Highlights
- Anthropic’s Claude is a powerful AI tool that can be used to help Salesforce Admins in their day-to-day tasks.
- There are a number of Claude native features that can and should be leveraged, as well as some Salesforce tools that can be connected with Claude to give more context to the LLM.
- Data security needs to remain top of mind when using Claude for Salesforce tasks.
- Salesforce Admins can make context management easier by leveraging Claude Projects, both in Claude’s main app and Claude Code.
- While AI is often criticized for causing human operators to become lazy, Salesforce Admins can use it to learn new skills and tools while improving their knowledge.
If you’re a Salesforce Admin and you’re trying to figure out how to create more value using Anthropic’s Claude in your Salesforce org, look no further! I’ve been using Claude and other similar AI applications for a while now, and I’ve curated a list of 10 tips and tricks that will make your life as a Salesforce Admin much easier.
As you read this, you may be wondering how much experience you need to have with Claude to benefit from these tips. I’ve designed this list to be accessible to anyone, regardless of their level of experience with Claude. Whether you’re a pro who’s been using Claude for ages, or are new and seeing what it can do for you, you’ll find something useful here.
1. Give Claude Appropriate Access to Salesforce
If you’ve ever worked with a Salesforce SI Partner or Salesforce Consultancy, you’ll know that it’s one thing to talk about your Salesforce org, and another thing altogether when they actually have access to it. The same goes for Claude. You can ask Claude (or ChatGPT, or Gemini, or whatever tool you’ve used in the past) hypothetical questions about your Salesforce org, or you can grant it access to see things for itself.
Giving Claude the right level of access to your Salesforce org means they have boots on the ground in terms of access to your metadata. Salesforce makes it easy to grant Claude access to your org by using hosted MCP servers.
At the beginning, you really should give Claude read-only access to your org, and nothing more. Think about the principle of least privilege; Only give individuals the access they need to do their job. Have a think about what you need Claude to do in your org. From the outset, probably not that much. Even if you’re getting it to work on some code or build metadata for you, you can (and should) handle the deployments yourself.
2. Leverage Projects for Specific Tasks
I’ve used the word “project” ambiguously in the header because I actually mean this in two different ways. Firstly, and the one you may find the most accessible as a Salesforce Admin, is a Project in the standard Claude app (or Claude.ai web app). Creating a project in Claude allows you to give it a set of instructions that apply for all conversations within that project, and load in additional files for a clearer context.
You could set up a project in Claude that has a PDF of the latest Release Notes, for example, and additional instructions like “We have an Enterprise Edition Salesforce org, and we use Agentforce for Sales and Agentforce for Service”. Then, when you start a new conversation in that project, Claude will have all of that context ready to go. If you ask it about new features that could be valuable to a specific task you’re working on, you don’t need to spend a lot of time directing it to the Release Notes or getting responses about features that are Unlimited Edition only.
The other meaning of “project” was referring to Salesforce DX projects in Claude Code. While you may not consider Salesforce DX projects to play a major role in your day-to-day as a Salesforce Admin, they can still come in handy. If you’re looking for a way to accelerate your learning of Apex, for example, you could build a DX project and use Claude Code to extract a class or two from your org and explain what they do.
3. Be Clear: API Names, Metadata, Problem Details
If a colleague taps you on the shoulder as they’re walking into a meeting and asks, “Hey, can you sort out that thing before I wrap this meeting? Thanks!”, you’re probably going to be a bit miffed at their lack of direction. Similarly, there’s nothing worse than working on building a complex report all afternoon because your manager asked for it, only to find out that they actually meant they wanted it as a dashboard, or wanted totally different data, all because they weren’t clear enough. The same goes for your Claude prompts.
You must be clear about what you’re asking for when it comes to Claude. Claude doesn’t know what you mean when you use internal slang that’s not documented, or when you use a generic term or phrase. It doesn’t know about how your team tends to use fields that aren’t labelled correctly, or other oddities that your org has. You need to be clear, giving direction and feeding it the information it needs to be effective.
The best and most appropriate way to do this is to use the API names, exact metadata names, and be very clear and descriptive about the problem you’re trying to solve. Don’t just say “fix the sales path, make no mistakes”, because that’s not going to get you any meaningful result. Explaining that when a specific user (it can help to specify which user too) clicks on a stage in their Opportunity record, it should update a hidden field that is not updating.
Tell it exactly which field you’re expecting to be updated, with what value, and whether it’s happening for all records or just some. Explain if it’s happening for multiple users, or just one or two.
4. “Interview Me” When Feature Building
You may have been doing your research on how to use various LLM tools and heard about the concept of asking it to interview you when you’re working on something. It’s a great place to start when it comes to building Salesforce features as well. You may be able to provide a lot of great detail about the problem you’re trying to solve, and how you plan on finding the solution, but there may be a grey area that you’re not aware of. Claude can identify this, gather additional information that it needs, and then take more effective action.
I’ve found this particularly useful when I’m getting Claude to assist me with some more complex LWC builds, or when I want to get it to help me plan out a more advanced flow. Despite the fact that I’ve had a decade of Salesforce consulting experience, there’s quite often an aspect or outlier that I hadn’t considered.
Having Claude there to ask me about it, and help me get my thoughts out in a conversational way, has been huge for me personally.
5. Get Claude to Show Its Working
I often hear the argument against AI that it makes people lazy, or is collectively making us all dumber. While I can totally see the argument, I don’t think it’s a death sentence. I believe strongly that it’s more about how you use AI rather than whether you use it or not. That’s why this next point is so important.
When Claude is building something on your behalf, ask it to show its working. Get it to explain, in detail, how it built something, why it did it that way, and any other decisions that it made along the way. In this way, you’re growing your own skills and understanding, rather than just sitting on the sidelines and literally letting Claude take your job.
Learning to become a better LLM orchestrator is a skill in and of itself, but getting Claude to show you new ways of doing things in Salesforce is also a great way to learn Salesforce better!
6. Teach Me!
Following on from the last tip, not only should you get Claude to show you how it did something, but you should continue to develop your own Salesforce skills by getting hands-on. You can learn new things in a dynamic way by sharing a problem you’re looking to solve with Claude, and asking it to run you through how to solve it. Not doing it for you, but showing you how to do it, and explaining its reasoning along the way.
Beyond just the Salesforce space, I’ve found working with LLMs to learn new skills, platforms, and approaches to problems has been profoundly useful for me. I’ve dabbled in technologies that I’d have never been exposed to otherwise, and taken some of those ways of thinking and applied them to my career in the Salesforce space too.
In The Matrix, Neo downloads the knowledge of Kung Fu and masters it in seconds. Claude can do the same thing, given the right skills. While you’re not able to do the same thing (yet, anyway), you can leverage Claude’s ability to download new skills as a way to accelerate your learning and solve problems together in real time. This beats needing to go and study a new tool from the ground up.
7. Turn Repeatable Tasks into Skills
Speaking of skills, you really should be creating a list of skills based on repeatable tasks that you and Claude do. A skill is simply a set of instructions for Claude to follow, and building them as a skill ensures consistency and saves you a lot of time prompting.
The simplest way to create a skill is to use Claude’s built-in skills creation ability. Depending on what you want it to do, it’s quite often as simple as having a conversation with Claude and getting it to write the skill for you. Then, you can head to the Customize menu’s Skills tab and upload the skill that Claude wrote for you. Simple!
The reason Claude Skills are so powerful is that they can be shared with other people in your team as well (either by sharing the skill files directly between individual Claude users, or by sharing within a workspace in a Team or Enterprise Claude plan). If you’re working to standardize a process or how your team approaches a certain task, you can leverage Claude Skills to make sure the whole team is on the same page and has access to the same functions.
8. Always Know Which Org You’re Pointing To
This may sound obvious, but… did I mention this list is a bunch of tips and suggestions from my own personal experience? Yeah, I’ve pushed packages to the wrong org before, and been arguing with Agentforce Vibes that a Flow has a problem while it tells me the flow itself doesn’t exist. It’s a CRITICAL part of the process to ensure you’re working in the right org.
This is especially important if you’re a consultancy that works with multiple clients, but it’s important for anyone. Pushing a package to the wrong org often ends in errors, but if it succeeds and someone suddenly sees another business’s metadata in their org, they’re going to have questions for you. Similarly, querying your Full Copy Sandbox about an automation that you’re building in a scratch org isn’t going to be helpful at all.
Make sure it’s the first step in any work that you do. Ask Claude to confirm which org it’s using at the moment, as well as any other orgs you may have connected at that point in time. It’s a small thing, and it’ll save you a lot of headaches. You could even consider adding it to your CLAUDE.MD file in Claude Code, or the instructions in a Claude Project.
9. Leverage MD Files, Not Chat History, For Context
When you talk to a friend, you do so verbally or with a text message. With a colleague, a Slack message or in the form of a document in more complex cases. When talking to Claude, you can communicate initially through text, but you’re better off using something a bit more permanent to keep track of longer conversations or projects.
When working in a Salesforce DX Project in Claude Code, you can use a Markdown (MD) file to keep track of the work Claude is doing for you and its progress through a task. I find this helpful for more complex automations, too. It ensures that Claude knows exactly what’s been done, what needs to be done, and it also gives you flexibility in case you want to switch to Codex or another model.
You can also get Claude to build you a full HTML file to explain a complex that you’re struggling to understand just in text format. Asking Claude to speak in websites is such a vibe, and a great way to communicate an idea between it, yourself, and anyone else who needs to be across an idea or decision.
10. Know When to Say No
Finally, this is similar to the advice I give when working with Salesforce Flow: know when to say no. This goes for access, and most importantly, your data.
You have a responsibility to protect your clients’ data, and you really need to take this seriously. A majority of Salesforce orgs that have had major breaches in the last few years have been due to a lack of security around their data. Sandboxes were breached, and they contained very real sensitive data. Your clients won’t care where their data was taken from, they’ll care that it was breached.
Summary
There’s a lot more where that came from! Anthropic’s Claude is extremely helpful in the Salesforce space, and with any tool, you need to make sure you’re doing things the right way. There are things you should do, things you should never do, and more efficient ways to do things. I truly hope that you’ve found the tips in this article helpful, and I’m even more hopeful that it’s so helpful you’ll want to bookmark it for later or share it with a friend!
What other tips and tricks do you have to share? I’m always keen to learn how others are using Claude in their day-to-day Salesforce work. Feel free to share this article on LinkedIn along with your own tips and tricks.
Comments: