Salesforce: Driving Trust in Agentic AI with robust integration powering Enterprises

Salesforce has been known Worldwide with their CRM having intricate, very complex but very intuitive solutions that gave companies the power of connecting various solutions into one platform. Most of the Top Companies were enrolled. Now, with the power of Agentic AI, it has opened to more possibilities bringing more solutions that can be offered to the clients with a single dashboard with automation that syncs flawlessly.

Models are brilliant, but they cannot run the enterprise by themselves.

• Models are probabilistic, which is great for reasoning but not business execution.
• Probabilistic reasoning interprets language, weighs possibilities, and proposes a next step. Its response can vary. Deterministic execution follows defined rules: with the same inputs and rules, a calculation or approval check produces the same result.
• Enterprise AI needs both. The model helps understand what someone wants; business systems determine what is permitted and execute the authorised action.
• Salesforce provides the deterministic foundation that turns AI reasoning into safe, reliable action: the right data, the semantic model to understand it, the tools to act on it, and the trust and governance layer that wraps around it all.

AI is creating an interface revolution.

• Every major era of computing has brought a new interface — from the command line to graphical interfaces, web, and mobile.
• AI is creating the next interface revolution.
• Instead of navigating fixed screens and menus, people can simply describe what they want to accomplish.
• The interface can become dynamic, intelligent, and composable around the person and the task.
• Salesforce can now be recomposed into these new AI experiences.

AIforce is your live interface.

• At Dreamforce, we announced AIforce, our live interface that unlocks the data, workflows, business logic, actions, and governance inside Salesforce and makes it available to any agent, model, and interface.
• For the first time, businesses can access the full intelligence of their business—their data, apps, and agents—from any interface they already work in, including tools like Claude Cowork, without ever logging into Salesforce. 
• So what does this live interface actually look like?
  • If you work in Claude Cowork, Claudeforce is for you.
  • If you work in Slack, Slackforce is for you. 
  • And if you work in Salesforce Lightning, Agentforce Coworker is for you.
  • And there’s more coming soon.
• What makes it possible for AIforce to show up everywhere you work is our architecture for the agentic enterprise.

Agentforce: pre-built agents and the platform behind them

• Building agents is a new skill — especially when working with probabilistic systems.
• So we’re packaging more of our expertise into ready-to-use agents that business teams can deploy faster.
  • We have Hunter, our Outbound Sales Agent; Piper, the Inbound Pipe Gen Agent; and Casey, our help agent, which we use on help.salesforce.com, resulting in $100M in annual savings. 
  • Paige is your IT & HR service agent, giving employees always-on help for common IT and HR needs and making internal service faster and easier.
  • Marshall is a Supply Chain Agent, and Fin is our Customer Service Agent.
• This means fast time to production and business value for organisations.

Introducing Salesforce’s Enterprise AI Harness

• • Trust is the most important requirement for AI transformation.
  • Salesforce’s Trusted Enterprise AI Harness provides a consistent foundation for every agent to act with confidence—connecting them to the context, actions, models, security, and governance needed to operate across the enterprise.
  • Brings together 6 capabilities + AI Control Plane to power the agentic enterprise 
    • First, the intelligence itself: the models. The harness gives models the context they need to understand the business while letting enterprises maintain control over how they use them.
    • Next: Trusted Context. Trusted context understands the customer’s business context: their history, products, contracts, interactions, and how work actually gets done.

• Then comes Trusted Agency: giving agents the ability to reason, plan, collaborate, and coordinate other agents and tools, while combining probabilistic reasoning with deterministic control when enterprise rules need to be followed.

• • • Trusted Action connects that intelligence to applications, APIs, agents, and workflows so AI can actually do the work.
    • Trusted Governance helps ensure answers are based on trusted information and that the agent follows the policies governing how the order should be handled.
    • Trusted Security applies identity, permissions, privacy, data protection, and runtime security to what AI can access and what agents can do.

• The AI Control Plane gives enterprises a way to discover, identify, govern, observe, and optimize AI across the enterprise, including usage and cost across models and agents.

Q&A:

1. Last year, you described the PH as your fastest-growing Southeast Asian market. Has this pace continued? Can you share a growth figure? Also, how many local customers have moved into live AI agent deployments?

• The Philippines is a super important market for us. We opened our office there a year ago, and during that time, it has expanded both physically in terms of space and in terms of headcount. Every time I visit, there are more people and further expansion.
• The Philippines is ripe for agent transformation. Many companies are bringing us their most complex, difficult problems—such as effectively and efficiently serving large customer populations—and AI technology is opening the doors for them to make seismic shifts within their businesses.
• We are seeing tremendous interest and excitement as we work with large Philippine enterprises to define their AI roadmaps. We have also shifted our delivery model to prove value quickly. Traditionally, software implementations involved long presentations, demos, and lengthy processes. Today, companies want fast time-to-value, clear results, and rapid experimentation. Moving past the era of science experiments and proofs-of-concept (POCs), businesses now demand to see how these solutions directly impact their key performance indicators (KPIs). Working closely alongside customers on the ground, our engineers rapidly build out these agents, tackling agentic use cases and demonstrating measurable value within just a couple of weeks.

2. Given high digital participation and large consumer bases in markets like the Philippines, how are local enterprises shifting their view of AI? Are they moving past simple efficiency gains toward using agentic AI for actual revenue growth and scaling?

• They are. Companies are looking beyond bottom-line savings to see how AI can dramatically boost productivity and unlock human capacity. However, there is still a significant amount of low-hanging fruit.
• Previously, many organizations attempted overly complex agentic use cases and pilots, leaving roughly 90% of them stuck in demo or non-production environments. Today, companies want fast, tangible value. Out-of-the-box agents like Casey, Finn, and Hunter are proving highly effective because they can be implemented quickly and linked directly to baseline KPIs.
• CFOs and COOs are increasingly insisting that AI expenditures be justified by clear, trackable benefits. The most accessible low-hanging fruit remains employee productivity and customer service. For example, by migrating customer support to an agentic model using Casey and Finn, Salesforce has achieved $100 million in annualized savings. Direct bottom-line savings linked to clear KPIs provide the essential justification for future AI investments, proving initial value before expanding operations into revenue growth and scaling initiatives

3. In your opinion, at what stage do you think the PH is in AI adoption? At a CXO Summit just last month, an executive mentioned that the AI bubble doesn’t exist in the PH because there’s still no bubble to burst. Do you have a bearish or bullish outlook for AI adoption in 2027? But also, does adoption mean job cutting?

• I hold a very confident and bullish view on AI and its power to dramatically transform how we work. Speaking from personal experience over the past 12 months, AI has fundamentally reshaped my daily routine. I spend significantly less time researching, drafting documents, clicking through systems, and completing administrative tasks. Consequently, I can focus on high-value priorities—spending time with customers, coaching, and mentoring my team.
• Looking ahead to 2027, tools like AIforce mark a revolutionary shift in software interaction. AI adoption does not translate into job cuts; rather, it provides an opportunity to realign human effort away from low-value administrative work.
• To illustrate: I frequently meet with up to 10 C-level customers a week. Previously, preparing a comprehensive briefing document for each meeting—gathering account history, open opportunities, support issues, executive profiles, and talk tracks—took 4 to 5 hours per meeting, totaling 20 to 30 hours a week across the team. Today, using Slackbot, we generate a comprehensive briefing, talk track, key messages, and initial presentation slides in just 3 minutes. This represents a massive productivity boost that frees people to do meaningful human work

4. How would the system work for teams that currently don’t use AI, and how would you walk businesses through getting started on their journey?

• Most companies are already using AI in some form, though often uncoordinated or through personal consumer tools. The fundamental limitation of consumer AI tools (like standalone ChatGPT) in an enterprise environment is that they are not connected to trusted corporate data, such as your CRM or core business applications. While public tools can summarize static documents or search the web, they lack internal business context.
• This is where Salesforce and AIforce provide immediate value by connecting AI directly into trusted enterprise data sources. For organizations looking to start, solutions like Agentforce Coworker can be enabled almost instantly. Rather than requiring a six-month deployment, companies can just turn on the switch, allowing teams to transition from manual menu navigation to natural language interaction in just a few weeks. It offers a scalable, secure, and manageable enterprise environment grounded in trusted business context

5. How is Salesforce driving trust in agentic AI and helping companies address concerns about agents making errors or taking unauthorized actions?

• Trust has been the single biggest issue with AI from the beginning. When using consumer-grade AI, users cannot verify where answers originated or whether the model hallucinated. Organizations cannot risk deploying ungrounded models directly to customers without verified accuracy.
• Salesforce overcomes this through Data 360 and the strategic integration of Informatica. Informatica establishes complete data lineage, providing full auditability and traceability back to source records. For instance, if an AI reports a revenue figure of $427,000, data lineage allows you to trace the exact underlying components back to their original transactions. Combined with the Enterprise AI Harness, this architecture supplies the security, governance, and contextual guardrails needed to ensure AI operates on trusted data without hallucinating.

6. To clarify, what is the source of the 90% pilot failure rate figure mentioned earlier, and does it refer to global or Philippine deployments?

• Please check the reference for the figure here: https://www.salesforce.com/news/stories/why-ai-pilots-fail/
• The basic premise—and I went through this myself a few years ago—is that many companies just wanted to implement AI as a technology. CEOs, directors, and chairmen were saying, “We need to do AI projects; why are we not doing AI projects? Let’s do more AI projects”. What happened was, because you didn’t have the trust there and didn’t have the ability to link to trusted sources with the right context, many companies developed AI that wasn’t linked to trusted sources. So you had AI that was hallucinating, which you couldn’t necessarily trust in terms of the results coming back. You had AI that wasn’t adopted and wasn’t really linked into business cases or key problems.
• As I mentioned before, we’re now changing to customers talking about, “Show me the real value, show me the KPI, what is the baseline of that KPI, what is the value that AI can deliver, and how do I measure and track it?” We’ve progressed from playing and experimenting with the technology—and I think the technology has been proven to work. Where we are now is finding opportunities to prove value, because companies are really looking for AI to start delivering. Many years have passed playing around with things in labs, and businesses are now saying, “Why don’t we have this in our hands, and why aren’t we doing this?”