Two New Ways to Debug Flows in Brazil

[laszloballa]

Every ServiceNow developer knows the specific misery of a flow that runs clean but produces the wrong result. No error, no red banner, execution details that say everything completed. You're left digging through logs line by line, and when you can't crack it, you open a support case and start the back-and-forth.

Brazil takes two swings at that exact problem. @Riya Kapoor , Senior Inbound Product Manager on the workflow automation team, walked through both on this week's Developer Passport. They tackle the same pain, debugging, from opposite directions: one leans on AI, one doesn't need to.

 

Flow Troubleshooting Agent

This is the AI one. When a flow breaks or misbehaves, the Troubleshooting Agent investigates on its own, pulls together a flow summary and a root cause, and hands you next steps in plain language, with links to the specific records and metadata it flagged.

Riya's demo was a real escalation the team had actually received, not a staged failure. A two-step approval flow where the second approval kept auto-approving itself. The flow was active. The execution log showed everything completing. No error anywhere. The agent dug through the approval rules, audit history, and logs, and pointed straight at the culprit: an active business rule with a script that was force-moving the record to Approved at step fifteen. The flow was fine all along, something else was fighting it.

A few things worth knowing:

• It runs from wherever Otto lives. Riya launched it from the flow execution record (so it inherits the full flow context), but you can also fire it from a problem record itself, or anywhere in the instance Otto is available.
• Drop the word "flow" in your prompt. That's the routing signal that tells Otto to hand the request to the Troubleshooting Agent instead of sending it somewhere else.
• The MVP covers approvals and triggers. Those are the two highest-volume case types the team sees, so that's where the first release focuses. More is being looked at.
• It respects your access. It won't surface scripts or tables the current user can't see. ACLs still apply, this isn't a backdoor to restricted metadata.

Availability: you need Otto (available with Now Assist for Platform). Note this one is not on personal developer instances yet, so you won't be able to test it on a Brazil PDI.

 

Input Modification

The second feature isn't AI, it doesn't need to be. It's an enhancement to the Flow Debugger (breakpoints and step-through shipped in earlier releases), and it lives in Flow Designer's new UI (built on Lux), which you opt into. It's also the one you can actually play with on a PDI.

The problem it solves: debugging a flow that only breaks under certain conditions. Normally that means the slow trial-and-error loop, tweak the data, rerun the whole flow, hope you hit the case you're after, repeat. Input Modification lets you edit a step's inputs mid-execution instead. Change the value, step over, watch what happens, all inside a single run.

In the demo, Riya edited a log message on the fly, forced an if-condition down both its true and false branches on demand, deliberately errored out a step by dropping a required field to test the catch block, and capped a lookup to two records by changing max results mid-run. Edge cases that used to take hours to reproduce, triggered in seconds.

One scope note: this covers flows and subflows, and handles standard input types (choices, numerics, text). It does not cover scripts, arrays, or Action Designer yet, though there's clear demand for the latter.

Availability: ships with Brazil and will be on your PDI as soon as Brazil lands, September 24.

 

The Common Thread

Both features are really about the same thing: winning back the time you'd otherwise lose to manual digging, and cutting down on escalations to support. One does it by having an agent read the evidence for you. The other does it by letting you bend a running flow to reproduce exactly the case you're chasing. Different tools, same goal.

 

Watch the Full Episode

The full session, both live demos including Q&A, is embedded above. If you want the Developer Passport badge, comment on the badge post before the window closes September 25. And Brazil hits personal developer instances September 24, so an Input Modification test is right around the corner, along with many more cool features that you can read about here.

Which one are you reaching for first, sending the agent after a production mystery, or reproducing the edge case that's been wasting your afternoons?

This episode covered forward-looking product plans. Development timelines and feature availability are subject to change, so base any decisions on what's actually shipped rather than what's on the roadmap.