More output, same control: How Salesforce leaders embrace AI

Share with

Description

Most Salesforce teams know AI is coming for their workflow — but knowing it’s coming and knowing what to do about it are very different things.

In this webinar, Jack McCurdy, DevOps Advocate at Gearset, draws on seven years of conversations with Salesforce teams to cut through the noise and share what the most successful teams are actually doing to put AI to work — without losing control of what gets deployed.

What you’ll learn:

  • Why AI increases the volume of changes moving through your pipeline — and why your existing delivery mechanisms need to be solid before you accelerate, not after.
  • How the four pillars of DevOps done right — automated testing, code reviews, pipelines, and observability — become the non-negotiable foundation for safe AI adoption.
  • Why vertical AI solutions built specifically for Salesforce outperform general-purpose tools like Copilot or Gemini, and how to apply them to well-defined, bounded tasks rather than sprawling ambiguous ones.
  • What “what is measured is managed” actually means for AI: how to set benchmarks, track real outcomes, and tell the difference between feeling faster and genuinely moving faster.
  • Why the human role gets more important — not less — as agents handle more execution, and what business context, stakeholder judgment, and interpretive skill agents can’t replace.

Learn more:

Relevant videos:

Transcript

Alright. We're gonna get going here, folks. We've got quite a quite a bit to cover today. It's an absolute pleasure to be here.

And more than anything, thank you so much for choosing to spend half an hour of your time today, with me to talk about a subject that is very much at the top of everybody's mind, at the top of their agendas, recently. My name is Jack McCurdy. I am a DevOps advocate here at Gearset. I've spent the last seven years with Gearset speaking to Salesforce teams of all shapes and sizes about their Salesforce release processes, how their teams work, especially about how the people in their teams work and how they like to engage with others around them, and really ultimately how they get the most out of their Salesforce implementation and the most out of their teams.

I run a podcast called DevOps Diaries. You can check that out on the website, Spotify, YouTube, all those fun places as well.

So that is a little bit about me. Obviously, an Eagles fan and a big hockey fan as well if you were here at the start of the call.

You might see this at many of the Salesforce conferences that that you have attended, especially Salesforce Salesforce conferences. This is a session that I presented a couple of times already this year at Salesforce conferences around the world, and you'll probably be familiar with a forward looking statement. And please make all purchasing decisions based on generally available products. But I think forward looking statements more than anything are quite a poignant place to start because I don't think we've ever been more forward looking than we have now than when it comes to AI and the journey that we're already on.

We're already looking forward, which in some sense is great, but I think more than ever, we need to look at what is happening right now in our teams and in our orgs, before we can get into that future.

But to address that future a little bit from right now, and the question that I ask myself the most regularly when I think about AI is what's kind of the opportunity that we're faced with? And one of those opportunities is I think more than ever, we have a decision on who we want to be and the kind of teams that we want to be. Do we want to be configurators and spend most of our time in Salesforce setup or the CLI, or do we wanna be architects of change and progress or and push boundaries and create impact and that has really not been possible before.

And I think we have a real opportunity to remove some of the shackles that have constrained us in the ecosystem for some time now and push us forward into a new era and deliver better outcomes for our customers. I at least hope that is a shared, shared desire for the people on this webinar right now or that are maybe listening back to it.

There's a future that I envisage where we get to work on Monday mornings where, the Sunday scaries have already happened. The dying embers of your weekend have been spent, thinking about, you know, the backlog of your Jira tickets or what work awaits you on Monday morning. And, normally, on that Monday morning, you spend time just scrambling, that first chunk of your day or the whole day figuring out what needs to happen in the rest of the week before you even get started.

But the vision I have for Monday morning and a vision that I think is possible is by the time you open your laptop on Monday morning, you've got a workload that is planned. You have an agent that has helped you pull together everything that's need to be happened, flagged areas for you that are most likely to be affected, whilst cross referencing a backlog, or maybe even drafted first pass requirements for you to look at so that you can actually spend your morning doing analysis and making decisions, judgment calls that only you can make. It's just that the prep work's already happened. An agent's already done it for us. Some of you may be working in that way now even.

But this month version of Monday morning that I'm envisioning, like, it comes down to you, and it comes down to your focus. You can get started on the week, and your team can get started on the week delivering that value. And that future of Salesforce development isn't just a pipe dream or a vision that I think is possible for everybody. It's already the way that the Salesforce team here at Gearset are working.

More importantly, it's not that agent and that journey and that place that we're all looking to get to. It's it's you and a virtual teammate working hand in hand. You know? The an agent that is handling well defined, repeatable, and time consuming groundwork, taking care of all the low hanging fruit in that backlog so you can focus on judgment calls, on context, on strategy, relationships, and the stuff that actually needs us as humans.

And this vision, of course, isn't about replacing you and your skills. It's just refocusing where and how you spend your time and ultimately being able to deliver all that value. It just means that you might not have to build the things anymore, and I think that's a good thing.

And I think the elephant in the room for a lot of this thing is that not everyone's an optimist, and teams are somewhat of a on a spectrum when it comes to this journey in AI. And when I sit here talking about AI and that journey, agents reshaping what it is to work in Salesforce and this ecosystem, like, honestly, what's your gut reaction, and how does that make you feel?

Really interestingly,

we found that in the upcoming state of Salesforce DevOps report, which will be coming out very shortly, this is a survey that we run every year of Salesforce professionals across ecosystem. It tells an interesting story and that most people aren't at the extremes of AI. So you have the optimist where the optimist where people think agents are gonna do absolutely everything. We're already in a new era.

The world is forever and completely turned on its head and completely changed. And the other side are the skeptics, the people who think AI is just hype, that the tech isn't there, and that their jobs are pretty much the same, and nothing's really changed thus far. But the majority of those people, seventy six percent of those, they're actually somewhere in the middle. So a bit uncertain, just curious, maybe a bit anxious, and watching and waiting to see what is gonna happen.

And I find that fairly resonant with my experiences, especially in the last three months or so of being out and about at various conferences in the community. And I wanna say that that is okay for Nine. You don't need to be an unmitigated enthusiast for everything that is to come or what teams are doing right now and some of the successes that teams are seeing with AI right now. You're allowed to have questions.

You're allowed to be a little bit unsure of your role and how you feel about that change.

But the time for kinda wait and see is over. So it's great that you're either watching this back or you're here live with me right now and looking to see what you can do to make whatever future awaits us a reality. And I wanna give you something concrete you can take away to embrace that future, and that's ultimately why you're all here. We should invest in our AI strategy now.

LLMs are getting really, really quite good. So I won't go deep into the technology and why. It's a whole other talk. Probably better suited to folks much more intelligent than myself and understand greater understanding of how they work.

But they have got genuinely good at a class of tasks that we previously assumed required human judgment. So if we think about synthesizing information, generating structured outputs from unstructured inputs or data, and following complex multistep instructions and being able to do that all relatively quickly and consistently, Those tasks, I think, matter, and that context matters for our world and the Salesforce ecosystem because so much of what we build and maintain is fundamentally about that structured process, the structured data models, the automation logic, the configuration, and things like that.

Those things in the right framing AI can help with. And for the skeptics that might be in the room, a lot of AI demos do look impressive in a vacuum and then fall apart in production. We've seen that.

Concerns about quality, hallucinations, context limits, all of those are definitely legitimate. But I think dismissing a whole wave of technology because of that early friction is definitely a mistake in this trajectory that we're on is very much real.

The capabilities have materially improved, so that window that I mentioned earlier, wait and see, most definitely closing. And for the optimists of you that are in the room, I completely love and appreciate that energy, but there is some realism that needs to be taken from here. Agents aren't fully autonomous yet, and if they are, then I would really like to talk to you. Please contact me afterwards to hear how you're doing it and what you're what how you're going about that.

But these agents, they make mistakes. They need guardrails. They need somebody accountable for the outputs, and you need to deploy AI with guardrails and infrastructure around it. Otherwise, it's not a shortcut and a value increase or it's gonna actually become a liability.

So the truth is actually somewhere in between, which is what I'm gonna get into here.

So the opportunity is basically comes down to this. So we with AI, we'll be able to do more with the same people. We'll be able to move faster and ultimately drive better business impact and results for the businesses and for ourselves in this industry that we work in.

On the flip side of that, the challenge that we have is our delivery mechanisms that support this increased velocity. We need some strategy, and we need some oversight into what we're doing. And there's three key components that I think that are gonna turn, that are already turning the most successful Salesforce teams, and Salesforce teams actually making great use of AI right now, turn it in from an opportunity into something sustainable that delivers real value. And those three areas are DevOps done right, the deliberate use of vertical AI, and, obviously, the human role, one of my favorite things to talk about, in this ecosystem overall.

So the practical bits. Let's talk about DevOps done right. So I've been in the Salesforce ecosystem for seven years, and all of that time, have dedicated to people, process, and technology that supports DevOps. And DevOps has always been important. You know? It encompasses every facet of every change that you deliver on the Salesforce platform.

And if you're going to increase the velocity at which changes move through your org, and AI is probably gonna push you in that direction whether you like it or not, then your delivery mechanisms need to be rock solid. Your manual processes that you might have right now don't scale.

And if you think about where your bottlenecks are today, so deployment nights, you know, a release manager maybe with production access, clicking through change sets at ten PM on a Friday, mapping information to spreadsheets to complete the tickets, testing, probably mostly vibes, caffeinated, a few manual spot checks here and there, and those code reviews that are inconsistent because there aren't enough people there and always time pressure, you know, those bottlenecks are real right now. If you imagine doubling the number of changes coming through that pipeline because of the velocity gains that AI gives you and the throughput that AI gives you, like, what's gonna break first? And the answer in pretty much every org is gonna be everything, including your brain simultaneously. So what we need is oversight and visibility at every stage, which is DevOps done right.

The four key here is four keys here. First one is testing. So we need high levels of automated test coverage so that you actually have confidence when you promote a change. And we're not talking just about the unit tests that are written for your Apex class. We need integration tests, integration tests, regression tests, UI tests, the full suite. If we can test to a high level of automation and make sure that everything that we're pushing through is thoroughly tested and validated, we're gonna succeed. And automation is the key here as well because we cannot be, we just don't have enough of us to be able to do this, on a manual basis every single time.

Closely joined to testing is code reviews. So automated analysis that catches common issues before a human even looks at it. So AI assisted review is totally okay here. Static code analysis, security scanning, and that security being baked into your pipeline and not bolted on at the end. Code views that have old context of well architected Salesforce applications as well. So, not just does this code potentially have a bug or an error, is this part of well architected practices for Salesforce?

We need pipelines as well. So automated pipelines that can move the changes through our environments consistently without somebody knowing the right sequence of manual steps. That repeatability in process gives you oversight into when things aren't working. You know, those guardrails that you set up really give you that sense of security.

They keep you safe, and they also keep you accountable. So everything that we're producing follows that same path, and we can look at the exceptions that we need to be looking at rather than every single change that goes through. And we decide what those guardrails are too. This isn't the AI saying this is our guardrail depending on your business, compliance, things like that.

We can set those up as we need it for our business.

And finally, we need to back up an observability because things, whether we like it or not, they can and they do go wrong. And there's always a horror story about broken integration, a rogue user, and we need to know when these things happen in our org as soon as they happen, not three or six months later.

You know, that covers your metadata, your data. We need to know if we're hitting governor limits and that error monitoring when maybe if your flows or Apex fails, we need to make sure that we can fix those as and when they happen, not when a user potentially tells us about. And the phrase I always come back to when I think about this is automating the oversight. So don't just automate the work because in a world where AI is generating more of the work, humans need to be looking at those signals and the exceptions, not doing every single manual check themselves. So we look at the things that really need our skill and need our judgment. So that's the real foundation before we are able to adopt AI effectively. Everything else becomes possible with this foundation.

Okay. Part number two. So if we have the delivery infrastructure sorted, we need to talk about where AI actually goes in the workflow. Most of the common mistakes that are happening in the Salesforce ecosystem right now that I see is people are treating AI like a magic go faster button. You know, someone spins up Claude or Copilot and developers start using it, app and start using it, and then victory, declaring that they're thirty percent faster, for example. Well, my challenge to that is how do you know?

Did you measure it? Did you measure what happened to the quality? What happened to the amount of times you might have to rework something? What happened to the number of instance in production? A lot of teams, they're finding that they think they're getting faster just because it feels faster, and the AI is able give them the output that they might have spent quite a bit of time on, the code appears more quickly, but that's not the same thing as actually being able to move faster.

The other point of note about vertical AI as well, it's tempting or it might even be forced through your company mandates to use general purpose tooling like Claude, Gemini, Codex, whatever it might be. However, studies have already found that vertical AI solutions and AI tools that have been built with specific tasks and industries in mind perform better and see ROI faster, than just that general purpose tooling. So we need to apply that, vertical AI in the, in the best way possible.

And in practice, this looks like this.

So when we're adopting AI for Salesforce, we need to look at the small, well defined slices of work that we can give it. So if you give an agent a sprawling ambiguous problem, you'll probably get a small ambiguous output. But if you give it a specific well bounded task that we understand, summarize these five user stories, or draft this Apex class with specific input and output, for example, it can do really well.

We can use it to, synthesize context. So your org has years of configuration history, field descriptions, automation logic, and a human can't hold that all in their head all at once. You may be working in an org that's very lucky to have great documentation. But if you don't, your org is likely a minefield. So org intelligence tooling is a real force multiplier when it comes to AI in the Salesforce space and definitely the area of AI use that I see most across Salesforce teams and the Salesforce professionals that I talk to. And this drastically reduces the amount of time your team spends in the discovery phase of any project or the plan stage of any project as well.

And as I alluded to just a moment ago, what is measured is managed. What is measured is managed. So we understand our benchmarks, and we can map our performance to those benchmarks. So for example, AI is good at generating tests. Your test coverage should be improving, or your regression rate's decreasing. Those measurable outputs, your executives, your teams, ultimately, you you can continue to get behind.

If you are able to release on a more quick cadence, if, your releases take less time, to run through all these things, you spend less time in debugging stage or your production releases doesn't need a two or three day window after that to start releasing again, then you know it's gonna be a success. So set your goals and make sure that it actually happened.

And then if we do all of this and we do get this right, we should then be pushing boundaries. So I mentioned a lot of folks like to push those boundaries early, those sprawling ambiguous tasks. If we do all the other things first, if we understand the small slices that we can do, we should then be able to push the boundaries of what is possible. What LLMs can do now is worlds apart from what LLMs could do, even a year ago, six months ago, eighteen months ago, should pick your time frame.

They get better and better and better. And the more comfortable you get with using it, you can then start to push the boundaries. So give yourself that ambition and give yourself that freedom to push it once you've got familiar with what you were doing here. The discipline here really is about being honest about what you expect AI to deliver before you start implementing it and checking whether it happens, not just whether it feels like it's helping you.

And third, the human role, one of my favorite areas to talk about.

It's one of my favorite areas to talk about, but I think it's also the area that people find hardest to talk about. And I think it's the most important conversation in any room.

So if the agent's doing more of the execution, the more of the building, what does we do? What do we do? What do the humans do in this process? And the answer isn't just manage tickets.

It isn't nothing. It isn't less. And the human role in an agentic world is more important, not less important than ever.

Somebody has to be accountable for the output, and an agent doesn't own an outcome. A team lead does, or the admin that has built the work owns the output. Your product owners do, an architect does, developers do. A well configured agent might be able to build something for you, but a human has to stand behind that thing and say, this is the right thing for us.

This is safe. This is right for our org, and this can go to production. This adds value to our business. And what that means practically is that the people who thrive in this new area and the shift that we're going through are the ones that really understand what they bring that agents can't.

And for most folks, that's gonna be business context. So you understand why the CRM in your organization actually exists. You know why specific fields exist, why things were built the way it was, who screamed about it in twenty nineteen, and what the downstream impact of changing it would be. That context is gonna be gonna be crucial, in this new world.

We need great stakeholder relationships. So you know what the sales vice president actually cares about even when they are can't articulate it. You know how to navigate a conversation between that sales VP and a customer success VP about the trade offs that might be required in your org.

Relationships really do take ideas and turn them into reality. And, also, those relationships generate buy in into projects and initiatives and involve multiple people at multiple levels. Being able to navigate that landscape complexities of any corporation or organization, is gonna be essential as we look to carry on creating the value that our organizations are looking to get out of the technology solutions that we build.

I mentioned this a couple of times already, judgment, especially under ambiguity. So when requirements are unclear, when two business units conflict, when something smells off and you just can't quite put your finger on it, that's gonna be you. That's not gonna be an agent. Something either even as simple as, you know, we should do it this way, overdoing it that way. AI can execute the instructions that you give it, but you need to make the judgment call and make the decision as to why it should do that thing over the other thing, with all that business context and the relationships that you have.

And finally, it's the interpretation. So the analysis of cause and effect. You know, AI can tell you what outputs are, or what happened based on the inputs.

But interpreting that data or knowing what to do with it is an inherently human skill. So for example, we're seeing a reduction in MQLs, turning into SQLs, marketing qualified leads turning into sales qualified leads. Why is that? What has changed?

What is our system telling us? And how can we turn the tides and interpret what is happening to make changes that are gonna move our business forward? And I think the interesting question emerging in a lot of orgs is is what does an app and a developer come in the world of agentic Salesforce? And I think the answer to that is an owner and a domain expert.

Someone who defines what agents do, validates what they've done as an accountable for the whole thing. You know, that verticalized business knowledge, knowing a health care workflow or financial services workflow, this specific way that your industry does things, that becomes more incredible, creates incredible value, as we move through in this new area.

And I think the theme that runs through any technology implementation of any kind, but especially in this new AI era AI era, is that technology is the vehicle, and humans are the driver.

So humans make the decisions, humans exercise judgments, and agents do the heavy lifting for us where it's applicable, and automation providing the guardrails that make the whole thing trustworthy.

And I don't think we should be, you know, just slapping AI on everywhere you can with sticky tape, and that's not what I'm advocating for. And what I am advocating for is for you to have a relentless focus about what your job to be done is, what are you trying to achieve at every stage of your development life cycle, What the outcome will be, and how that outcome will impact people in your business? Does it enable them? You know?

Does it restrict them? And is AI the most appropriate thing to solve that job to be done? And really coming down to that is up to you. If we go this route, if we're using AI sensibly and deliberately, you know, what does all of this get you?

Then I came back. We got the business impact. At business level, better change, deliver faster, more confidence, fewer regressions, faster cycle times, more visibility into what's in your environments and why. When something goes wrong, you'll be able to fix it.

Ultimately, what it comes down to is your Salesforce org is mission critical.

It runs sales pipelines, your customer service maybe. It houses the data. You know, everybody's talking about how important data is and that data being used to make decisions. Supporting systems need to move at the speed of your business, and I think businesses are moving pretty darn quickly at the moment. And for a lot of teams and a lot of organizations, those systems aren't moving fast enough to meet that pace of change.

Many companies have AI mandates that, looking to see where they can leverage AI in a way that provides value. And for my money, there's no real better place than Ewing seeing AI in a smart practical way that supports a tier one business application like Salesforce with far reaching far reaching impact and consequences. And I think that's something your executives can get behind if you have those types of mandates.

I think the most important place to end on, however, for myself and and for you is ultimately the human impact. And the best version of this feature isn't one where people are redundant. It's one where people aren't stuck doing things that, quite frankly, aren't a good use of their time or their abilities. I don't think anybody got into this ecosystem because you wanted to manually click through deployments at night or repeatedly click through hundreds of set of Salesforce setup screens.

You got into this thing because you wanted to solve problems. We're all problem solvers in this ecosystem. You wanna help your business run better to build things that actually matter. And to me, humans, we're the stewards of trust.

We we are trusted to get things right for this business, and that requires all the things that make us implicitly human. It requires conversations, the relationships, connections, thoughtful judgment being exercised in a nuanced world as we look to translate all of those business processes into the technology solutions.

The agentic future, if we're doing it right, gives you more opportunity to do that, more time for the high judgment work, and more space to actually understand the business that you are serving, and more ability to be that person who's gonna be accountable and visible and valuable to your organization.

But the catch to that is, obviously, you need to do it right. So, done right requires an investment in your DevOps foundations, you know, your deliberate vertical AI adoption, but most importantly, in your role as well, thinking about what you bring to the table, what use skills and value you have to offer, and how that will ultimately shape the future of your business, your own career, and how you deliver value as a person.

So for my money, the teams that make investments in what the human role is now and truly get down to what it means to be a human in your business are the ones who are gonna look back in a couple of years' time and say, we got ahead of this, and we got this right.

So your future, if you choose to accept it, has DevOps done right, deliberate AI implementations and understanding human impact? Those are the things that I will leave you with. If you would like to understand more about how Gearset can help with some of the things that I talked to today, then please reach out to me. Please reach out to somebody, at Gearset.

There is there is lots of us. We have a little In app chat widget, on the website there as well, and one of us will be more than happy to help guide you, on the journey that you're on. That is it for me. I have just come in just over time.

But if there are any questions and folks are still here, then I'm happy to take a couple of questions, in the chat here.

Sean, thanks for your question. Perhaps I missed it, but mention of vertical AI implies some other angle, perhaps horizontal.

Could you elaborate on this distinction you are trying to make by calling it out vertical specifically? Yeah. So horizontal AI, I would describe as, you know, the platforms, you know, like, the big four platforms for AI. So Gemini, Claude, OpenAI, ChatGPT, etcetera.

You know, AI that they apply to most things. With vertical AI, there are AI tools and solutions. Gearset has one specifically that looks just at Salesforce, with all the Salesforce specific context it needs. So vertical AI solutions within those specialist fields.

So just for Salesforce implementations is obviously the example that we're gonna go for here, but another vertical might be this AI is built and designed specifically with health care organizations in mind or health care workflows in mind. So those vertical AI solutions in studies, there are a bunch of them online already, but companies that have looked at those vertical solutions have seen much quicker ROI on their investment than just adopting something like Claude and building skills and things like that, to push them forward.