Automates Workflows: Creates tickets, sends reminders, generates decks, and publishes updates so teams can spend their time on decisions, not busywork.

Every output shows sources and reasoning, because Genie was designed to be a collaborator, not a black box.

Our work was evolving, and so was the industry around us. Like many of my teammates, I was spending the bulk of my time on tasks that kept projects moving, but that steadily pulled my attention toward repetitive work that didn’t take work to the next level. It was a lot of effort for what was ultimately table stakes.

The early work pattern in a project is always the same: gathering information from scattered sources, shaping it into user stories, recreating requirements written many times before, re-running audits, and assembling backlogs that shift as a project unfolds.

None of this work is wasted, but it consumes the energy needed for strategic problem-solving. Designers, developers, and delivery leads were spending their sharpest hours on setup instead of judgment-driven decisions.

The challenge wasn’t about squeezing more output from teams. It was about protecting the creative and analytical focus that makes delivery meaningful. The question became simple: if we could automate the predictable work, could we reclaim time for the moments that move projects forward?

The first attempts to solve this weren’t part of a formal initiative. They emerged in pockets— 30 minutes between meetings, an hour on a Friday, small automations and custom GPTs stitched together to run audits, generate user stories, and map components. These were practical attempts to make daily work less repetitive and to explore emerging tooling—Nn8n, workflow orchestration, agentic patterns—through real problems.

Then something unexpected happened: everyone wanted it.

It became obvious that the pain wasn’t isolated. Every team felt it. And the experiments hinted at something bigger: if these tools were connected, orchestrated, and expanded, Orium could build an agent capable of supporting the entire delivery lifecycle.

What started as lightweight experimentation quickly turned into momentum. The company wasn’t just looking for efficiency; it was looking for a way to shift the balance of work back toward what humans do best. The emerging agent systems weren’t replacing expertise— they were clearing the path for it.

That’s the moment Genie moved from individual hacks to an organizational initiative. A shared recognition that Orium needed an intelligent layer capable of unifying context, reducing manual load, and giving teams the space to focus on the work that actually moves projects forward.

The fundamental question at the heart of Genie was big: How do you design for an intelligence that's partly autonomous, partly responsive, and operates across systems that users might not even see?

Designing Genie wasn’t like designing a typical interface, and though the design patterns for chatbots and assistants are well-established, agentic AI is different. It's proactive, multi-modal and capable of acting across systems autonomously. We couldn't just apply existing design patterns— we had to pioneer new ones.

1. Capability Clarity: "What Can Genie Actually Do?"

When I started sharing early experiments, it became obvious that people didn’t know what to ask for. I had built tools around my own workflow in my experiments, but everyone else approached projects differently. They weren’t sure whether Genie could help, or even what to try.

• Project leads wondered if it could draft requirements the way they would.
• Delivery managers asked whether it understood our estimation patterns.
• Designers wanted to know if it could flag gaps in scope.

But no one instinctively turned to Genie. They didn’t yet have a mental model for where it fit. The design challenge became helping users discover capabilities at the moment they needed them, not buried in documentation.

2. Trust & Transparency: "Should I Actually Believe This?"

This was the hardest problem, because it goes to the heart of working with autonomous systems. Agentic systems make decisions with consequences. When Genie flags a risk, drafts a backlog, or creates a Jira ticket, someone has to judge whether to trust that output, and what to do next.

• A project manager gets a risk alert. Do they act, escalate, or ignore it?
• A 20-ticket backlog appears. Is it ready to approve, or does it need rework?
• A scope summary is generated. Did it capture the essentials, or miss something important?

We had two choices: aim for high accuracy and let users navigate the gaps, or design for transparency so they could verify every output. We chose transparency. Genie shows its sources, explanations, and confidence signals so trust isn’t assumed, it’s earned.

3. Autonomy vs. Control: "Who's Actually in Charge?"

Genie works across several modes—answering questions, surfacing insights, executing workflows, generating artifacts—and the real risk is when users can’t tell which mode they’re in.

• A user asks for something → Genie produces it on demand
• A threshold is hit → Genie proactively issues an alert
• Genie spots an opportunity → It suggests an action, but needs approval
• A scheduled trigger runs → Genie executes a workflow on its own

Without clear signals, users could be confused: Did I make this decision or did the agent? Did I approve this action or did it happen automatically? And in enterprise contexts, confusion + autonomous systems = serious problems. We needed absolute clarity about who’s in control at every moment.

The measure of Genie’s success isn’t throughput. It’s what teams are doing with the time they now have back.

Project leads used to spend days gathering documents, synthesizing information, building spreadsheets, and manually creating Jira tickets and requirements. Today that work takes a fraction of the time (and most of it is reviewing what Genie generated, not doing it from scratch.)

That means project leads have time for what matters far more. As one delivery leader put it, "Instead of filtering beans on spreadsheets, we can actually be developing deeper relationships and better understanding our customers, their brands, and their needs."

We didn't just automate work; we freed up cognitive space for the strategic, human-centered work that actually drives project success.

There’s a business impact too. Early context means fewer risk escalations. Shared understanding means fewer requirement clarifications. And because teams start with stronger foundations, Orium can offer fixed-cost, fixed-scope engagements with more confidence.

Across new projects, teams are adopting Genie enthusiastically because the outputs are reliable, transparent, and easy to trust. But the real signal is this: the time it saves is being reinvested in judgment, creativity, and collaboration—not more routine work.

Genie hasn’t replaced expertise. It’s amplified it.