Dreamforce 2025 Day Three: vision, agents, and responsible innovation

Day Three at Dreamforce 2025 closed out the conference experience with some of the biggest ideas, the sharpest warnings, and the boldest promises for how the agentic enterprise will reshape business. In this post, we’ll help you catch up with the key highlights of the day.

Today’s sessions reinforced that AI agents aren’t just futuristic dreams — they’re being deployed now, and the real challenge is scaling them with trust, orchestration, and accountability.

More vertical takeaways

Yesterday saw even more keynote sessions covering industry verticals like Energy and Utilities, Tech, Non-profit, and Media and Entertainment. With Agentforce use cases being demonstrated across this wide variety of verticals, there were some key thematic takeaways spanning across many of the sessions:

• Sustainability and ESG meets AI: Salesforce leaned into how agents and Net Zero Cloud can power carbon reporting, climate workflows, and supply-chain transparency, aligning innovation with values.
• Cross-cloud synergy: We saw more integration narratives — Slack as an agentic OS, Data 360 underpinning real-time activation, and agents deployed across multiple Clouds.
• Ethics, governance, and risk: Many sessions were grounded in the reality that agent models bring new hazards — bias, misalignment, lack of interpretability — and that governance must be baked in, not bolted on.
• Iterative adoption: Customer stories and demos emphasized that no org starts “perfect” — growth is iterative. The message: pilot, learn, scale with guardrails.

Cloud keynotes and more

IT Keynote

The IT Keynote focused on what it really takes to build and govern the Agentic Enterprise, showcasing Pearson as a customer success story. Salesforce leaders showed how IT teams can now connect, monitor, and manage both apps and AI agents within a single, trusted framework — introducing new tools for observability, multi-agent collaboration, and governance to keep innovation safe and scalable under the new Agentforce IT Service. A major theme was lifecycle management: from planning and building, to deploying and auditing, teams are getting a clearer blueprint for how to scale intelligent systems responsibly across the enterprise.

Marketing Keynote

From the Marketing side, attention turned to agent-augmented engagement: predictive outreach, automated escalation workflows, continuous optimization, with the announcement of Agentforce Marketing (Marketing Cloud rebranded). Salesforce product leaders and customers like Indeed showed how AI agents can scale marketing by making every campaign more conversational, personalized, and adaptive. They laid out a marketing maturity model — from predictive to custom models to full agentic models — and demonstrated how to turn channels into two-way dialogues using Agentforce 360 and Data 360 integrations.

Sales Keynote

In the Sales Keynote, Salesforce rebranded Sales Cloud as Agentforce Sales, unveiling AI-powered prospecting and engagement that prioritizes high‑value accounts from signals across CRM, calls, email, and the web, drafts personalized outreach, syncs with Slack for collaborative follow-up, and handles inbound from service to booking meetings. The platform is bolstered by the Bluebirds acquisition and new partnerships with ZoomInfo, 6sense, and Moody's, with customer examples like Uber for Business showcasing impact. Most features are available now, with advanced prospecting rolling out in February 2026.

Service Keynote

Similarly to the Marketing and Sales keynotes, the Service Keynote rebranded Service Cloud to Agentforce Service. The rebranded solution — including Command Center for Service, Agentforce Service Console, and Agentforce Service Setup — will combine human agents and AI to deliver proactive, personalized support and improve CSAT while lowering costs across channels. They also featured customer stories, including PepsiCo, Southwest Airlines, and Gaggle, to demonstrate how real organizations are already applying agent-augmented service to transform their operations.

Customer Success Keynote

The Customer Success Keynote shone a spotlight on how leading organizations are already realizing the value of the Agentic Enterprise. Speakers shared practical ways teams are scaling faster, cutting costs, and driving customer loyalty through responsible AI adoption — with live examples from this year’s Customer Success Award winners, Asymbl, Engine, and Lennar. The session’s core message was clear: humans and agents working together create the biggest impact, and now’s the time to move from pilot projects to scalable, enterprise-wide success.

Salesforce on Salesforce Keynote

In this internal spotlight, Salesforce shared how they themselves are using agentic systems across support, internal operations, and product development. The keynote underlined:

• Agentforce as backbone: Agents are being deployed across multiple internal teams — from service ticket triage to internal process automation — not just in isolated functions.
• Governance-first mindset: Salesforce echoed that audit logs, fallback paths, human oversight, and transparency are foundational to their architecture.
• Roadmap signals: Expect deeper orchestration capabilities (multi-agent coordination), expansion of core agent templates, more connectivity across Clouds, and smarter change-control frameworks.

This was a full-circle moment: Salesforce is not just preaching the agentic future; they’re building it for themselves.

Salesforce and Google

One of the most anticipated moments of the day was the in-conversation between Marc Benioff and Sundar Pichai, where they dove into how Salesforce and Google see the evolving frontier of AI agents, innovation, and enterprise impact. Here are some of the highlights and key themes from the session:

• Reframing AI as co-author, not replacement: The tone of the conversation leaned heavily into synergy — AI agents should extend human capability, not supplant it. Benioff and Pichai both emphasized how the future lies in humans and agents teaming up to unlock productivity, creativity, and new business models.
• Speed of innovation outpacing adoption: Benioff underscored a recurring tension — the pace of AI breakthroughs is far ahead of most organizations’ ability to safely and effectively adopt them. He urged leaders to build with deliberation, with guardrails, and to prioritize meaningful adoption over flashy launches.
• Google’s AI journey and lessons from OpenAI’s breakthrough: Pichai reflected candidly on how ChatGPT’s market arrival shifted expectations. He acknowledged that while Google has had deep investments in AI infrastructure, the sudden jump in external attention forced recalibration — a moment of learning around timing, readiness, and trust. 
• Interoperability and model ecosystems: Both leaders talked about how agents will need to operate across platforms, Clouds, and contexts. They see a future in which models, data fabrics, and APIs blend: Google contributes model infrastructure, Salesforce contributes agent orchestration and business context. The idea is that no single vendor owns the full stack — collaboration and openness will matter.
• Ethics, trust, and responsibility as differentiators: Pichai and Benioff converged on the idea that in the coming agentic era, how you build matters just as much as what you build. Transparency, auditability, governance, explainability — all these will be the canonical expectations, not optional extras.

MuleSoft Keynote

In the MuleSoft Keynote, we learnt how MSC Cruises and Barco are shaping the Agentic Enterprise with developer workflows, and governing and orchestrating cross-ecosystem agents with MuleSoft Agent Fabric.

• The emphasis was on composable architecture: agents need to be able to invoke APIs, coordinate services, and move across systems — MuleSoft is how you do that reliably.
• Demos showed agents using MuleSoft to orchestrate multi-step business processes (e.g. order-to-cash, procurement, inventory sync) under the hood.
• Salesforce’s narrative reaffirmed that you can’t isolate agent logic — it must live in a mesh with APIs, business logic, data clouds, and legacy systems.

Security Keynote

Security, trust, and resilience were front-and-centre in multiple sessions, but most notably in the Security Keynote. Here were some of the key takeaways:

• Zero trust and agent authorization: Agents don’t get blanket permissions — you need fine-grained scopes, principle-of-least-privilege, and runtime access checks.
• Audit trails and observability: Every decision, every data read, every hand-off needs traceability so you can reconstruct agent behavior.
• Fallbacks, overrides, dead-man switches: Agents must yield to humans or safe states when uncertainty is high or anomalies are detected.
• Adversarial testing, drift detection, and continuous validation: Models age, contexts change, and guardrails must evolve.
• Regulatory alignment and compliance: For industries such as healthcare, finance, and government, agent decisions may need to be explainable and auditable — you can’t assume opaque black boxes.

Wrap-up: top 5 takeaways from Dreamforce 2025

And so Dreamforce 2025 concludes with grand vision, real momentum, and a sharp reminder: the future isn’t agentic by chance, it’s agentic by design. Salesforce has laid out the roadmaps, shown early proofs, and charged the community with responsibility.

Across all three days, Salesforce delivered a powerful story: from the vision of the Agentic Enterprise on Day One to the wave of innovation and impact explored on Day Two, and finally, the focus on trust and orchestration that brought everything together on Day Three.

After three packed days of keynotes, demos, and deep dives, a few big ideas stood out, not just about where Salesforce is heading, but what it means for DevOps teams building on the platform every day.

1. AI agents are officially part of the fabric

Dreamforce 2025 confirmed that AI agents are being baked into every Salesforce Cloud through Agentforce 360. The focus has shifted from “what’s possible” to “how do we roll this out responsibly and effectively across real teams?”

2. It’s all about orchestration and connection

The magic doesn’t happen in isolation. The sessions made it clear that the real power lies in connecting agents, systems, and data. With Data 360 and MuleSoft acting as the glue, Salesforce showed how orchestration is what turns individual AI tools into coordinated, business-ready workflows.

3. Trust and governance are the make-or-break factors

As excitement builds, so does responsibility. Dreamforce hammered home that every AI rollout needs human oversight, clear audit trails, and strong guardrails. Trust is the foundation for adoption.

4. Values and innovation can (and should) coexist

From sustainability in Net Zero Cloud to the emphasis on ethical design, Salesforce used Dreamforce to remind everyone that innovation means little if it’s not grounded in purpose. Progress only matters if it’s responsible.

5. Adoption takes time — and that’s okay

Even with all the hype around AI, speakers like Marc Benioff and Sundar Pichai were refreshingly honest: technology is moving faster than most organizations can adopt it. That’s not failure — it’s an opportunity to be intentional, experiment safely, and build lasting change.

Keep learning with DevOps Launchpad

Dreamforce might be over for another year, but the learning doesn’t stop here. The pace of change in Salesforce — from AI agents to automation and observability — shows just how important it is to keep upskilling and future-proofing your career. Check out all of our free, expert-led training designed to help you grow your Salesforce DevOps skills and stay ahead of what’s next.