Description
In this rapid-fire talk from Dreamforce 2025, discover 10 ways Gearset simplifies Agentforce DevOps, from versioning and testing to visualizations and intelligent analysis — all while keeping production stable.
Speakers:
- Eamonn Boyle, VP Engineering
- Javier Suquia, Software Engineer
Transcript
If you're watching this talk and guessing AgentForce is something that's important to you, this talk is part of a trilogy presented at the AgentForce Pavilion at Dreamforce twenty twenty five. And in it, we're gonna talk about the practical real world challenges of adopting AgentForce and how a DevOps mindset and some great tooling can really help.
My name is Eamon Boyle, and I'm the VP of engineering at Gearset. Hopefully, you've all heard of Gearset, but if not, we're the leading Salesforce DevOps platform.
I'm a software engineer with background working on many different stacks and platforms with about twenty years of experience.
And hi. I'm Hayao Sukia. I'm a product manager at Gearset working on our agent for solution. I also come from a technical background, though I haven't got as many gray hairs as Eamon yet.
As an engineer, talking to customers, understanding their problems, and finding solutions was the most rewarding part, so I made the transition to product management.
We've got a lot of material to cover, and I'm originally from Spain and Eamon from Ireland, so we've got a tendency to talk fast. Thankfully, this is being recorded, so it shouldn't be much of an issue.
Here's what we have planned for this session. We'll start by exploring what makes AgentForce unique, then we'll discuss why Gearset is the ideal partner for your AgentForce and Salesforce DevOps journey. We'll dedicate the bulk of our time to offering practical advice and demonstrating how Gearset can help, and then we'll end with a quick wrap up.
Are you pushing agents to production yet?
As our CEO, Kevin, said in the first AgentForce Pavilion talk, AgentForce is a leap forward that changes how we build, how we deliver, and what our customers can ultimately achieve.
Salesforce themselves have described as the future of the platform. It's exciting and energizing and all that good stuff. But it's also disruptive and challenging because every leap forward comes with obstacles. The scale of the paradigm shifts and agent forces unique challenges such as nondeterminism, data reliance, and new testing requirements, means that we have to be at the top of our game to succeed. And this means having robust DevOps practices, processes, tooling so that our execution meets our ambitions.
So why listen to us? We're not salespeople. We're not here to give you a really hard sell, but Curacet is fundamentally a product and engineering company. We really do believe in this stuff.
Because of this, we focus deeply on the challenges that Salesforce teams have, helping customers achieve their outcomes. That's what gets us out of bed in the morning. We support more organizations than anyone else in the ecosystem, over three thousand teams. And because of this, we learn and deliver more than anyone else in the ecosystem.
We also practice what we preach, releasing our service multiple times per day, which is especially important in a fast evolving agentic world.
And this is demonstrated by the trust that our customers place in us. Our solution is so flexible and powerful that we cover all segments and all industries. We can meet you and your teams where you are on your DevOps journey and gradually help elevate you to that next level.
We work with small teams who are early in their Salesforce DevOps journey through to huge corporations with the largest, most complex Salesforce teams. Ten percent of Fortune five hundred companies use Gearset, including names such as Amazon and McKesson. We understand the full gamut of customer problems and build solutions for them. Okay. Maybe a smidge of a sell in there, but folks like evidence. Right?
So what do we think is the secret to success?
While AgentForce does present us with some unique challenges, fundamentally, crux of solution is well established. There are a wealth of learnings taking from the broader software engineering discipline that we can apply, techniques and practices that make for a successful delivery team.
At Gearset, we talk a lot about DevOps on right, those practices and processes that you should adapt in order to be successful, covering the complete DevOps life cycle.
In this talk, we'll cover ten pieces of advice for success. This is not an exhaustive list. This talk's only twenty minutes. But we'll go through them and along the way show how Gearset helps and guide you through a bush tooling. We've grouped them under three themes, having a solid foundation, building a smooth delivery process, and leveraging automatic automated protections to ensure your foundation doesn't crumble. And remember, this is not new knowledge. This is well established battle tested DevOps practice.
Alright. So let's start by looking at solid foundations. I'm sure we've all had an agent or other AI hallucinate in the past. You can't build a house on shaky foundations, and this is true for software development too.
Work health and addressing clinical debt is a really important practice generally, but this is especially true for agent force development. For example, do all your custom elements have consistent naming and good descriptions? All those objects, validation rules, flows, this will impact the effectiveness of your agents.
Is your data in a good state? Garbage in, garbage out applies to agents too. Are you using standardized pick lists? Do you have validation rules enforcing the integrity? This will impact the effectiveness of your agents.
Do your teams understand the org and its structure so they can reason about it to maintain the quality? Because this is a continuous process. Our jobs are tough, but we can make them easier by building on solid foundations.
From a solid foundation point of view, we should consider understanding the structure of your orgs to take corrective action and reduce technical debt and using representative data.
Alright. So let's look at number one, understanding the problem. It's important to always start by understanding that problem that you're trying to solve. Figure out what the job to be done is and only then work towards solution. But a key part to them building that solution is that you need to understand the state of your org to define what changes are required to deliver the solution.
Gearset's org intelligence feature gives you a bird's eye view of your org's health, its structure, and metadata.
So let's say you want to know more about what your new agent does and what its dependencies are. With Org Intelligence agent, you can easily ask that question. This will scan across your entire org and help you understand what the agent does.
And if you want to improve the agent, our dependency analysis, also visualization, and item explanation will help you figure out what changes are required and plan how to implement the new features.
Number two, using representative data.
Agents require good data for development and testing. You need representative data to ensure that your agents perform accurately. Manually creating test data is time consuming and not always representative. Your real data would be ideal, but your customer's data privacy and security is nonnegotiable. So how do we resolve this dilemma?
With Gearset, we provide you with a sandbox seeding and data deployment solution that allows you to easily deploy representative data to your sandbox and development orgs with the ability to automatically mask out that sensitive customer data such as PII. We also offer in place masking so that you can protect that sensitive data after refreshes.
Okay. Let's talk about smooth delivery. Do you ever have to carry out manual steps when deploying agents?
We're not making one change and then stopping job done, retire, let's go to the pub. We are in a continuous pipeline of increment delivery, new features, bug fixes, improvements, and we are in the middle of an AI and agentic revolution. This is a new paradigm for most people, and things are moving at breakneck speed.
As you move from building simple chatbots through to fully autonomous agents, you want the delivery process to be robust and repeatable and just get out of your way. As new versions are dropped that contain new or breaking features, you need to have confidence in your release processes to get those updates out quickly. That's how you're gonna succeed. That's how you're gonna go fast and deliver value again and again and again. By automating the boring stuff and letting the tools do the heavy lifting, building a repeatable process, that's how you move fast. And it'll free you up to concentrate on the more difficult stuff, the stuff of actually creating solutions.
We've got a bunch of things related to this, so let's just get into them.
Number three, understand the metadata.
Metadata is the cornerstone of your orgs, but come on, there's a lot of it.
You need context to be able to track your changes, all those prompt templates, topics, actions, flows, Apex, etcetera.
At Gearset, we have numerous visualizers for metadata, including the table views like you're seeing here for agent force actions. You can easily see the internal structure and navigate through it with ease.
And what about flows you've attached to your agents? Like, have you ever looked at the XML representing a flow? It's absolutely crazy. Right?
We've got you covered with our flow visualizer navigator. You can examine the changes that you're deploying in the way that you built them. You can see the differences right there in the flow, navigate and drill into the structure so you have all of the contacts to make the right decisions.
This is all awesome, but agent forces evolving at breakneck speeds, so it can be difficult to keep track and stay on top of the changes across API versions. Gearset has you covered here. We're leasing twice a day. We are able to stay on top of this for you.
Alright. Let's look at understanding the dependencies.
Another challenge with all this metadata is the dependencies. Each functional change you build touches many different metadata types and to successfully deploy your agent force changes, you need to ensure that you have all the relevant dependencies in place. Your agents to actions, topics, flows, Apex, and prompt templates. Gearset makes reasoning about this easy by allowing you to drill into the dependency tree to ensure you have all the relevant pieces in place for the deployment.
Here you can see us drilling in from the planner all the way down to the Apex class, ensuring that we are capturing everything to make this agent work.
Number five. This one's bit of a mouthful. Be careful with profiles and permissions.
Profiles and permissions are especially difficult due to their size and complexity. With agents, you need to create a bunch of new permission sets specifically for your agents to ensure that they have the right permissions, and we give you a bunch of tools to help here.
As you can see, we provide a combined profile view that allows you easily drill into permissions that have changed, visualizers to help you see those differences easily, and precision deployment for selecting only the changes that you want to deploy.
You can also select permissions within a specific object easily too just like other dependencies.
Number six, review your deployment.
Okay. So you've selected your changes and you've got your deployment all ready to go. It's still easy to have missed something or included something invalid. So it's always a good idea to give the complete deployment that last once over. Gearset has over eighty problem analyzers automatically scanning your deployment for issues with helpful auto fixes to streamline making your deployments a success.
Here, you can see the analyzer has identified that we've missed a dependency with our agent, which as we all know will cause the deployment to fail. The analyzer has identified the issue and has an automatic fix, which we can easily accept to add the missing dependency so our deployment succeeds. As you can see here, Gearshed has added these missing dependencies to our deployment package, which will ensure a success of validation and deployment.
In fact, we've brought some of these analyzers forward during the comparison phase, shifting left to intercept issues even earlier.
Number seven, watch out for configuration.
Not everything to do with successful agent force deployment is about deploying metadata. There's also configuration to consider such as version selection for your agents and flows, activation and deactivation, and others that you need to control in order to have your agents work properly.
Doing this manually is just more friction on recruiting a repeatable deployment pipeline. That smooth delivery process we want. Luckily, Gearset's user friendly interface and intelligent deployment can make this configuration easy.
You can easily select which version of an agent you wanna deploy, and coming soon, we'll have automatic deactivating and activating agents throughout the deployment process, which means less manual steps for you and your teams. And you can see here those early problem analyzers suggesting suggestions, keeping you correct as version change. It's these sorts of features that allow our users to experience a ninety nine percent successful deployment rate.
Have you ever released a bug to production that needed immediate rollback? We've all done this. Right?
When our solution makes it to production, it's a different ballgame. We need to understand how our system and our agents are behaving. Your systems are only getting bigger, and you need that automation in place to alert you when things are not behaving as they should. You don't want your users alerting you to the issues. You want to see when that flow fails, who it failed for, tracing it back to a change, and undoing that change if necessary. All of this feeds back into protecting that solid foundation and org health that we established is so important.
So let's take it home with our three protection concepts, quality, testing, and observability.
Let's talk about maintaining a high quality bar.
The natural place we want to get to is putting all of the things we've talked about together into a repeatable pipeline. This is a place to coordinate across multiple deployment streams, multiple environments, and even multiple geographies.
This pipeline allows you to deliver at a rapid but sustainable pace.
I'm not going to get into pipelines in detail. That could be a couple talks in itself, but something to reflect on is that with all of this ease of delivery and automation, you want to make sure you have the right guardrails in place to protect your environment.
We've already talked about the problem analyzers protecting you, but we also have protection around review gates, merge conflicts, post deployment steps and checkpoints, testing, and code coverage gates, and of course, our code review tooling to scan your code for security and quality issues.
With large dynamic teams and with technology that is moving as fast as AgentForce is, it can be hard to maintain quality. You're an early adopter and you've got your agent in production, but suddenly something changes in the latest release.
New configuration is added or something gets deprecated or that new member to the team just doesn't have all the knowledge yet. Gearset's code review scans leverage an intelligent understanding of the complete Salesforce platform, not just Apex code, but your metadata, your flows, more to better reason about your entire system and flag issues before they make it to production.
We have over a hundred rules including many agent for specific rules. Here, you can see that the scan has detected several issues with this latest PR.
From simple best practices, like having good descriptions, to having the correct configuration, to important security best practices, such as having an agent verify the identity of a user.
You'll see as well that for many of the code review issues, there's an auto fix option, such as removing deprecated features or setting the correct configuration.
Code reviews also fits back into our foundation as there's a whole host of team auditing and guidance on how to incrementally improve quality of your code over time and reduce that technical debt.
Number nine, harness testing.
Testing has never been more important. The nondeterministic nature of agents makes testing a crucial component of agent force development. Testing helps to identify and fix bugs, vulnerabilities, and performance issues, and it's essential to ensure the reliability, security, and effectiveness of your agents and the overall org health for that matter.
DevOps testing center is therefore a critical weapon in your arsenal when deploy developing agents. Gearset seamlessly integrates with testing center, triggering test runs for agents and pipeline workflows and showing test center results directly within Gearset.
And since we have a complete end to end solution, each part of the process has interconnections. Here, you can see that we navigated from the test results through to the deployment that triggered it. We also integrate with a host of third party test solutions directly, which could surface right in your pipeline. Our colleague Lawrence Boyce talks about this in the third talk of this trilogy, so I'll not steal his thunder too much, but I recommend that you go and watch that talk. He'll show how you can leverage UI testing with agents, showing off Gearset's own AI powered UI testing solution.
And finally, leveraging observability. Observability is another crucial component in a robust DevOps process. Understanding the state of your system allows you and your teams to proactively identify and solve problems, understand system behavior in real time, and make data driven decisions to improve performance and reliability. Gearset has a rich observability solution allowing you to view in real time your errors for flows and Apex code, visualize where in the flow errors occurred, view trends over time, trace back to users, get insights and alerts around governor limits to understand when you have errant Apex or flows eating into your budgets, link to work items in Jira and Azure DevOps, and much more.
And of course, we also have our popular change monitoring solution so you can track changes. Alright. So I may have lied slightly. There's an sneaky bonus one and an important one too, backup and restore.
If the worst does happen in production with a robust backup and restore features, you know that you'll be able to get back to a good state in no time at all.
Okay. Let's wrap up.
We've talked about how AgentForce is a powerful and rapidly evolving tool that can unlock so much for you and your teams, but it comes with its own set of challenges. The advice we've laid out along with some great tooling from the gear set DevOps platform can help you overcome those challenges.
Here, you can see the complete list. And as you can see, this spans the entire DevOps process.
I hope you enjoyed the talk and the other talks in the series. If you haven't seen them, I really recommend you go and watch one of the many talks Gearset presented at Dreamforce twenty twenty five. And thanks if you're at Dreamforce and came to see us at one of our booths or came to one of our talks in person. I hope you find Dreamforce as valuable as we did. You can scan this QR code for more information on Gearset and AgentForce.
We always love talking to customers. So if you're working with AgentForce and would like to talk, please reach out in the in app chat. Or if you're not a customer but having difficulties with AgentForce, then also please reach out to see if Gearset can help. Or simply sign up for our free trial. It takes five minutes, is noncommit, and will give you everything that you need for thirty days to evaluate the platform.
