Teach it the work like a new hire. It builds the agent.
Share your screen and work through the job, or describe the outcome and let it plan. Loop Studio asks the questions that get the rest out of your head, writes the plan, and builds the agent that runs it.
No spec docs·Deterministic runs·On the free planEmail triage, then. Walk me through it.
Screenshare and explain1:15Screen recording · Email triageThanks, that is everything I need.
Building your automation…Here is what you do. I will do it without fail, and flag anything out of the ordinary.
Show it, or tell it.
Most processes are easiest to demonstrate, some are easier to describe. Either way ends in the same plan.
- Screenshare itShow the work out loud
Open the Chrome side panel and share your screen. Loop Studio tells you what it needs to see, follows along in your real tools, and asks about the cases you would never have thought to mention. No spec document, no process map.
- Describe itSay the outcome, and it plans the loop
Some work has never been done by hand, or is spread across too many people to demonstrate. Give it the source, the destination and the rule, and it drafts the plan and names what it still needs to know.
A hybrid agent, not a prompt.
The plan becomes an agent with two kinds of step, and the line between them is drawn in the plan where you can see it.
- Fixed steps where the work is fixed
Opening the record, pulling the report, filing the document, updating the stage. Deterministic code does these, so the same input takes the same path every time.
- Judgment where it is not
Classifying a message, matching a name that is spelled two ways, deciding whether a document is complete. AI does these, inside a step with a defined input and output.
- Held when it should not proceed
Anything the plan says to stop on, it stops on: a missing signature, an amount that does not reconcile. It raises the case for a person and does not write.
It asks what is only in your head.
A screenshare shows the normal case. Loop Studio asks for the rest in plain questions, and every answer becomes a rule in the plan before the first run: what if only one party has signed, what if the matter is not in the system yet, what if the amount is off by a cent.
When the process changes, tell it or show it.
The process will change, and someone new will inherit it. Tell Loop Studio in a sentence or share your screen again, and it proposes the edit to the plan for a person to approve.
The loops nobody was doing by hand.
Some work never sat on a desk to be demonstrated: data that has to move between systems, records that have to match, a process still held together by a shared sheet. That is what the second way in is for: describe it, and it runs on a schedule.
- On a scheduleMove data between systems every morning
Pull the day's records from one system, reshape them, and post them where the team actually looks. The kind of push a spreadsheet macro or a nightly export does today, without the export.
- ReconcileMatch records across two systems
Deposits to matters, invoices to purchase orders, signed agreements to the CRM stage. Post what matches, hold what does not.
- ElevateLift a new process off spreadsheets and email
A new process runs on a shared sheet and an inbox for its first quarter. When it settles, describe it as it is now and build the loop around it.
Questions we get asked
Do we need to document the process first?
No. The screenshare is the documentation. Caddi writes the plan from what it watched and what you said, and the plan is what you review. Firms with a clean process map are welcome to bring it, but the ones without are not behind.
Is it only watching the screen?
No. While you share, Caddi reads what the browser and the tools are actually doing underneath the page, not only what is drawn on it. That is how it knows which system a step happened in, which record you opened and which field changed, instead of inferring it from a picture. The plan is built on those operations, which is also why it runs over APIs rather than by clicking around a screen.
What if the process is messy, or three people do it three ways?
Talk it through as it is. Loop Studio asks about the cases it cannot see, and every answer becomes a rule in the plan. Where people disagree, the plan makes the disagreement visible, which is usually the first time anyone has written it down.
What actually runs the work?
Deterministic code, generated from the plan and run in the cloud. AI is used inside specific steps that need judgment, with a defined input and output. Nothing re-decides the process mid-run.
What does it need access to?
One scope per connected system, under the access the person already has there. No standing admin credentials. You can see the scope each step used in the run record afterwards.
What happens when the process changes?
Say so in the chat, or share your screen again and show the new part. Loop Studio proposes the change to the plan, a person approves it, and the next run takes the new path. No rebuild and no consultant.
Can it move data between systems on a schedule?
Yes. A loop can run on a schedule as well as on a trigger, and reading from one system to write into another is most of what back office work is. Describe the source, the destination and the rule.
One piece of honest advice: if the source is a large reporting system, have your data team consolidate what the loop needs into one view or table first. A loop that runs a few precise reads is cheap and fast. One that runs broad queries across many sources on demand is neither, on any automation platform.
Which plan is it on?
Loop Studio is on the Individual plan, which is free for one user with no time limit. See pricing.


