A connected system can still lose the work.

In a private warehouse pilot I’m building and testing, the screen said the systems were connected. A receiving record had been entered. The delivery planner’s view showed a job to plan. The dashboard looked healthy.

Then we compared the records field by field.

Some details had not made the trip. A reference from the paperwork was blank in the next view. An incomplete job did not appear where the delivery planner needed it. The connection was working in the narrow technical sense: data moved. It was not yet carrying the full operating story.

That distinction has occupied a lot of my week.

When we build a CORE workflow, we are not only connecting a form to a sheet or a sheet to a dashboard. We are connecting decisions. The person at receiving records what arrived. The delivery planner needs to know which job it belongs to, what remains open, and whether the paperwork can be trusted. The next person should not have to reconstruct a missing field from a note, a phone call, or memory.

Test the record at the other end

A green “Connected” indicator tells me the connection responded. It does not tell me that a reference on a bill of lading—the shipment document—kept its punctuation, that a job with a pending contract number stayed visible, or that a correction flowed into the next person’s view.

Those are separate tests. We found places where a field existed in one workflow and did not arrive intact in the next. We also found that a record with an unfinished contract detail could be left out of a delivery view, even though the physical work still needed planning.

After correcting which information transfers and how pending jobs are handled, we checked those problem cases again. Text references from shipment paperwork now survive the transfer, the pending job appears in the delivery list, and a delivery update reaches CORE. Those specific checks passed; the broader pilot is still being tested.

I want a system that can say, “This part is pending,” without making the entire job disappear. A blank field may mean nobody knows yet. It should not automatically mean the work is irrelevant.

The lesson is simple to state and harder to practice: test the handoff with real, imperfect records. Include a reference with punctuation, an incomplete job, a corrected entry, and a record that needs a person to verify the next step. Follow each one from intake through the view where someone will act on it.

If it vanishes, changes meaning, or appears complete when it is not, the handoff is not done.

The source and the decision both matter

There is a temptation to make every dashboard clean. Clean screens are useful, but a clean screen achieved by hiding incomplete work is expensive. It creates false confidence at exactly the moment a team needs a clear exception.

In this pilot, we are making the pending state visible and checking that updates refresh the next person’s view. The goal is for an operator to see the source fact, the unresolved item, and the next decision in one place. That is a much stronger definition of “connected” than an indicator light.

This matters beyond receiving and delivery. A lead can move into a proposal without a confirmed scope. A purchase request can reach approval without a final price. A customer handoff can look complete while one promise lives only in somebody's inbox. Every business has a place where a missing detail becomes somebody else's surprise.

Workspace Intelligence: reusable instructions need reliable records

On September 30, Google announced skills for Gemini and Google Workspace: reusable instructions for repeated tasks. Workspace rollout begins October 5 for Rapid Release domains and October 19 for Scheduled Release domains, with gradual availability. Google lists Business Starter, Standard, and Plus among the supported editions, and limits Workspace skills to users over 18. Skills in the Gemini app and Workspace do not currently sync, so a business using both must create them separately.

I see a useful opening here. Imagine a weekly operations summary. A skill could tell Gemini to read only approved receiving and delivery views, separate verified work from pending work, list exceptions with their source references, and draft questions for the person responsible. The owner reviews the summary before anyone relies on it.

But a reusable instruction cannot recover a field that never arrived. Before asking AI to summarize a workflow, make sure the workflow preserves the record. Then give the skill boundaries: do not turn an estimate into a verified fact; do not silently fill a missing reference; show the exception and the source.

That is how I want to use AI in an operating system: help people notice, prepare, and move faster while the consequential decision remains visible and owned.

BUSINESS TIP & TREND: Before skills reach your Workspace account, write one reusable instruction for a low-risk task. Name the approved source, the expected output, what the AI must never change, how it should label uncertainty, and who approves the result. Test it with an incomplete record before using it in routine work.

Source: Google Workspace skills announcement, September 30