Report from Salesforce's TDX 2025: AWS, Google Cloud, and New Tools to Build and Scale AI Agents
How do you grow to 5, 10, or 20 agents? With data, APIs, templates, actions.
Welcome to the Cloud Database Report. I’m John Foley, a long-time tech journalist, including 18 years at InformationWeek, who then worked in strategic comms at Oracle, IBM, and MongoDB. I invite you to subscribe, share, comment, and connect with me on LinkedIn. This post is sponsored by Salesforce. The views are my own.
The gist of this post:
At TDX 2025, Salesforce introduced Agentforce 2dx, the latest version of its digital-labor platform
Also new: a developer environment, marketplace, community
Early adopters are beginning to scale to “Day 2” projects
There are new capabilities for integration, agent actions, testing, and more
Salesforce has 5,000+ Agentforce customers and a growing list of references
Agentforce 2dx takes center stage
AI agents represent a potential multi-trillion-dollar market opportunity as businesses add these AI-enabled doers to the flow of everyday work. This kind of digital labor promises to not only boost employee productivity and improve the customer experience, but to augment human labor so we can add value in other ways.
That’s the vision. What’s the reality?
Last week, I attended Salesforce’s TDX 2025 developer conference in San Francisco to gauge the progress. It’s clear that, like previous digital transformations, this one will require a blend of business strategy and technical execution.
At TDX, Salesforce continued laying the groundwork to let its customers shift into higher gear. The company introduced Agentforce 2dx, a new and improved version of its digital labor platform, with additional capabilities for developers, administrators, IT pros, and other devotees. The pieces include:
Agentforce 2dx, its latest and most comprehensive agent-building platform, with pro-code and low-code tools to build, deploy, and supervise autonomous agents. There are analytics for monitoring, debugging, and optimizing performance.
Agentforce Developer Edition, a free environment with 10GB of access to Salesforce’s Data Cloud and 150 large language model (LLM) generations per hour. It can be used to build, customize, develop, and test agents.
A marketplace, AgentExchange, with prebuilt actions, topics, and templates from more than 200 partners, including Box, Docusign, Google Cloud, and Workday. There are hundreds of ready-made components and solutions in areas such as sales, service, finance, operations, and talent.
A certification program for Agentblazers, a new community for skills development and education. Salesforce also announced that it’s expanding TDX to Tokyo, Bengaluru, London, and other cities.
According to a recently released Salesforce survey, developers are generally optimistic about AI and agents, but there’s still a lot of work to do. Over half (56%) say their data quality and accuracy isn’t sufficient for successful development and implementation of agentic AI. And nearly half (48%) say their testing processes aren’t fully prepared to build and deploy AI agents. (See the report for details on methodology.)
The big picture
Most businesses are still in the early stages of AI agent strategies and projects. According to Salesforce, 5,000 customers had signed on to use Agentforce by the end of Q4 FY25, its most recent quarter.
However, that relatively modest uptake may be deceiving. The number of organizations evaluating agents is probably much higher, but the numbers don’t reflect it because they haven’t signed on yet. “Companies are still a little cautious, kicking tires,” said David Schmaier, Salesforce President and Chief Product Officer, in a roundtable meeting with journalists.
That will change as more tire kickers become adopters. Many CIOs and CTOs are already moving ahead with GenAI and LLMs, and agents are the natural next step. In fact, one could argue that moving too slow could be as risky as moving too fast. Because if digital labor proves a competitive advantage, CXOs don’t want to lag behind.
Agents don’t exist in a vacuum. Salesforce is building out an ecosystem to give its customers more choices in how they create and deploy these highly connected digital workers. The big news prior to TDX was Salesforce’s expanded strategic partnership with Google, enabling customers to build Agentforce agents using Google’s Gemini model and deploy Salesforce on Google Cloud.
With that top of mind, I was surprised to come across AWS on the TDX show floor at Moscone Center. Turns out that 60% of Salesforce customers already use AWS through Salesforce’s Hyperforce public-cloud offering. In China, they use Alibaba. So choice of cloud infrastructure is an emerging factor in agentic AI.
Of course, other tech vendors — Microsoft, Google, Amazon, Meta, SAP, Oracle, Workday, ServiceNow, OpenAI, and many more — are racing in the same direction, each with its own approach to agents. Salesforce feels it has advantages, including a head start and three essential elements — a data cloud, cloud apps, and its agentic layer. But the market is moving fast, and any advantages could be short-lived.
Digital labor at enterprise-scale
Salesforce CEO Marc Benioff, who has become the Chief Agent Evangelist for the entire tech industry, set a stretch goal last fall of deploying 1 billion agents by the end of 2025. To get there, Salesforce is now looking to empower a community of 1 million Agentblazers, the people who will make it happen.
That will require scale, a word I heard repeatedly at TDX 2025, both from Salesforce and its customers. To deploy at scale, developers and IT teams will need good, clean data, all-points integration, and robust testing.
“How do you go from one agent to five, ten, twenty with speed and quality?,” asked Gloria Ramchandani, SVP of Product at Copado, a Salesforce partner that offers a DevOps platform for managing the agentic software lifecycle.
It was a rhetorical question, but the right one for early adopters advancing to “Day 2” implementations. Ramchandani’s answer: Get your data right from the start.
Salesforce has been building out the tools and platform to enable this kind of agent expansion. It started last September with Agentforce 1.0. Two months later, the company introduced additional capabilities for enterprise deployment, including a sand box, test center, and monitoring/usage tools. Then, in December, Benioff rolled out Agentforce 2.0, with Slack and Tableau integrations and more agent skills.
Integrating data, workflows, agents
Now, Agentforce 2dx represents the next step in enterprise-class agent development, customization, and deployment. There’s a long list of new features and capabilities. Let’s start with integration.
An Agentforce API can be used to integrate Agentforce with other systems, apps, and sources that feed data into agents.
Invocable Actions enable agents to be embedded within Salesforce business logic, like Flow and Apex, so they can become part of workflows.
MuleSoft for Agentforce comprises an API Catalog with APIs from MuleSoft, Salesforce, and Heroku, and an Agentforce connector.
Agentforce Steps in Slack Workflow Builder let developers embed Agentforce into no-code automations in Slack. So Slack conversational context can be passed to Agentforce automatically without a user having to do it.
Context, actions, use cases
Also, here is some of what’s new and needed for agent-building.
An Employee Template creates agents that can be configured and deployed for employee use cases across Slack and Salesforce apps.
Surfaces let developers customize Agentforce for specific devices or channels. And Cards make it possible to reuse existing Salesforce components to give context and functionality to Agentforce.
Tableau Semantics can be used to create data structures for Agentforce and provide real-time business context for insights and answers.
Configure, test, deploy
And finally, there are new capabilities to control, understand, and fine tune agent behaviors.
AI assistance in Agent Builder troubleshoots answers that Agentforce provides and offers guidance on how to improve topics and instructions.
Testing Center can automatically generate and run test cases in the sandbox. Importantly, developers can evaluate how well Agentforce configurations adhere to guardrails like faithfulness or relevance.
DX Inspector provides visibility into the metadata and data used by Agentforce, including agent topics and actions.
Interaction Explorer provides reporting on how Agentforce is performing and session tracing. Developers can review user requests and the reasoning steps associated with responses and get recommendations to refine topics and instructions.
Growing proof points
Agentforce 2dx is the upper layer of Salesforce’s agentic tech stack. Other layers include the Salesforce Platform (Mulesoft, Flow, Atlas Reasoning Engine), Data Cloud, and Salesforce Customer 360 apps.
With all of these pieces working in concert, Salesforce execs argue it doesn’t make sense to “do it yourself” (DIY) by stitching together LLMs and other disparate pieces. “We have a data and AI platform that’s better than DIY AI,” said Schmaier. “It’s faster than DIY, purpose-built, and the time to value is much better.”
There’s a growing list of Agentforce customer references, including 1-800 Accountant, The Adecco Group, Engine, FedEx, Goodyear, Heathrow Airport, Oregon Humane Society, Precina, Royal Bank of Canada, Saks, Vivint, and Wiley. At TDX, I heard from Shayne Smyth, CTO at Saltbox, and George Pokorny, SVP of Global Customer Success with OpenTable.
OpenTable is using Agentforce to help its customer service personnel better serve restaurants and diners. After just three weeks, Pokorny said, Agentforce agents were already handling 73% of all restaurant web queries — a 50% improvement compared to the tool it had been using.
Adam Evans, EVP and GM of Salesforce’s AI Platform, was asked during a media briefing how customers typically get started with agents. He pointed to customer service, sales teams, and self-service as common starting points. In my own experience, I have talked to several early adopters who began with internal use cases before turning agents loose on external customers.
If all goes as envisioned, it won’t be long before these autonomous agents begin to appear in more of our everyday experiences. And they won’t be mere chatbots. They’re becoming more proactive, reasoning and taking action without direct human oversight.
Here’s another way to think about it: Salesforce built its business on the software as a service (SaaS) model. Agentforce flips that — it’s service as software. And the beauty of digital labor is that the more you use it, the more you can potentially save through productivity gains, according to Schmaier. “It’s a virtuous cycle,” he said.
I hope he’s right because I’ve been working hard, maybe too hard, for 35 years now. Bring on the agents.
