Agentforce is the latest game-changer from Salesforce, unveiled at this year’s Dreamforce.
This is what a more advanced, thought-out, and active AI looks like – not just the ability to discuss your business and make recommendations, but the ability to take action on tasks within your business. So where does this leave Salesforce Flow?
Understanding Agentforce
When Agentforce was first introduced at Dreamforce ‘24, many thought that it was just Einstein Copilot rebranded, but this is not the case. There has been a lot of new functionality added to give Agentforce its own “persona”.
Agentforce brings a necessary layer on top of just being a large language model (LLM), adding the ability to perform actions as well as read, understand, and make suggestions around your business data. This adds a whole new dimension to the functionality that Agentforce can do for you, and this functionality is what helps it stand apart from other tools.
What drives this intelligence is the Atlas Reasoning Engine, and the focus for Atlas has been giving it agency – the ability to do for itself rather than assist a human.
You can configure Agentforce to take action on your data, reducing the headaches for your business users and allowing them to focus on more important work. Wait – that sounds awfully familiar to what automation tools such as Flow have done in the past. With Agentforce now having the ability to read your data and act on it accordingly, what use does Flow have anymore? Cue the existential crisis!
Is Flow Still Necessary With Agents Doing Everything?
While those of us who have just spent the last few years upskilling on the Flow tool may be a little shaken when the Agentforce and other AI announcements came about, I believe it’s all in vain and there is not much to worry about.
When you think about it, it’s the ultimate “AI is coming for our jobs” discourse that’s been circulating for years. This development of fear is bound to occur when there are gigantic advancements in technology, like those offered by Agentforce, but it’s always best to do your research and keep your head on straight.
Believe it or not, Flow and Agentforce will actually work hand-in-hand together in a larger capacity in the future; at least that’s my guess. It doesn’t make sense to completely replace Flow with Agentforce given they behave quite differently and have different purposes, which we’ll explore further below.
Currently, flows are the building blocks of Agentforce, used to give it the ability to take action. I believe we’ll see them work even closer together in the future by allowing Agentforce to suggest Actions that it could perform.
Given that Flow is typically a lot easier to show to a business user and have them understand it (in comparison to Apex), I could see a world where Agentforce identifies a way it could assist as well as generate flows that perform an action and shows it to the admin (or even the end user) to ask permission before actually using it. All the pieces are there, it’s just a case of tying it together and making it happen.
As the AI boom has shown us, we can never truly know exactly what the future will hold – but I believe one thing is certain: Flow isn’t going anywhere in this new Agentforce world.
How Agentforce and Flows Work Together
Earlier in this article, I talked about how Agentforce adds an action layer on top of standard LLM functionality. You’re likely familiar with an LLM like GPT by OpenAI or Google Gemini – these are both fantastic, but can only offer you advice on what to do.
If you give ChatGPT a scenario (with relevant fake data to back it up) and ask it to perform a task for you, it would typically tell you that it’s not able to perform the task and would instead provide you with detailed instructions on how you can perform the task.
Handy, but not quite what we’re looking for. This is where Agentforce and flows begin to shake hands and really save you some time.
When building with Agentforce, you can use your existing flows, Prompt Templates, Apex, and APIs into what Agentforce calls actions. Instead of the AI figuring out what you want to do and how to do it best, you can now enable it to take action on your business’s behalf.
Another way to look at it is like this – to date, you’ve made do with record-triggered flows, input-triggered flows (screen flows), schedule-triggered flows, and platform event-triggered flows. I’d look at the Agentforce and Flow partnership as the ability to create logic-triggered flows.
Agentforce’s powerful Atlas Reasoning Engine paired with the entire library of flows that you’ve used in your business to date (as well as those you’ll continue to build in the future) allows your CRM to take action based on logic and reason, rather than a defined set of criteria.
What Could the Future Hold for Agentforce and Flow?
I’m excited about the future of Agentforce and Flow. I think we’ll see a lot more people adopt Flow who have previously avoided it now that it is being made far more capable by Agentforce. I think we’ll see a lot of businesses harnessing the power of AI in a way that they never thought possible before by having agents activate complex business automations using advanced reasoning.
We’ll likely see more flows being created for very specific tasks and more modular design overall. subflows will be used more to call the smaller ‘action flows’ from a more traditional record-triggered or similar flow, which allows that same functionality to also be called from within an agent.
We will see more flows, not less. We’ll see more ability to streamline business processes using a combination of Flow and Agentforce, and not a replacement of Flow by Agentforce.
This is also one of the reasons that I have high hopes for Agentforce as a whole. Salesforce has been able to deploy an advanced AI tool that leverages decades of data architecture within CRMs that can take action with automation tools and APIs that are in use and have been in use for many years. There’s no starting from scratch here, only enhancing beyond our wildest dreams.
Dream or Nightmare: Your Choice
Any time the topic of AI is raised we can expect to hear the usual tropes – Terminator being a good example – but that’s not what I’m talking about in this case. Building out a tool as powerful and innovative as Agentforce can save your business time and energy while creating spectacular experiences for your customers.
However, we all have learned by now that a powerful tool like Agentforce needs to be planned out carefully and implemented correctly to ensure it gives you value.
There’s a saying from Spider-man I use a lot when talking about Salesforce Flow: With great power comes great responsibility. Flow is an incredibly powerful tool, and that power used without caution can be destructive to your Salesforce environment. The same sentiment applies to flows built with and for Agentforce.
Given that these flows are going to be used by Agentforce, you’ll need to take extra care when building them. All the standard best practices are important, but an often overlooked one is critically important here: document your flows, and specifically fill in description fields in every element of your flows. This provides guidance for users and artificial intelligence alike to understand what each element is used for and how it functions.
Think about it: if you’re watching a movie and the plot is “Someone named Tim got a coffee. He was attacked. He later left the coffee shop and beat his attackers. The end,” you’d be a little confused – not to mention disappointed. Description fields in Flow elements provide the same substance as the actual plot of a real movie. It fills in the gaps, explains the why, not just the what, and gives greater insight into the purpose the Flow actually serves.
In the example below, we can see that the instructions pulled into the Flow Apex Action come directly from the description field on the input variable inside the Flow.
As mentioned earlier, it’s not just documentation that is important – ensuring your flows are architecturally sound and follow best practices is just as critical here, if not more so. Agentforce will only have access to what you give it, so you want to make sure that the tools it has are working as expected to avoid issues in your org.
Given that Agentforce and other AI tools only know what we tell them, it’s important to share the whole story. This will help create more useful agents, and help ensure they know what they’re doing with the tools they’re given!
Final Thoughts
What are your thoughts on the future of Flow in an Agentforce world? Do you share my optimism, or is there another train of thought that you’re following? I’m curious to learn what others believe the future of Flow will look like with AI now much closer than just “on the horizon”.
AI is here, and here to stay. By leveraging existing tools such as Flow, it will be relatively easy to embed Agentforce deep within your current Salesforce environment and have it perform tasks for you much sooner than other tools.
Comments: