Agentforce Vibes – what can it do?

When writing an article about the latest Salesforce Spring ’26 features last week, I noticed an Agentforce Vibes option on the setup menu in the new pre-release developer org I had created.

Being curious, I just had to jump in and take a look.

Setting up Agentforce Vibes

I accepted the Terms and Conditions – knowing this was relatively new and could still be a bit buggy. I was just experimenting anyway and I was doing it safely in a developer org completely isolated from any “real” Salesforce environment.

Note: Agentforce Vibes was formerly known as Code Builder.
(of course, a name change makes it better)

It took a few minutes to set things up:

I had to click Install a few times to add some new extensions:

When all the extensions finished installing, I chose the resource panel icon (cloud) because it looked interesting.

This has lots of useful resources to help get started.

But what I really needed was the next icon down (Agentforce):

I had to tick I agree to the Terms then click Enable & Start Building

This (highlighted box) is where you enter your instructions:

Plan or Act Mode?

There are two ways to use Vibe: Plan or Act mode.

Plan is just as it sounds, where Agentforce Vibes will help you plan out what you want to achieve but it will not action and build anything.

Act is where you give Vibes the ok, so start building stuff – but it still asks your permissions before it does many things so you are still in control.

Plan was the obvious place to start, so I clicked the Plan button

Now I have done some work with Agentforce Vibes before, so this time I really wanted to push the boundaries and go as far as I could. Not satisfied with just a simple LWC, this time I wanted to see if I could build an entire app with help from Agentforce Vibes.

I decided a recruitment app would be a good test because I understood the business process fairly well. I drafted my requirements in business-like language as follows:

Build My App Instructions

Here is the exact prompt I used:

I chose the Plan mode, pasted those instructions in the task window, and hit the airplane icon to get started:

Off it went, spitting out the following:

The output was very detailed, so I will summarize here:

Proposed Data Model (Objects, Record Types, Fields, Relationships)

Approval to Proceed

Next, I was asked to make some decisions before proceeding further:

Once I had reviewed the proposed architecture, I was asked to toggle to Act mode to continue:

Now I toggled to Act

I chose the first option – Native Approval process and Auto create the offer record.

Now Agentforce wanted to use a tool on the Salesforce DX MCP server to create the custom objects, fields, etc locally.

I approved that option, then was asked to use my default org (the Salesforce Dev Org I was working in) as the target for the configuration we were creating:

I then went through a long process of approving the creation of each object and then deploying them. It was relatively quick and easy, just clicking save or ok each time.

After about 5 mins, this message popped up:

I pretty much ignored it at this stage and seemed to be able to continue.

However, that’s when the problems started.

What went Wrong?

To put it in simple terms Agentforce lost half its brain. It ran out of the 50 tokens available for a free Developer org and switched back to a simpler model that was far less capable than the Pro model.

As I continued trying to get the application completed, it ran into errors with the XML syntax in validation rules, had problems getting the XML right for the flows and had problems with the email templates.

There were also issues creating Lightning Record Pages.

Those things aside, the majority of what was required had been successfully built, but not all of it deployed.

What I should have done here is stopped immediately when I ran out of tokens and returned the next day with a new set of tokens and continued where I left off.

What was Delivered

The following was the result as written by Agentforce. Even if some of the items were not successfully deployed, it certainly got very close!

Deployment Status

The detailed deployment status showed what was complete and what was outstanding:

Evidence of the deployment: the custom Application object:

A validation rule to ensure an applicant has an email address (otherwise, a reject/accept email will fail). Agentforce decided this was necessary without any prompting from me.

While I did specify to use a record type for contacts for the job applicant, Agentforce decided another was required for the interview panel members.

Final Conclusion

From start to the aborted finish was about 60 minutes. This included me writing the relatively simple requirements brief and clicking Save/Ok/Next many times. If we had unlimited tokens then within one hour, I would have had a very comprehensive working application, at least good enough to use as a Proof of Concept or an MVP product upon which improvements could be added. As an experienced Salesforce administrator, it would probably have taken me about 8 hours to complete this app if I had built it manually.

First, I have to say I was impressed by the Plan/Act stage. This made logical sense to do the initial design first, review the design, and confirm it meets requirements before proceeding. If I had taken 24 hours to do this review, I would have had a new set of tokens to do the build and deploy, and possibly not run into the issues I had encountered.

The architectural/design decisions made were quite sound, it thought wider than my relatively simple set of instructions, and asked “intelligent” clarifying questions and made some great suggestions too. It also created validation rules and a record type without any hints from my specification.

Running out of tokens was a problem that could have been prevented had I chosen a simpler task to begin with. But doing something simple was not what I wanted to test this time. I really wanted to throw a relatively complex request at Agentforce Vibes to see how it would handle it.

Overall impressions: I was pleasantly surprised at the results; they certainly exceeded my expectations.

Create a new Pre-Release Org, select Enterprise and start experimenting. Or, after Spring ’26 is released to production environments, create a Developer Sandbox and start experimenting. Just remember to keep your experimentation well away from your production org for now – until both the technology and your competence reach production-ready status.

Similar Posts