Anthropic’s Claude is currently one of the biggest stars of Salesforce’s show. There’s no question as to why, either. It’s such an incredibly powerful tool that is accessible to almost everyone, and easy to wrap your head around once you get started.
That being said, SF Ben recently learned that over three-quarters of Salesforce Admins have never touched a vibe coding tool of any kind, which leaves a lot on the table in terms of the potential of Claude for Salesforce Admins! The intent of this article is to share 10 practical examples of how Salesforce Admins can use Claude in their day-to-day.
Before you proceed, I want to be very clear that you should never blindly trust the output of any AI tool, including Claude. You should always validate and test the output before you deploy it, or share it as fact. This goes for any and all AI tools, not just Claude.
Considerations
Prior to moving forward with any of these use cases, I should be clear about what I mean when I say “Claude” in this article. I’m referring to a combination of Claude.ai (the web app/desktop app) and Claude Code, the latter of which will then have limited access to your Salesforce CLI as well.
I’ve found that the easiest way to do this is to build a new Salesforce Project in VS Code, and retrieve relevant metadata from Salesforce into it. Then, you can tell Claude to work with that metadata and you don’t need to give it direct access to your Salesforce org, especially if your business prohibits this.
You MUST follow human-in-the-loop practices when using Claude. Claude is a tool, not a replacement for a human being. In the above examples, your goal is to save yourself significant amounts of time and energy, not try to make yourself redundant. The latter simply doesn’t work, and isn’t a good idea to attempt.
If you’re able to, you should consider leveraging the Salesforce DX MCP. I’d only really suggest doing so if you’re more advanced and comfortable with the Salesforce CLI, because some of the tools in this suite are quite powerful. That being said, if you know what you’re doing and have the ability to do so, it’ll be a great way to expand your use of Claude in your day-to-day. Doing so will reduce hallucinations, give the agent more ground truth about your org, and create less friction for the agent, usually resulting in more efficient task completion (which will cost you less).
Task 1: Write and Debug Formulas
Formulas are used everywhere in Salesforce, and some can be more complex to build than others. Not necessarily because the requirements are overly complicated, but rather because there are so many different ways to approach a formula, and not all of them make sense for a given scenario.
Some may even require a helper formula to be leveraged for the best results, but using these unnecessarily makes things complex for no real reason.
This is where Claude can come in. Rather than testing and figuring out what happens best over multiple iterations, you can ask Claude to assist you with architecting the best solution for your requirements and providing the formula text for you to use as well.
While you may be fully capable of writing and debugging your own formulas, leveraging Claude to do so will save you significant amounts of time, both in building and in testing. This is a great use case for Salesforce Admins who are looking for ways to put Claude to work in their day-to-day.
Task 2: Write and Debug Validation Rules
Yes, validation rules are a form of formulas, but they come with their own special set of challenges. Validation rules have always been the bane of my existence (don’t ask me why; I can’t quite put my finger on it). Something about the criteria needing to be written in such a way that it ONLY fires when something is wrong just doesn’t compute in my head.
In my over a decade of Salesforce experience, any time I’ve written a validation rule, I’ve accidentally written it so that it fires when it shouldn’t, as opposed to when it should, and I then need to reverse it.
All this stopped being a problem for me when I started using ChatGPT to draft my validation rules a few years ago, when consumer LLMs were just starting to become a thing. Today, unless it’s a super simple validation rule, I’ll use some form of LLM to draft it for me every single time. With the more advanced LLMs that we have today, I also find that tools like Claude will suggest edge cases for me to consider while I’m building the validation rules.
When building or editing a validation rule, make sure you’re providing Claude with all the context that it would need to be able to accurately draft the validation rule for you. This can be done by either providing the API names of objects and fields that are involved and ensuring it has visibility into your org’s metadata if necessary.
Task 3: Requirements to Solution Mapping
Capturing requirements is a critical part of any Salesforce project, and it can be a headache to do. Often, it requires multiple rounds of conversations with different stakeholders to get the full picture of what’s needed. Then comes the fun part: translating those conversations into a clear set of actionable requirements.
If you’re like me and want to speed up this process, then Claude can be a great way to do this. Most conversations with stakeholders will happen online in tools like Microsoft Teams or Google Meet, each with its own ability to transcribe the conversation.
With this transcript, you can use Claude to extract the requirements from the conversation and map them into actionable requirements. You can take all of this one step further by creating a repeatable skill that you use to generate project specs based on conversation transcripts or solution design files.
This is where I’ll remind you that it’s important to follow human-in-the-loop processes. Do not take Claude (or any other AI for that matter) at face value. It’s your responsibility to check that what it outputs is correct, accurate, and complete. In this instance, it ensures that your requirements are going to achieve the desired outcome that your stakeholders want, and spent the time to discuss in detail with you. Make sure you respect their time by paying attention, even when using an LLM tool to help you.
Task 4: Document Automations in Plain English
Those of you who have inherited a Salesforce org with little to no documentation will know how much of a pain it can be to figure out what’s actually happening in that org. It gets especially tricky when there are hundreds of automations across various different tools (flows, Apex, and potentially even legacy Process Builder automations).
This is where a tool like Claude will shine. If you give Claude the ability to see your metadata, it can help you by translating what the automations are doing into plain English. Yes, it’s entirely possible for you to do this yourself, but using Claude will save you a lot of time. Additionally, if you’re not so familiar with apex, Claude will be able to fill in the gaps for you.
Particularly for those who are starting from scratch when it comes to your documentation, Claude will make your life a lot easier. Once you give it the ability to see your org’s metadata, it will be able to construct the full story of what’s actually going on, even when one tool calls another (like a flow calling invocable Apex).
It’s here that I’ll remind you of the important note at the top of this article. Do not blindly trust the output of AI tools. Make sure to validate Claude’s output, and use it here as a guide, not Gospel.
Task 5: Generate Complex SOQL Queries
SOQL, or Salesforce Object Query Language, is a powerful tool that can be used for extracting data from Salesforce quickly. It can also be used in Apex code to build dynamic queries, or even in custom applications to surface relevant data. Building SOQL queries can be time-consuming, especially when you’re not so familiar with it.
Leveraging the power of Claude to generate SOQL queries will save you stacks of time when it comes to both the construction of the query, as well as the testing of them. Nested SOQL queries and queries with complex WHERE clauses can be a nightmare to build, but Claude can do it with its proverbial eyes closed!
You can leverage the new Web Console to preview the output of a SOQL query before you put it to work in your code. It’s also a lot quicker than spinning up a new VS Code or Terminal instance of Salesforce CLI, in many cases.
Task 6: Construct Accurate Descriptions
Hands up if you’re guilty of not populating your description fields in Salesforce properly. I’m certainly guilty of it myself, despite the fact that I’ve been doing this for a long time. Additionally, if you’re not the only Salesforce professional who has had access to your Salesforce org, there’s likely a significant number of description fields that are missing entirely or have been populated without much thought.
Claude’s ability to read and understand Salesforce metadata means that it can be used to help you clean up your missing description fields with ease. Whether it’s the description fields in a flow, on a custom field, or anywhere else, Claude will be able to assist you and make it easier to populate them with relevant information.
Keep in mind that description fields are designed to give context both to humans and agents. It may seem counter-intuitive to get Claude to assist with the population of these fields, but when you consider the amount of context you’re saving your end-users once it’s done, it’s a no-brainer.
Task 7: Clean Up Messy Requirements
This is an add-on to an earlier task, but I’ve experienced this myself so many times that I figured it’s worth separating into its own point. Extracting requirements from stakeholder conversations is one thing, but trying to make sense of messy requirements is a whole other beast.
Once again, if Claude has the ability to read your org’s metadata and relevant documentation, it’ll be much more able to assist you in cleaning up your messy requirements. If you give it as much context as possible so that it has everything it needs to do the job, you’re more likely to get a better result.
While Claude can help here, it’s critical that you get buy-in from your stakeholders if you’re trying to change requirements after they have been provided to you. Just because your requirements aren’t great doesn’t mean you can change them without approval. Leveraging Claude just makes your job a bit easier.
Task 8: Plan and Build Security for Your Org
Salesforce security can be a bit fiddly, especially when you’re trying to clean up an existing permissions model. You need to map out what your users currently have access to, compare it with what they actually need, and then ensure you’re building a model that supports this.
I’ve found that one of the easiest, most low-effort ways to put Claude to use in your Salesforce org is to get it to assist with the creation of Permission Sets and Permission Set Groups. Give it the ability to read your current permissions model, and instruct it to design a new one based on the current best practices (moving permissions away from profiles and towards permission sets). This works for any security features in Salesforce that you can deploy through the CLI as well.
Once again, it’s critical that you test the output before you deploy it. You’re dealing with security here, and it’s something that you really don’t want to mess up. Claude is a tool that you’re putting to use, which means that you are still responsible for what you deploy to production. Test it thoroughly before you deploy it, ensuring that you adjust anything that Claude didn’t get quite right.
Task 9: Write Release Notes for End Users
When you make significant changes to your Salesforce org, it’s important that you communicate those changes to your end users so that they know how to take advantage of them. There’s no point putting the time and effort into building something if no one is going to have a clue how to use it.
Typically, when you make significant changes to your Salesforce org, you’ll deploy them using the Salesforce CLI or another CI/CD tool. This will be in the form of a package that contains all the metadata relevant to your change. Give this to Claude as context, and instruct it to draft the release notes for your business according to your business’s standards.
For additional context, you may want to give Claude access to the requirements that led to the metadata change as well. This helps Claude to understand the full story and communicate it clearly to your end users. Your documentation shouldn’t just tell users what is coming and how to use it, but also clearly communicate why the changes will be beneficial to them.
Task 10: Construct Metadata Quickly
I briefly touched on this in Task 8, but once again, I think this deserves its own section. Leveraging Claude to construct metadata quickly is a superpower that you’ll be addicted to once you get started.
Once your requirements are clear, you’ll often have a document full of things you need to build. Let’s say, for example, that you need to create four new custom objects, and each has about twenty custom fields on it.
Add to that the permissions and permission set groups that you’ll need to create for each, and you’ve got quite a bit of work cut out for you. However, you can speed this up significantly by providing Claude with that same document and asking it to build the metadata on your behalf.
Claude will construct your new objects, fields, permissions model, and can even assist in developing the relevant Lightning Pages that you need as well (although this has been hit-and-miss for me).
Your workload will be significantly reduced, and you’ll be able to focus more on orchestrating Claude and testing its changes. That beats the repetitive work of creating everything by hand any day!
Summary
In the 2026 Admin Survey SF Ben found that 43% of admins use AI tools regularly to do their jobs. Next year, I would love to see that the 2027 Admin survey showed that three-quarters of you have of you had! The power that tools like Claude have for Salesforce Admins is immense, and it feels like a lot to be leaving on the table!
My suggestion would be to identify one or two of the above tasks that you could leverage Claude for, and give it a go in your own org. Remember, you are responsible for what you deploy into production, whether you built it yourself or used Claude to help you. Test thoroughly, and make sure you’re comfortable with the results before you deploy.
