TDX last week was an absolute blast. Being back in San Francisco, co-presenting with Salesforce, and sharing insights on Agentforce was a real highlight. Aquiva led three presentations – one on Security Review for Agentforce and two others diving into the top ten tips for getting the most out of it.
The Partner Summit the day before TDX was buzzing with exciting announcements for technology partners. But during TDX itself, the focus was squarely on Agentforce. It was everywhere – from the keynote to almost every session. It was amusing when someone quipped during the ‘True to Core’ session, “Are we really true to core? Because everything’s about Agentforce!” And they weren’t wrong! Even though there were updates on the core platform, they were overshadowed by the big reveal: Agentforce 2dx, Salesforce’s enhanced AI platform that now operates autonomously in the background of any business process. It was clear: Agentforce took centre stage.
Why Agentforce 2dx Matters
So, why does Agentforce 2dx matter? It’s a big deal because it takes AI from simple chatbots to agents that can actually reason, act, and automate complex tasks. One of the major challenges businesses face is scaling human capacity. As highlighted in the keynote, we can scale infrastructure and digital capabilities, but not human capacity. That’s where Agentforce 2dx steps in, enabling AI to boost productivity autonomously. Adam Evans summed it up perfectly: “Software is changing. We’re moving from clicks to conversations, from static apps to dynamic reasoning.”
We see it first hand at Aquiva – this shift is already resonating. Over the past three months, we’ve collaborated with five partners to develop applications for the new AgentExchange. These applications are already powering Salesforce agents with specific topics, instructions, and actions. With Agentforce 2dx, companies can integrate these autonomous agents into their data systems, business logic, and user interfaces, letting them anticipate needs and unlock value quicker. It’s a big leap from a simple chat window to a fully integrated AI layer across Salesforce’s ecosystem. While it’s not quite Agentforce 3.0, 2dx definitely packs a punch, signaling Salesforce’s push to make AI a core part of their suite.
Key Capabilities of Agentforce 2dx
Salesforce is throwing a lot of weight behind Agentforce 2dx, and you can really tell. They’ve rolled out some impressive updates, like the enhanced Atlas reasoning engine, plus new automation features with variables and filters. They’ve also expanded integrations, improved UI options, and added asynchronous background processing. There’s a ton happening behind the scenes, and it’s clear they’re serious about pushing the boundaries of what AI can do on the platform. Key highlights:
- Enhanced AI Reasoning (Atlas 2.0): Advanced decision-making capabilities enable proactive, independent actions and reduce human intervention. This probably deserves a separate article, but lots is happening here with Planner (SFR-Embedding, SFR-Reranker), Tool selection (xLAM), Evaluation (SFR-Judge) and Feedback.
- Proactive Automation: Agents can act independently rather than just responding to user input.
- Deterministic Logic with Variables & Filters: Ensures precise and compliant AI behaviour, critical for regulated sectors.
- Integrations: Better embedding of AI across web platforms, mobile apps, and bespoke interfaces, transforming daily business operations.
- Expanded UI & Surfaces Integration: AI agents can interact with custom UI components, mobile apps, and websites.
- Agentforce Cards: LWCs can be embedded within the response of Agentforce actions.
- Asynchronous Background Processing: Agents autonomously manage complex workflows, significantly boosting operational efficiency.
Empowering Developers & Admins
Agentforce is all about making life easier for developers and admins. It mixes pro-code and low-code tools to help you create AI solutions quickly. The magic lies in what they call the DATA framework:
D – Data: Structured & unstructured data + metadata & semantics + connected (Zero Copy, MuleSoft, Open APIs).
A – AI: Predictive & generative AI + advanced RAG + reasoning.
T – Trust: Security + guardrails + compliance & data residency.
A – Action: Automation + Salesforce & MuleSoft APIs + applications & analytics.
Here are some of the key announcements:
- AI-Assisted Agent Builder: Accelerates agent creation with intelligent recommendations for topics and instructions.
- Agentforce API: Agentforce can be integrated into back-end processes, other systems, and directly into applications. It’s a lightweight REST API for embedding agents into any app, workflow, or process.
- Invocable Actions: Agentforce can also be embedded within Flow and Apex. Incorporates AI-driven reasoning directly into Salesforce’s core automation, simplifying complex workflows.
- Testing Centre: Stochastic testing for nondeterministic AI behavior, ensuring agents operate within expected parameters.
- DX Inspector: Offers deep insights into agent metadata and data usage, enhancing governance and compliance.
- Interaction Explorer: Provides detailed analytics for real-time monitoring and continuous agent optimisation.
- VS Code & CLI Integration for Agentforce – Salesforce provides a dedicated VS Code extension for Agentforce development with YAML-based Agent Specification and configuration for repeatability, scalability, and CI/CD integration.
- CodeGen – Powering the Agnetforce for Developers’ Experience. Open Source LLM for Program Synthesis, ensuring trust & security, booking accuracy and minimising latency and resource usage.
- xGen-Code: purpose built for Salesforce use cases with flexibility and adaptability in mind. Used for test case generation and multi-turn chat responses (dev assistant).
- Automated Testing for AI Agents – CLI-based testing allows continuous validation of AI logic.
- Dev Assistant – Multi-turn chat interface for code suggestions & refinements.
- Code Autocomplete – Inline AI-powered Apex & LWC completions.
- Apex & LWC Test Case Generation – Automated test case creation with a new test panel.
- LWC Optimization (Preview) – AI-powered LWC performance & accessibility improvements.
- MuleSoft Integration: MuleSoft Topic Center lets developers use natural language to create topics and actions for agents from MuleSoft APIs.
- Slack Integration: Developers can now integrate Agentforce into Slack automations, allowing it to be triggered by Slack events.
- Tableau: Build structured semantic models that Agentforce can leverage for better reasoning. Tableau enhances agents with real-time business context, enabling more accurate insights and responses.
- Observability & Monitoring: Real-time agent reasoning visibility with alerts and interaction explorer.
- Variables & Filters: More structured agent logic to improve reliability and control.
- DevOps Integration (Beta in Summer ‘25) – Allows agent tests to be triggered automatically in CI/CD workflows. Works with DevOps Center, GitHub Actions, and partner tools (Copado, Tricentis, etc.).Supports automated validation via Code Analyzer & Quality Gates.
- RAG for LWC (Summer ‘25) – Generate code using Agentforce, grounded in your local LWCs as well as sObjects and Apex.
- Agent Versioning – not quite QA but March-April. Makes perfect sense and is very much needed.
- Bring your own LLM – no-code Model bilder, BYOM and BYO-LLM all GA now.
- 2GP Packaging for AgentExchange: Actions, Topics and Prompt Templates can now be packaged in both 1GP and 2 GP.
Salesforce is making Agentforce and Data Cloud available in Developer Edition:
- 150 LLM outputs per hour
- 10 GB of data
- Persistent environments that never expire (as long as logged into every 45 days)
This enables developers to experiment, build, and deploy AI agents without constraints.
AgentExchange: Salesforce’s AI Agent Marketplace
Salesforce introduced AgentExchange, a new marketplace that makes rolling out AI way faster with ready-made agents and actions. If you think about AppExchange’s growth, it’s rather impressive – over 700,000 app installs in the last year, and more than 91% of Salesforce customers use AppExchange solutions. Now, we’re in the middle of this huge shift with “digital labour”, which is a $6 trillion opportunity. So, AI is naturally a top priority for CIOs.
That’s where AgentExchange comes in. For partners, the big news is that actions, topics, and templates are now both 1GP and 2GP ready. For customers, the vision is that you’ll be able to try and buy partner-created actions directly inside Salesforce. So, when you’re customising your agent, you can grab, say, a DocuSign action and plug it right in.
Christophe Coenraets summarises the advantage: “If you’re building an agent from scratch, you’re doing it wrong.”
What’s next? Where do you start?
Salesforce’s global vision is all about building a community of over a million AI specialists, aiming to meet the rising demand for intelligent automation. At a high level, when you’re designing your strategy, think about the “trinity” of Data, Apps and AI:
Data – Business, structured, unstructured, and real-time event data.
Apps – APIs, enterprise applications, analytics, and integrations.
AI Agents – Autonomous systems that reason, act, and learn.
Here are our Top 10 Best Practices for Building with Agentforce – it will be a good starting point for you and your organisation. But let’s be real: with so many announcements and new features, it’s easy to feel a bit lost. That’s where working with a partner like Aquiva comes in.
We’ve been deep into AI since Einstein’s early days, working on everything from predictive tools to Agentforce. So, whether you’re a Salesforce ISV partner or customer looking to get the most out of Agentforce or Data Cloud, we’ve got the experience to guide you. If you’ve got questions or need a hand, just reach out.
Author
Greg Wasowski
SVP, Consulting and Strategy
