3 comentarios
Grupo
Agentblazer Community Group
Habibi, engineered for 100-year storms & summer heat!
🏙️⚡
Tomorrow is Salesforce Well Architected - Session 3: Built to Bend, Not Break!
Just like Dubai’s iconic skyscrapers, your org needs to sway with the wind without collapsing under heavy load. Come grab your coffee, join the community, and learn how to build systems with zero breaking points!
🔗 Lock in your spot here: https://trailblazercommunitygroups.com/j/rsnter9h3mhnb/
@Salesforce Developer Group, Dubai, UAE @Architect Trailblazers @* Salesforce Developers * @Agentblazer Community Group
🚨 Action Required: Transition your legacy Agentforce Builder model
What’s Changing?
OpenAI is deprecating the GPT-4o model currently powering our legacy Agentforce Builder reasoning engine. Starting October 15, 2026, you have two options to keep your agents running smoothly:
- Option 1 (recommended): Upgrade to the new Agentforce Builder
- Move your agent to our new builder to unlock enhanced features and long-term support. Review this Help article for instructions on how to upgrade your agent to the new Agentforce Builder.
- Option 2: Change the reasoning engine model on your agent within the legacy builder
- Toggle your reasoning engine to the newly supported OpenAI GPT-5.6 Luna model. Review this Help article for instructions.
Next Steps & Testing:
Changing your reasoning engine model or migrating to the new builder can impact your agent’s responses. We strongly recommend testing your agent thoroughly and updating its configuration as needed to ensure expected behavior before switching in production.
Please reference the Help articles linked above for next steps on how to change the model in the legacy builder or consider migrating the agent to the new Agentforce Builder.
Explore the 🆕 Agentforce Builder with Agent Script
Explore the Agentforce deployments that are in production today with customers using the new Agentforce Builder.
Read more about the new Agentforce Builder and here’s the page for you to learn about Agent Script.
🗣️ Questions specific to your Agentforce deployment
Please reach out to your Account Executive if you have additional questions or need support.
4 comentarios
Hi @Lim Chong Meng – So glad to hear you're able to see the articles now!
OpenAI GPT-5.6 Luna model is specifically noted above as a model available to customers who wish to remain on the legacy builder, so they can switch the model that's powering their agent in production today.
In the new builder, you'll have the option to select the model that best aligns to your use case from the set of available ones from a number of different providers.
Could Agentforce help improve a duplicate review process, particularly for the scenarios where existing Salesforce matching logic does not identify potential duplicates? Could Agentforce flag likely duplicates for a user to review?
3 respuestas
,
Yes, and it’s a good use case, but I wouldn’t hand the whole job to the LLM. Here is what we tried and suggest:
- Narrow the candidates first. Use SOSL, a looser matching rule, or Data Cloud Identity Resolution with fuzzy match rules to pull a small set of possible matches. Don’t ask the LLM to scan the whole database.
- Let the LLM judge similarity. Pass the record and its candidates to a Prompt Builder template that returns a likely duplicate yes or no, a confidence score and a short reason. This catches what string matching misses: nicknames, abbreviations, “Ltd” vs “Limited”, typos, and moved addresses. It’s a semantic problem, not a string match.
- Keep a human in the loop. Invoke it from Flow or Apex, create a review record or task for likely matches, and let the user decide on the merge. Don’t auto-merge.
Hope this helps.
The Agentblazer Adoption Playbook is live 🚀
How do you get real value from AI agents? Nine builders shared what they built, where they got stuck, and what they'd do differently. The playbook has 5 chapters and 4 take-home tools.
🎯 Customer outcomes from Agentforce
- Workday: 21% case deflection by the end of Q1, double the 10% target
- Datasite: 82% case deflection on its Agent Script rebuild
- MIMIT Health: 459% ROI in three months
- Southwest Airlines: $6M projected annual savings and 7× ROI
📖 What's inside
1️⃣ Start with a Metric that Matters: Andrea Simon and @Rushikesh Choudhury from Workday, plus @Chris Emmett from Capgemini
2️⃣ Ship the Smallest Winning Use Case: Sarah Morton formerly with Engine
3️⃣ Context Beats Perfection: Dr. Romi Chopra with MIMIT Health and @Cecilia Chiderski with CLD Partners
4️⃣ Build It Right: @Sally ElGhoul from Code with Sally and @Grant Roberson from Datasite
5️⃣ Scale What Works: @Ishrat Bhatti from Southwest Airlines
🧰 Take-home tools
- Plan Your Use Case
- Choose Your Pilot Path
- Build Your Business Case
- Set Up Your Center of Excellence
We have more users than agentforce seats, but enough to give some heavy users unmetered access
4 respuestas
The quickest path is to query the TenantEnrichedUsageEvent object via the Query Editor, or use Data Cloud to build a custom report mapping those consumption insights directly to individual user IDs. If you just need a quick look at token volume, you can also spin up a regular Salesforce report using the GenAIGatewayRequest logs grouped by user.
New Salesforce Break content:
This AI Coworker Saw What Humans Missed: Agentforce Coworker(YouTube video)
Agentforce Coworker Applied: Enterprise Intelligence at Your Fingertips(article)
Is Setup with Agentforce Ready for Prime Time?(article)
Should You Use Roll Back Records in Salesforce Screen Flows?(article - best practice series)
Is Hardcoding Ids a Good Idea in Salesforce Flows?(article - best practice series)
Enjoy
#Agentforce #New Releases #Flow #Best Practices
What's new on Salesforce Break:
Winter '27 Flow Tips Every Admin and Developer Needs Right Now(YouTube)
Collection Filter Criteria Resource in Salesforce FlowKeep Business Logic Outside Your Flow
#Flow #Flows #New Releases #Architects
3 comentarios
Collection filters flew under the radar in this release, but is probably my favorite feature. Going to be very useful.
I am working on an Agentforce Employee Agent and created a prompt template to summarize records and generate clickable record links.
Prompt used:
“Use your intelligence and provide the full clickable URL to access the record in the format: lightning/r/<Object>/<Id>/view.”
In the prompt preview, I can see the full link correctly, for example:
https://<mydomain>.lightning.force.com/lightning/r/Case/500XXXXXXXXX/view.
However, when I add this prompt as an Action inside the Employee Agent, the clickable link is displayed as URL_redacted
. If I ask the agent again in a follow‑up utterance to provide the links, it then returns them correctly.
Please advise if there is a solution or configuration that can prevent the link from being redacted when used within an Employee Agent Action.
13 respuestas
The Agentblazer Collection 🗞️
What is it?
The Agentblazer Collection is a weekly post shared in the community with resources from Salesforce related to Agentforce, community-contributed resources, and other exciting items ✨
Community Contributed Resources:
- Episode 1: AI + MuleSoft Architecture: Why Specification-Driven Development Matters | MuleSoft Vibes by @Stuti Tank, Software Engineering Specialist, LTM
For a resource be considered for a future post of The Agentblazer Collection, please share content with us by using the workflow.
Featured Resources this Week:
- 🎥 One Size Doesn't Fit All: Matching The Right Agentic Tool to Your Business Needs [Webinar Recording]
- 3 Signs You’ve Outgrown a Single Agent
- Fisher & Paykel resolves more appliance issues with Agentforce
- Your Agent Can Sprint, But Can It Go the Distance?
Hope your week is going well!