3 reasons I am looking forward to the new Agentforce Agent builder experience

We are only a few weeks away from the new agent builder, and I am already looking forward to using this new tool. The demos I have seen look really promising, and I feel like this will be a huge improvement over the existing agent builder. If you spent any time in the existing tool, I'm sure you can understand the frustrations and challenges and making your agents work the way you expect. I've often used a strange analogy to describe this, which is wrestling with the jellyfish. Sometimes you feel like you get it under control, and you have everything in your grasp, and then it slips away, and sometimes you might even get stung. This is the nondeterministic nature of AI in general, and why we need new tools to be able to address this. Because of this, I have taken more deterministic approaches when building solutions in the past, such as lightning web components or screen flows, but now I am looking forward to the new builder to help me solve these challenges, and here are the three biggest reasons why.

1. Hybrid reasoning - Hybrid reasoning combines logic with AI reasoning and delivers, as Salesforce states, "Creativity when we want it and certainty when we don't". This means we can apply logic to follow fixed paths to ensure we raise the right topic at the right time, and utilize the AI when we need dynamic reasoning. Originally, I was a bit worried we were going to get an advanced version of flow builder for the new agent builder, but this is worlds better than what I expected. Agent script is the magic behind all of this and what this allows us to have three different options when building agents. The first experience is working in the no code interface, which looks very much like the way you would type instructions to an employee as to how a process needs to be accomplished, providing excellent readability. The second interface is a traditional coding interface, where agent script can be coded like any other language. Lastly, we have AI guided assistance where we can interact with an LLM to and ask to make changes to our agent, working in a vibe coding type of way.
2. Portability and maintenance - Previous, it was a great deal of work and a disjointed experience working within the agent interface. Agent properties were on their own screen, topics in one area and agent actions in another. Multiple layers to work with and difficult to see how it all fit together. In the new agent builder experience, everything is flattened and easy to see within one page. Even better all the agent's properties are exposed in code and those changes that were frustrating to make before are easy to access and change now, in fact the entire agent is one long piece of code can easily be copied and ported to another agent.
3. Testing - Testing and troubleshooting in the builder offers a number of useful new features such as mock mode (testing logic without running all the actions) and proper traceability, which is a headache in the current version of Agent builder. The interaction is fully searchable, and you will be able to track variable state throughout the process. We will now have full visibility into the latency so that we can optimize our agents and improve user experience.

Next

Now, I am still waiting to see how actions could be simplified and made easier. I have to say I am quite frustrated with the existing action experience as it takes a lot of work just to build out a single action, and after it's been created it is very difficult to make changes which ultimately leads to re-creating the action. This needs to be better, and I hope this is on the road map.