Site manager
- Behaviors
- Planning, assignments, leading the on-site crew
- Goals
- Keep the plan current; run the week at high operational efficiency
- Pain
- Missing timesheet data, delayed invoice approval
ShiftForge AI
AI agents that handle the busywork of running a hybrid workforce — so the people running it can focus on the decisions that actually need a person.
Overview
Operators spend most of their week on reactive problem-solving and manual data tracking. Only 7% have real-time, end-to-end network visibility.
Three things break the same way every time: staff key in line items by hand, invoices stall waiting to match POs and receipts, and approvals sit in someone's inbox.
Hand-keyed line items
Staff type timesheet detail into a second system by hand, every week, from memory.
Invoices that stall
Nothing ships until a PO, a receipt, and a timesheet agree — and no tool owns the matching.
Approvals in an inbox
The next action lives in someone’s email, so exceptions surface days after they cost money.
Design goals
- Auto-build the plan — let AI match work orders to POs and draft invoices, so a week stands up without hand-keying.
- Give every task an owner — assign clear tasks so invoices don’t get lost and exceptions get caught early.
- Show overtime before it happens — at the moment of assignment, not on the invoice.
Research
Not one vendor fit both sides of the problem.
I scored the top 14 vendors in labor operations software on two things: how well they fit a factory floor, and how well they fit a staffing agency's economics. None scored well on both.
| Vendor | Category | Staffing fit | Mfg fit | Hybrid |
|---|---|---|---|---|
| Bullhorn | Staffing OS | 5 / 5 | 1 / 5 | 2 / 5 |
| TempWorks | Staffing OS | 4 / 5 | 2 / 5 | 2 / 5 |
| Shiftboard | Mfg WFM | 1 / 5 | 5 / 5 | 2 / 5 |
| UKG WFM | Mfg WFM | 2 / 5 | 4 / 5 | 3 / 5 |
| ShiftForge | Orchestration layer | 4 / 5 | 4 / 5 | 4 / 5 |
What that gap costs, every week
~4 hrs
of a coordinator’s week, reconciling timesheets by hand
15–30%
of invoices need a rate or overtime correction
3–5 days
to resolve a billing dispute over email
Personas
Two roles, one week, two systems that don't talk.
Plant supervisor
- Behaviors
- Data-driven calls on the floor, leading shifts
- Goals
- Operational efficiency and quality assurance
- Pain
- Backfilling positions fast, unexpected overtime
“If a temp worker goes into overtime, I don’t find out until the invoice comes.”
“We approve time in one system, then re-enter it in another to generate the invoice.”
Design Challenge 01
How might we hand the repeating parts of the week to an agent without handing over the decisions?
The operator’s week is four repeating jobs across two systems that do not communicate: onboarding a worker, assigning them to a shift, backfilling when someone drops, and reconciling the invoice at week’s end. Basic automation handles the happy path; an agent has to handle the exceptions.
The AI constraints I designed around
Made-up references
The agent sometimes cited time entries that did not exist. The app now checks every reference against real data before showing it.
Version drift
The same question got different answers from different model versions, so I locked the answer to a fixed shape the app validates every time.
Over-narration
The agent wanted three paragraphs where one sentence would do. Length is capped in the prompt and again in the app.
Fake confidence
It would confidently invent what it did not know. I gave it a way to say “I don’t know” — and made that the only option when it isn’t sure.
Cost per use
Paying a vendor per question added up fast, so for the pilot the agent runs on the customer’s own machine on a free open-source model.
Multi-agent design
Three agents. Each gets a bounded scope, a typed output, and a named place where a human takes over.
| Agent | Input | Output | Approval boundary |
|---|---|---|---|
| Reconciliation Agent | This week’s time entries and rate cards | 3–5 typed flags with citations and proposed resolutions | Plant supervisor approves any flag that affects billing |
| Coverage Agent | Next week’s demand and the current roster | Coverage flags, ranked candidate suggestions per open shift | Coordinator dispatches, supervisor confirms |
| Workspace Assistant | The active surface plus the palette query | Free-form chat scoped to the current view | Never moves state, only proposes |
The two surfaces I redesigned
Planning
Standing up next week's coverage
Timesheets
Reviewing the week and catching overtime
How the week got designed
Research
- Conversations with staffing coordinators and plant supervisors
- Mapped weekly workflows and pain points
- Identified primary users and jobs to be done
Scope
- Mapped user jobs to features, grouped by role and approval boundary
- Ranked by impact and effort, cut to what could ship as v1
UI design
- Pulled references from operations and fintech tools
- Studied approval and reconciliation patterns in adjacent products
Prototyping
- Several rounds of HTML mockups, low-fi to high-fi
- One shared shell across planning, timesheet review, and reconciliation
- Handed off to build with Claude Code in Next.js
Key Solution
The five capabilities that make it work.
Assign work without lifting a finger
When a shift needs coverage, a ranking algorithm sorts available workers across every connected agency — by certification, availability, cost, and distance — and hands the supervisor a shortlist with the reasoning attached, not just a name.
Human approval when it matters
Overtime, missing punches, geofence issues — all flagged automatically with a proposed fix attached. Nothing touches pay or a worker’s record until a named person clicks approve.
Take action before it’s too late
The moment a worker crosses the overtime threshold it is flagged — not discovered a week later when the invoice lands and there is nothing left to plan around.
Accountability that highlights exceptions
Before the agent writes a word of an invoice, it retrieves the real timesheet, rate card, and policy behind that line, and shows what it looked up. It can’t invent a number it can’t point to.
Keep track of how it all happened
Every flag carries its evidence. Every approval is logged with who, when, and why. Every action can be reversed by a person with the right permissions.
Impact
Simulated against baselines from prior research.
~80%
less planning setup, with auto-built plans
~100%
of surprise overtime designed out at assignment
0
chasing — the next action is always on screen
Next Steps
Sign design partners and run the product against their actual weeks.
Next is a small group of staffing agencies and manufacturers as design partners. Running against their real weeks is how we measure the three things that matter: overtime caught before it billed, hours cut from the invoicing cycle, and follow-ups avoided on exceptions and reconciliations.
