This week, I went to Salesforce TDX 2025 in San Francisco to network and learn more about Agentforce. “AI Agents” unlike co-pilots are intended to take action semi-autonomously (or without human intervention at all).
I know first hand that the most valuable aspect of attending TDX are conversations and insights that happen from direct conversions and hearing vendor’s real take on what’s happening with real world implementations. When the bright lights of the stage or recorded sessions are off, the real story surfaces. Here is what I found walking Moscone West this week.
Good Energy: Excitement for production grade agents
Salesforce AgentForce was built with the goal of deploying production grade AI agents. In that regard, attention to security, testing, and validation have produced a really good experience. We built our first agent at DataTools Pro last week. Salesforce is building an enterprise ready platform where security, testing, and validation are front and center.
Of course, Agentforce is not the only AI agent platform. Every technology behemoth wants to be the center of your “agentic” development ambitions. You have to carefully consider your cost and risk variables up front:
- Your business case
- Where the data comes from
- Volume of interactions (chat sessions, automations, volume of data processed, etc),
- Your team (who will build, deploy and own the agents’ performance)
These were key themes and interesting side conversations I had. I could put every person into 2 groups:
- Professionals learning the tools in the toolbox in search of a viable problem to solve with agents.
- Professionals with a reasonable problem, rationalizing the Salesforce flavor of agents and what can and can’t be done with the current set of features.
TDX had technology focused pods allowing me to break down AgentForce into its smallest parts and ask a ton of questions. I got a crash course grounded on real world use cases drawn from my customers. The Salesforce folks working the pods were fantastic.
Bad Habits: Technology in search of a problem
I wish I had met more folks or sat in more sessions that focused on real world problems worth solving. In that regard, I left un-inspired with the generic chat bot examples shown in presentations. That could my fault for not curating my sessions. That opinion may change as I go back to the agenda to see what I missed.
Self deprecating play on over marketing Agentforce made the presentations light hearted, but most were absent of real world knowledge, experience and examples. The pinnacle of this experience was watching two partners share tips like “document your use case” and “start simple, then iterate.”
TDX 2025 helping building confidence to build agents
The transition to agentic inside of Salesforce ecosystem requires an evolutionary step forward from looking at Salesforce solely as a business application. Salesforce DataCloud is a critical component for this transition. A lot of enterprises are maturing outside of Salesforce and now have to grapple where to start building and managing AI “agents.” Microsoft, AWS, Google, DataBricks and Snowflake direct integrations with DataCloud are important players for data and AI.
Salesforce also announced developer editions for DataCloud and Agentforce: Signup
My guess is this time next year we will see a ton of unique and interesting uses where adopters get expected or surprising results. There was a great article covering Salesforce vs Microsoft agentic ambitions that I recommend if you want to dig deeper into this topic. At DataTools Pro are working on our first agent for the AppExchange, knowing Agentforce long term will be a winner for Salesforce!
