The 10-second win

You open ChatGPT. You type:

Write a WQL query that returns all active workers in the Finance supervisory organization with their hire date and job title.

Ten seconds later, you have a query:

SELECT workdayID, fullName, hireDate, jobTitle
FROM allActiveWorkers
WHERE supervisoryOrganization = 'Finance'

It looks clean. It uses reasonable field names. It seems right. Now what?

The 10-minute reality

Here's what actually happens next, every single time:

  1. Copy the query from the ChatGPT window.
  2. Open your Workday tenant in another browser tab. Sign in (possibly with MFA).
  3. Navigate to the WQL query tool.
  4. Paste the query. Run it.
  5. Get an error. The data source name is wrong. ChatGPT used allActiveWorkers, but your tenant's data source is actually called allWorkers with an isActive field you need to filter on. Or the field name is workerHireDate, not hireDate. Or jobTitle isn't a direct field — it's nested under a related object.
  6. Go back to ChatGPT. Paste the error message. Explain what went wrong. Wait for a revised query.
  7. Copy the new query. Go back to Workday. Paste. Run.
  8. Get a different error. Repeat from step 6.

After three or four rounds, you either have a working query or you've given up and written it yourself. Total time: 10 to 30 minutes for a query that ChatGPT generated in 10 seconds.

Why this keeps happening

ChatGPT has no access to your Workday tenant. It doesn't know which data sources exist, what fields are available, how fields are named in your configuration, which fields are filterable, what data types fields return, or your tenant's custom objects and custom fields.

ChatGPT generates WQL based on patterns from training data — documentation, forum posts, examples. That training data reflects some Workday tenants at some point in time. Your tenant is different, and Workday may have changed since the model was trained. Every query is a guess.

What closing the loop means

Imagine a different workflow:

Show me all active workers in Finance with their hire date and job title.

The AI agent:

  1. Looks up the actual data sources available in your connected tenant
  2. Identifies the correct field names from your tenant's WQL catalog
  3. Writes a query using those real names
  4. Runs the query against your tenant
  5. Shows you the results — actual data, not a text prediction

If the query returns unexpected results, you say: "That's only showing 12 workers — Finance should have about 40. Can you check if it's filtering out contingent workers?" The agent adjusts, re-runs, and shows updated results. The entire iteration happens in one conversation, against real data.

Beyond WQL: the same gap everywhere

WQL is a concrete example, but the same gap applies to everything a Workday admin does with AI:

  • Calculated fields: ChatGPT can write an expression, but does it use the right function names for your Workday version? Does it reference fields that exist?
  • Custom reports: The AI can describe how to build a report. But you still need to navigate to Report Writer, find each field, apply filters, and check the output.
  • Business process configuration: The AI can explain how to add a step. But it doesn't know your current process, your security policies, or your condition rules.
  • Security configuration: The AI can explain domains and policies. But it can't see your tenant's actual security model or verify whether a change will break access.

In every case, the AI produces text about Workday. Turning that text into changes in Workday requires manual work that the AI can't help with — because it can't see or touch your tenant.

What kiweely does differently

kiweely is a desktop app that connects directly to your real Workday tenant. For WQL work:

  • The agent discovers your tenant's actual data sources and fields — no guessing
  • It runs WQL queries against your connected tenant — you see real data
  • It can query through the Workday UI when a direct API isn't available
  • It iterates in one conversation — adjust, re-run, see results, refine
  • It handles the WQL ceiling honestly — when a row limit applies, the agent tells you

For broader Workday admin work, kiweely connects the same way:

  • Read and modify tenant configurations through Workday's APIs, with consent for every write
  • Open a browser already signed into your tenant for visual verification
  • Access Workday's API contracts for accurate field references
  • Search Workday Community and the Developer Forum for the latest knowledge

A simple test

Next time you ask ChatGPT for a WQL query, time how long it takes from the moment you get the query to the moment you have the data you need in your Workday tenant. Include every tab switch, every error, every paste, every re-prompt.

If the answer is "less than a minute," your query was simple and your tenant is vanilla. If the answer is "10 minutes or more," you've experienced the gap. The AI wrote the query in 10 seconds. The remaining 9 minutes and 50 seconds were the cost of disconnection.

AI that writes a query is a convenience. AI that writes a query, runs it, and shows you the data is a tool. The difference is connection.