Agentic Scheduling | Skunk Works
Bringing AI Agents to WorkWave.
6 Weeks to decide validity. Fail fast, fail forward.
What took hours, now takes minutes
Many of our products require constant monitoring of schedules, and planning future schedules for service workers. We want to take the monotonous tasks that can take hours for our customers, down to minutes to complete these tasks.
Don't replace any humans
We don’t want to take jobs away, but rather aid and assist our schedulers to free up their time to focus on other tasks to complete. In developing these AI assistants, we worked with our users, and wanted them to feel comfortable that we weren’t creating tech to replace their people.
Build trust in our AI
All this said, if our users don’t trust the AI/algorithms, then what benefit did we provide? Building trust in our systems is key for customer utilization and to make our services worth the cost.
Current results:
4000%+ Faster
At filling 100 open shifts
24hrs/day
AI Agents working
$6.2M
Generated in 2026 bookings
Define JTBD, Scope, POD teams
Week 1
Design & Infrastructure prep
Week 2
Prototype Development & User Research, User Conference in TX
Week 3
Experimentation, Testing & Integration
Week 4
User Feedback, Finalizing Design Decisions & Iterations
Week 5
Deliver MVP, Review, Final decision to continue or scrap.
Week 6
Project Background
Bringing first AI tool to our workflow, building Skunkworks teams
I was assigned to the skunkworks team developing the Agentic AI designed to integrate seamlessly across all four verticals of our products. As the lead designer for this AI scheduler, my role involves creating a user-centric interface that enables users to efficiently fill open posts, assignments, tasks, or other unassigned duties with suitable, certified, and available workers. The AI’s primary function is to streamline the scheduling process by handling the legwork, saving time for our users, while still allowing them to retain final control over who is assigned to the work. This balance ensures that the tool enhances productivity without compromising user oversight.
Target worker levels
Optimal Schedule
Employee data
Availability
Skill matching
Annual Leave
Work Regulations
Fairness
Overtime Avoidance
Productivity
The Skunkworks Team
We are operating on a new team setup similar to the Pilot Project from the TG project showcased on my home page. The main idea is one Product Manager, one Product Designer, and one Product Engineer who has a team of developers/engineers working with them. Below is a quick breakdown of the structure of the “Triad” from a presentation on the team structure. It started as a six week sprint to gather as muich data and develop an MVP to see if our efforts were even worthwhile. If we fail, that's okay, but we want to fail fast. I'll breakdown the six weeks and some of the action that took place.
12 persons
Between dev, PM, and Design
4
Indepedant Skunkwork Teams
6 Weeks
To determine future of project
Week 1 - Define JTBD, Scope, POD teams
Service Blueprinting, JTBD, How does AI even work!?
I started our project with some ground work, this included service blueprinting, defining the jobs to be done, existing customers at a conference in Texas, and attending sessions with AI trainers who taught our engineers how to build AI tools and for me to wrap my head around it. I'll start with service blueprinting and JTBD. and what we learned from it.
Persona:
Scheduling Manager
User Story:
As a manager, I want to schedule the right employees for the right shifts, so that the jobs I’m responsible for always have a qualified guard, to prevent open posts, to lower our overall budget, avoid overtime, reduce no-shows/call outs, and also make our workers happy with the shifts they’re scheduled.
Scenario:
A Worker is scheduled for a 10:00-18:00 shift, but doesn’t call-in, and doesn’t show up. WAIve scheduler finds a replacement with minimal time for the scheduling manager, but still gives them the final say.
What did I find?
There were some areas that grabbed our attention. After we completed our passes of Service Blueprinting, and mapping our JTBD, we went into Story boarding. Here's where we focused our attention to get the most value we could out of a six week sprint.
Transparency: Show all open shifts across all jobs in one place
Best Fit:
Rank users on Availability, OT avoidance, Compliance
All or None:
One click will assign all selected workers, no option to go shift by shift.
Week 2 - Design & Infrastructure Preperation
Wireframe, Prompting AI for proof of concept prototypes, building AI with AI?
After defining the scope and areas to target in the next 6 weeks, I started putting the pen to paper as they say by quickly sketching in figma using old components, shapes, and rough outlines. Instead of just immediately jumping into Figma Make or Bolt, we wanted to make sure design aligned with the scope and what was achievable in the next 5 weeks. At the same time our team was being coached on developing AI from an outside AI consultant.
Lo-fi flow
Functionalities to focus on (Green are our target, red are for future implimentation)
Core Functionality:
Ability to schedule all shifts for all jobs
Core Functionality:
Hard Constraints
PTO
Banned workers
Legal Hours
Availability
Core Functionality:
Soft Constraints:
Distance from job
Worker Preferences
Past Experience
Certifications
Seniority
Core Functionality:
Ability to schedule all shifts for all jobs
Core Functionality:
Ability to see why a shift was given to an employee. What was the reasoning?
Core Functionality:
View of individual workers schedule for the week? What about for their month? Ability to see snapshot of the workers other shifts
Core Functionality:
Ability to schedule single shifts
Week 3 - Prototype Development & User Research, User Conference in Texas
Figma Make for AI prototyping & Talking to users at Beyond Service User Conference
This is where the fun started, I used Figma Make, and Gemini to start prompting Make to craft some usable prototypes for me. I uploaded some of our design system in a previous project, so Make already had a starting block for our projects. I'll show you the prototype, and what we ended up with. Week 3 also had a 2 day Beyond Service conference in Texas, where I facilitated workshops on our Agentic AI, ran some sticky note exercises to determine pain points, and determine what they, our actual customers, need solved. The clear pain points were last minute call-offs/no-shows & the overall time spent filling open shifts for a week
Running workshops at our Beyond Service conference in Texas this year with our customers.
Rapid Prototyping for ideation:
My original prompt for getting some designs from Figma Make:
Design a high-fidelity enterprise SaaS dashboard for a Workforce Manager titled "Open Shift Command Center."
Layout: A clean, professional desktop view with a left sidebar for navigation and a main content area. The main area features a header with a prominent "Auto-Schedule" primary button (incorporating an AI icon) and a "Confirm All" secondary button.
Main Content: A list or card-based view showing "Unassigned Shifts" across 15 different job sites. Each shift card must display:
Site Name (e.g., "North Warehouse," "Downtown Retail")
Shift Time & Duration (e.g., 08:00 - 16:00, 8hrs)
Employee Slot: A placeholder for "Open" or a suggested employee name.
Badges/Chips: Visual indicators for "High Pay Rate," "Cert Required: OSHA," or "Urgent."
AI Features: Include a "Draft State" design. Show some shifts filled with a "Suggested by AI" badge and a subtle purple/sparkle background to indicate the Agentic AI made the match. Include a few "No Match Found" error states with a red warning badge.
Interactions: Ensure that a user can select/deselect which of the matched shifts they want to confirm to the schedule.
Visual Style: Modern, minimal, using our design system attached
First prototype shown to customers
Refined Prototype from Figma Make
Week 4 - Experimentation, Testing, and Integration
Rapid prototype testing, here we come!
I have experienced the benefit of testing prototypes with actual users over the past few years, and this project was no different. The catch? We don't have a ton of users on WT to pull from. We have thousands in WT Legacy, but this agentic scheduling is for the modernized WT Web, meaning our pool to choose from was smaller than previous projects. This wasn't a big issue, as we have a strong relationship with some of our key customers, and were able to experiment with them. The Agentic AI project also gave us a bit of a carrot, to entice our existing legacy users to switch over to the modernized web platform.
One of many prototypes to test usability, adoptability, and scalability with our customers.
13 interviews
Across several companies
4
Different prototypes
8.7/10
Ranking on likelihood of adoption
Week 5 - User Feedback, Finalizing Design Decisions & Iterations
Race to the finish to get the Most Valuable Product out
I have experienced the benefit of testing prototypes with actual users over the past few years, and this project was no different. The catch? We don't have a ton of users on WT to pull from. We have thousands in WT Legacy, but this agentic scheduling is for the modernized WT Web, meaning our pool to choose from was smaller than previous projects. This wasn't a big issue, as we have a strong relationship with some of our key customers, and were able to experiment with them. The Agentic AI project also gave us a bit of a carrot, to entice our existing legacy users to switch over to the modernized web platform.
One of many prototypes to test usability, adoptability, and scalability with our customers.