The Genie Doesn't Decide. You Do.
The first time someone uploaded a PDF to the Genie, I watched what happened very carefully.
They'd been emailing that PDF to new clients for four years. It was a client intake form — twelve fields, a signature block, a section for authorization. The Genie analyzed it, classified each section, and extracted the fields into a draft packet in about eleven seconds. Then it stopped and presented the result for review.
The person leaned forward, checked one field the Genie had labeled as "text input" that should have been a date field, fixed it, and clicked confirm. Total time: under two minutes, including the upload.
What I was watching for wasn't the speed. It was whether they trusted it. Not blind trust — the kind of trust you extend to a capable colleague who does the tedious work and hands the result to you to verify. That's the mode I'd designed for. And it's the mode I think AI should operate in for any workflow where the stakes are real.
In 2026, the pitch for AI in business software usually sounds like "automate everything." Upload your documents, the AI processes them, the workflow runs. You don't have to look at it. But in onboarding and compliance — where a miscategorized field in a W-9 or a misread entry in an I-9 becomes a legal record — "don't have to look at it" is not a feature. It's a liability transfer.
The principle behind the Genie is human-in-the-loop AI: the AI does the time-consuming analytical work, and a human reviews, adjusts, and approves the final output before it becomes part of any official workflow. The AI is not autonomous. It is transparent, specific, and correctable.
What does "human-in-the-loop" actually mean in practice?
Human-in-the-loop means the AI produces a draft, not a finished product. When you upload a PDF to the Genie, powered by Vertex AI Gemini 2.5 Flash, it returns a structured proposal: these are the fields it found, these are the field types it assigned, this is how it's suggesting they be ordered. You see exactly what the AI interpreted and why.
If the Genie reads a field labeled "Date of Birth" and marks it as a text field instead of a date picker, you see that. You change it. The system doesn't go live until you confirm. That's not a limitation of the AI — it's a design decision about where human judgment belongs in the workflow.
The alternative — AI that publishes directly without review — trades short-term speed for long-term risk. If the AI misclassifies a form field and the packet gets sent to 40 new hires before anyone notices, the correction is not just a software fix. It's a re-send, a re-completion, and potentially an explanation to anyone who already submitted the wrong format.
Why does AI transparency matter more for small firms than for enterprise?
Enterprise deployments have compliance departments, legal review, and IT oversight. When an AI-generated workflow produces an error, there are multiple layers to catch it before it becomes a problem.
A 12-person consulting firm does not have that infrastructure. The founder or office manager is the compliance department, the legal reviewer, and the IT department simultaneously. If an automated AI workflow produces a silent error in an I-9 or a W-4, there is no second layer. The error becomes a record.
For small firms, the cost of "set it and forget it" AI is higher, not lower, than for enterprise. The firms that can least afford compliance errors are the ones most likely to be pitched on full automation. That's the gap I was designing against.
What the Genie does, and what it doesn't do
Here's the honest breakdown of where AI operates in OnboardingGenie and where it doesn't:
| Task | AI involvement | Human involvement |
|---|---|---|
| Analyzing uploaded PDF forms | AI extracts fields and classifies document type | Human reviews and confirms all extracted fields |
| Suggesting packet structure from a prompt | AI proposes steps based on described workflow | Human edits, reorders, and finalizes before deployment |
| Routing packets to recipients | Automated | Human initiates each send |
| Completing I-9 Section 2 | Not involved | Human (or authorized representative) examines documents |
| Generating compliance reports | Automated PDF/CSV generation | Human reviews before sharing with auditors |
| Making hiring or compliance decisions | Not involved | Human, always |
The Genie is fast at analysis and structure. It is not involved in anything that constitutes a legal decision, a compliance determination, or a personnel action. Those stay with you.
Frequently asked questions about AI in OnboardingGenie
Can the Genie send packets to recipients automatically without my approval?
No. The Genie assists with building and structuring packets. Sending a packet to a recipient is always initiated manually by the account holder. The AI does not trigger outbound communication on its own.
What AI model powers the Genie?
The Genie runs on Vertex AI Gemini 2.5 Flash. It is accessed via Google Cloud's Vertex AI infrastructure, which means the processing happens server-side rather than in the browser. Documents you upload for analysis are not used to train any model.
What happens if the Genie misreads a field in my uploaded PDF?
You'll see it in the review step before any packet goes live. The Genie's output is always presented as a draft, not a finished workflow. Every field it extracts is visible, labeled, and editable before you confirm. Nothing is hidden or finalized without your explicit approval.
Is the Genie available on all OnboardingGenie plans?
The Genie document import feature — uploading a PDF and converting it to a digital packet — is available on the Pro plan. The prompt-to-template feature, where you describe a workflow and the Genie builds a draft packet, is available on the Pro+ plan.
The discipline of knowing where AI belongs
I've spent years in software development and education, and the pattern I keep coming back to is this: the best tools are the ones that amplify what a person already knows how to do, not the ones that replace the knowing entirely.
The Genie is fast at reading PDFs. You are better than any AI at knowing whether a field labeled "Authorization Signature" should be a drawn signature or a typed acknowledgment in your specific client relationship. That knowledge matters. The Genie removes the eleven seconds of reading time so you can apply the two minutes of judgment time. That's the division of labor that produces reliable results.
Transparent, correctable, human-approved — that's what responsible AI looks like in a compliance workflow. See how the Genie works in a 30-day free trial.
Founder, OnboardingGenie