jordancbaron
Administrator

 

now[dev]26 Keynote Recap: The Debt Behind the Wall (and How to Fix It)

 

Missed now[dev]26 in San Jose? Not to worry! We're bringing the keynote to you. Sit back, relax, and enjoy the video above!

 

Video too long for the middle of the workday? Totally fair. 

 

Here’s the scroll-friendly version of everything we covered on stage in San Jose: five familiar developer headaches, one fictional (but very relatable) company, and a lot of “wait, it can do that?”

 

 

TL;DR

 

  • Green doesn’t always mean right. A flow can run perfectly and still send your P1 to the wrong team. Clean flows need clean data underneath them.
  • Duplicate CIs are sneaky. Don’t delete them. Merge them into the real CI and retire the twin.
  • Inherited an app nobody understands? Build Agent can explain how it works and map what might break before you touch it.
  • Big modernization project? Autonomous Engineer turns a PRD into a plan, then builds it piece by piece while you stay in control.
  • Love your own tools? Use Claude Code in VS Code with the ServiceNow SDK, and your team’s Build Agent rules come along with you.
  • ToolSmith builds integrations and agent capabilities in minutes, not weeks.
  • One Notify step in EmployeeWorks can reach every employee on Slack, Teams, email, text, mobile, and more.

 

261105-NowDev26-EL1_7756.JPG

 


 

Built for builders

 

Amit Zavery and Bhavin Shah (founder of Moveworks, now leading Employee Experience at ServiceNow) kicked things off with a promise: no more AI hype. now[dev] is a space to build together, with hands-on labs, Ask the Experts sessions with the PMs behind the products, and of course, Hackzones.

 

Bhavin compared enterprise development to renovating an old house. AI makes planning easy: floor plans, designs, even material costs. Then you open the wall and find wiring with no diagram, a pipe going nowhere, and tape where screws should be. 

 

AI doesn’t always know your integrations, your data models, or the decisions your team made years ago. But someone in this Community has probably already solved the problem behind your wall.

 

Amit backed that up with a few numbers:

 

  • 97% of answers in the Community come from other builders. That’s over a million people who hit the same bug and shared the fix.
  • Active Personal Developer Instances have nearly doubled in just 8 months, so we’re adding more capacity.
  • Developer videos are watched almost 3 million times a month, and we’re investing in even more video content.
  • User groups are running in 29 countries.

 

His message to the Community was simple: this platform exists because of you.

 


 

Demos 1 & 2: The P1 that went to the wrong team

 

Picture it: 4:47 on a Friday. Online checkout is down. Customers are literally trying to give you money, and you can’t take it. 

 

Sujan and Laszlo started with a confession. Not everyone is ready for AI offerings or has access to them yet. So this demo went back to fundamentals: workflow debt and data debt.

 

 

Part 1: The workflow debt

 

Monitoring raised a P1 for checkout-app-03, but it landed with the Service Desk instead of the team that supports the server. The weird part? The flow run said complete, and every step was green.

 

When they dug in, they found the culprits:

 

  • The update step grabbed the wrong data pill. It pulled the assignment group from a Service Desk lookup instead of the CI’s support group.
  • The flow had an If, an Else If, an Else, seven lookups, a script stuffed into a Log step, and no error handler. Every admin had added a bit over the years, and nobody had ever cleaned it up.

 

Laszlo’s point: dragging in the right pill only fixes the symptom. The messy design is why someone grabbed the wrong pill in the first place. So he rebuilt it with two reusable subflows:

 

  • Resolve CI replaces all the CI lookups with exactly one. It takes a host name and returns the server, matching only operational servers with an exact name match.
  • Route P1 was packaged straight from the existing actions and pills using Workflow Studio. If the CI has a support group, the P1 goes there. If not, it falls back to the Service Desk. (As Laszlo put it, the bug just became the fallback.)

 

Then they tested it, and the P1 still went to the Service Desk.

 

261105-NowDev26-JM1_3245.JPG

 

 

Part 2: The data debt

 

The fallback did its job, which meant the flow found a checkout-app-03 with no support group. Sujan explained that when the right name returns the wrong record, you’re dealing with a duplicate CI, one of the most common CMDB problems out there.

 

A quick search in Service Graph Workspace turned up two checkout-app-03s: the real Linux server, and its “evil twin.” On the map, the twin had no relationships and exactly one active incident: our P1.

 

So, do you just delete the twin? Sujan asked the room, then answered for them: no, no, no. Delete it, and every ticket pointing at it suddenly points at nothing. Instead:

 

  • Use the de-duplication task and the Duplicate CI Remediator to merge the twin’s attributes and related items into the real CI.
  • Retire the duplicate and label it, so Resolve CI never matches it again.
  • At scale, use de-duplication templates. If an integration created one twin, it probably created hundreds. Make the decisions once, then run the template on demand or on a schedule.

 

After the merge, they ran the same test on the same P1, and it routed to Commerce Platform Support. They didn’t touch the flow this time. All they fixed was the data.

 

The takeaway: Good data makes good workflows possible. Good workflows are the ones you can trust to automate more. And all of that is the foundation agents need. In Laszlo’s words, give an agent a CMDB with two checkout-app-03s and it’ll pick one, just like the flow did. “It just sounds a lot more sure about it.”

 


 

Demo 3: The app nobody will touch

 

Kristy opened with a question a lot of us can relate to: who has migrated a process off a legacy tool and rebuilt it exactly the way it used to work? 

 

Custom tables, custom scripts, custom approvals. It mostly works, but the team that built it is gone, and nobody wants to touch it because nobody knows what might break. Then the business asks for a change to a core approval workflow.

 

 

Step 1: Understand it with Build Agent

 

Patrick had never seen the app, so he asked Build Agent what he’d ask the original developer: How does this app work, and how is it different from out-of-box Change Management?

 

Build Agent researched the actual instance (tables, business rules, script includes, flows), not a three-year-old doc. It came back with:

 

  • An overview of how the app fits together
  • Everywhere the app rebuilds something the platform already does
  • Security risks in the implementation
  • The blast radius of changing the save process: every table, record producer, and script connected to it

 

Ten minutes in, Patrick had enough context to actually get to work.

 

 

Step 2: Modernize it with Autonomous Engineer

 

Patching one script would work, but the Change team really wanted change models, approval policies, and a modern UI. That’s where Autonomous Engineer comes in. Build Agent handles one task at a time. Autonomous Engineer takes on the whole job.

 

Here’s how Patrick ran it:

 

  1. Set the rules. In Build Agent’s Rules tab, you define custom skills and rules (naming conventions, commenting guidelines, footguns to avoid) that shape every session.
  2. Drop in a PRD. Autonomous Engineer reads it, explores the instance, and asks clarifying questions like any good engineer would.
  3. Review the plan. It splits the job into Work Items (think user stories for the agent), each with a description, acceptance criteria, and prerequisites.
  4. Let it run. It works in the background for hours, across as many sessions as it takes.
  5. Review and accept. Each deliverable comes with acceptance criteria, ATF tests it generated and ran, and a tidy update set. Your approval unblocks the next task.

 

Patrick’s point: it took on a massive pile of work, but he stayed in control the whole way.

 

261105-NowDev26-EL1_7630.JPG

 

 

Step 3: Build it anywhere

 

Kristy asked the question a lot of you were probably thinking: is this a Studio-only party? 

 

Nope. Patrick opened a Fluent project in VS Code with the Claude Code extension and the ServiceNow SDK underneath, plus a Jira MCP connection to pull in the PRD and user stories.

 

  • A custom skill pulled down the team’s Build Agent rules first, so you write your rules once and they follow you wherever you build.
  • While working on the approval story, Claude caught a real bug: changes that needed CAB approval could close with no approval at all. The rule only added a work note instead of blocking the save. 
  • Patrick moved quickly through low-risk recommendations, but slowed way down for anything that touched the instance: readiness checks, rollback info, manual steps.

 

Built, installed, and ready to test, and Patrick never left his editor.

 

The takeaway: Whether you work in Studio or your own tools, you get the same skills, the same rules, and the same outcome.

 


 

Demos 4 & 5: The request that keeps coming back

 

Ajay set it up: Build Agent is great at building ServiceNow apps. ToolSmith specializes in building AI assistants for your employees, and it’s trained on over 75,000 enterprise APIs, so your agents can take action against pretty much any system.

 

 

Part 1: Build the capability with ToolSmith

 

HR wants employee referrals to be easy. Sounds simple, until you find out recruiting uses different systems for technical and non-technical roles, and bonuses change based on role level. Suddenly it’s weeks of API research, security reviews, logic tables, and agent design. Into the backlog it goes. 

 

Kevin had an HR Assistant that could answer questions about the referral program, but HR wanted it to act. So he pasted the talent team’s feature request into ToolSmith, which:

 

  • Researched the API docs from its action catalog and vendor documentation
  • Checked existing connectors and automations, so you don’t end up with your thousandth duplicate connection
  • Explained its design decisions and wrote a plan in plain language, so you can catch misunderstandings before anything gets built
  • Asked for credentials through a secure form, never in the chat, with connector details already filled in

 

In the test, Kevin gave the assistant a LinkedIn link instead of typing details. It pulled live open roles, filled in the candidate info, and asked him to confirm before submitting.

 

 

Part 2: Get it in front of every employee

 

Next, HR wanted to invite every employee to submit a referral after their first 30 days. Easy, until you ask where everyone works:

 

  • The acquired company is still on Slack (Block Kit + DM)
  • Everyone else is on Teams (Adaptive Card)
  • Store teams don’t have laptops (text message)
  • Everyone else gets an HTML email
  • Plus a push notification, timed for 7 AM local, because nobody wants a referral ask at 2 AM

 

That’s 23 steps, 6 integrations, and 5 message formats for a single message. With EmployeeWorks, it’s one step: Notify. Look up the employee, write the message once, and add a CTA that launches the referral experience. The assistant handles delivery end to end.

 

And that same experience goes wherever your employees are:

 

  • The EmployeeWorks web app, Slack, and Teams
  • An Embedded Assistant snippet for your intranet
  • The ServiceNow Mobile App for teams on the go
  • The Conversations API for other channels, even WhatsApp

 

The takeaway: That request sitting in your backlog? Build it in minutes, not days.

 

261105-NowDev26-EL1_8035.JPG

 


 

Where to go from here

 

There’s probably something in your backlog you’ve been putting off. A workflow that technically works but you don’t trust. An app nobody wants to touch. Something you’ve built three different times.

 

Pick one. When you get back to work, take another look at it. You might have a much better way to handle it now.

 

Plus, check out these other resources when you have the time:

 

 

Now we want to hear from you: what’s the thing behind your wall? A gnarly flow, a duplicate-CI horror story, a backlog item that won’t die? Tell us in the comments.