Salesforce is moving from the hype phase of generative AI into doing the harder engineering work to create the agentic enterprise. Dreamforce 2025 showed us that the company which created the original SaaS model now wants to lead what we call service as software. In our view, this represents a profound revolution, not just in technology but in business. In a world where AI agents deliver outcomes across systems of record, systems of engagement and systems of intelligence. Our assessment is that Salesforce’s AgentForce 360 platform is crossing day one early versions to attack day two problems and customer requirements. What does that mean? Its no longer just demoing copilots inside Customer 360. Rather, the company is going after the messy problems of observability, orchestration, data quality, etc. that determine whether agent systems can scale in the enterprise.
This is where the rubber meets the road in our view. The emergence of a new operations layer features critical new functionality like agent observability and analytics. We liken this to the “Datadog for agents,” where the lineage of every decision, every workflow transition and the reasoning steps are tracked for provenance, performance and continuous learning. This learning includes absorbing the reasoning traces of humans. It’s all an extension of Data Cloud and sets up Salesforce to be a software-only agentic hyperscaler.
At the same time, Salesforce is fortifying its data cloud as the foundation of trust and accuracy. We see a new software stack emerging that will evolve from today’s isolated SaaS apps into a System of intelligence (SoI), where data governance and metadata are unified to support agentic execution. This is the day that customers, investors and CxOs have been waiting for. We’re talking about less showmanship and more system architecture.
In this breaking analysis, we will take you through an analysis of and takeaways from Dreamforce 2025. The strategic question that we’ll address in today is whether Salesforce can convert this discipline into a durable advantage and beat the competition before the hyperscalers, other SaaS players and open source ecosystems commoditize the agent platform. We’ll also explore how Salesforce’s open posture can attract new partners. We’ll test our thesis that Marc Benioff is betting on a similar integration playbook that won the SaaS era – unifying apps, data and engagement around a scalable cloud-based operating model – around trust in the enterprise . We believe the outcomes will define not just Salesforce’s future, but the competitive balance between the builders of AI and the operators of agentic Enterprises.
How Quickly Can the Giant Reclaim Its Dominance?
We chose the “King Kong” visual to set the tone because Salesforce led the SaaS revolution. The image (courtesy of Nano Banana, Gemini’s new image generator) is intentional. Salesforce has been the King Kong of applications for decades – dominant, visible and battle tested. Not quite a Gulliver tied down by cords, but undeniably mature after a period where the company rounded out the suite. The real question now is whether that maturity can be converted into a new kind of leadership as the industry shifts from cloud-delivered automation tools to delivering work itself – i.e. business outcomes implemented in software.
We believe this marks the start of a 20-year reset in enterprise software. The center of gravity moves from “sell an app that users operate” to “deliver outcomes that the software executes with human supervision.” That change is not cosmetic and it forces a different operating cadence and proof metrics. Instead of showcasing feature depth, the leader must prove it can orchestrate agents that perceive state, decide, and act across a customer’s environment to produce measurable outcomes. The years Salesforce can be an asset if the breadth of the estate becomes a platform on which outcome delivery scales with less labor. The competitive benchmark is already shown above with Palantir and Celonis, which are, in our view, the only vendors with offerings mature enough today to credibly claim “agent-based Service as Software” – i.e. systems that deliver outcomes, not just SaaS. Their presence sets the bar. We believe the category is real and sets expectations that demos and copilots must meet or exceed. For Salesforce, the edict is to transform platform scale and install base into operating leverage for customers – i.e. faster paths from intent to result, fewer handoffs, and verifiable business impact.
Our premise is that the mantle of “King Kong” in the agentic era will belong to the first vendor that consistently ships outcomes at scale. Salesforce has the surface area but the test is whether it can convert its current maturity into a new velocity and prove that Service as Software – not traditional SaaS – defines its next act..
Service as Software Must Integrate 60 Years of Silos
The slide below frames the following question: Salesforce led the SaaS revolution; can it lead the next one – and how long will it take? The graphic walks through the familiar on-prem → IaaS/PaaS → SaaS progression and then introduces a new “green zone,” the high-value real estate – the System of Intelligence – where the next layer emerges.
We believe this transition is profound because everything that changed over the last 50–60 years – interfaces, operating models, architectures – never eliminated fragmentation. Each wave produced more silos – e.g. silos of work automation inside applications and silos of data, even inside “single” data lakes, data across company divisions is not harmonized. The net effect is a patchwork where outcomes require brittle handoffs across tools, teams, and schemas. The green zone shown above is a potential response – i.e. an abstraction layer above apps and data that enables end-to-end work. In our view, this is not just another middleware tier; it is a system that perceives business state across sources, reasons about what should happen next, and then acts, closing the loop with governance and measurement. The incumbents that dominated prior layers can reclaim dominance only if they capture this high value piece of real estate quickly, because it is where value concentrates. In other words, whoever orchestrates outcomes across silos owns the control plane for enterprise work. Time is of the essence in our view. Pilots that stall in glue code will cede way to platforms that turn intent into repeatable results. The “managed services” vendors of prior eras optimized pieces of the stack, while the green zone integrates the pieces into a continuous outcome path. That marks a shift in our view from selling tools to selling work.
Our premise is that the players who win the green zone – abstracting silos into end-to-end, governed outcomes – will define the agentic era. Speed to credible, at-scale execution in this layer will determine whether past leadership translates into future dominance.
From SaaS to Service as Software: The Next Model Shift
As described below and in a prior Breaking Analysis argued that the on-prem → SaaS transition rewired everything from the technical architecture, the operating model, and the business model (including pricing). We believe the industry is at an equivalent inflection, moving from SaaS to Service as Software (SaSo). This time the shift extends beyond tech vendors and IT; software-like economics will flow to every company that digitizes certain expertise and delivers it as outcomes.
What changes now is not just where software runs but what software delivers. In SaSo, vendors provide end-to-end outcomes – legal resolutions, call-center performance, wholesale banking services—executed by governed agents. That forces simultaneous change across three planes as described above. Three implications surface in our view:
- Marginal costs decline because more of the expertise is in software.
- Differentiation increases as systems learn from every execution.
- The result is winner-take-most dynamics—simultaneously lowest cost and highest differentiation.
The implication is that firms can scale without proportional labor growth. That is the economic engine of SaSo. Vendors and enterprises that operationalize digitized expertise as software services will capture outsized value; those that remain feature-or app-centric will see value pool migration to the outcome orchestrators.
Our view is that SaSo extends the SaaS-era playbook in that a full-stack transformation of tech, ops, and business models results. The winners in our view will be those can progressively convert human expertise into governed, observable software services that learn with use and unlock declining marginal costs at volume and compounding differentiation at production scale.
The Emerging Service-as-Software Stack: System of Intelligence as the Key Enabler
Above we show our view of how the software stack is evolving. The green band is the high value System of Intelligence (SoI). We’ve discussed this repeatedly as a “4D map” of the enterprise. This section attempts to explain why that layer is different from conventional software platforms and why agents deliver declining marginal cost at scale.
Traditional data platforms (the Databricks and Snowflakes of the world) remain an essential contributor as they capture transaction snapshots and event tables streaming out of operational apps. But they generally excel at what happened. The SoI re-shapes those raw events to reflect the dynamics of the business in terms of causal relationships, constraints, rhythms, and allowable actions across time. In our view, the SoI extends the outcomes by addressing what happened, why it happened, what is likely to happen, and what should be done. That makes it a dynamic, four-dimensional map – state + time + causality + policy – rather than a passive store.
Agents require this mapping. Similar to robots navigating a plant floor, enterprise agents must orient (perceive the current state), reason (infer intent and likely outcomes), and act (invoke governed tools) and absorb human feedback. Without a living model of the business, agents either overfit to narrow workflows or devolve into prompt chains, which are brittle. With the SoI, agents inherit shared context, policies, and goals, so actions become consistent, auditable, and improvable.
This is how we see the stack evolving, not merely adding a tier. Systems of Record remain the singular truth; Systems of Engagement remain the human/agent interface. The SoI sits in between, blending data semantics, business logic, and policy into a control surface that agents can use. Governance, metadata, and observability weave through all three layers so every perception, decision, and action is traceable.
We believe the SoI is the high-value real estate because it turns disconnected data and apps into a navigable, causal map that agents can operate on. That is the prerequisite for Service as Software in our view, where outcomes are executed by agents that perceive, decide, act, and learn across the enterprise with trust and control.
Mapping the Stack to Agentforce 360
This next section maps the emerging stack to Salesforce’s Agentforce 360 as shown below. Our view is that a system of agency is much more than an individual agent embedded in Sales, Marketing, or SDR/BDR workflows. It is the agent inside the app plus the governance fabric, observability, testing, and control surfaces that make fleets of agents safe, auditable, and scalable. In this model, the System of Intelligence (SoI) – the green layer – connects directly to Data 360, which is not merely a data store but a living model of the customer profile and engagement journey. That model sits atop a conventional data platform. MuleSoft provides the governed bridge back into systems of record to operationalize decisions in legacy applications.
We believe Salesforce’s real play is integrating a system of agency to a system of intelligence so agents can perceive, decide, act, and learn across Customer 360 with provenance. The narrative on stage in the keynotes at Dreamforce emphasized new agent development features, but the durable moat forms in a new operations layer that integrates and extends the SoI – this is the scaffolding that sustains Day-2 scale, specifically:
- Governance and policy enforcement across all agents and actions
- Observability and reasoning traces for provenance and continuous learning
- Testing, versioning, and rollback to harden workflows before broad rollout
In our opinion, this is where differentiation will emerge. The market is flooded with agent runtimes and dev tools – “more startups than fleas on the average camel” – George Gilbert. Most runtimes look similar; what actually separates agents in practice are the tools they are natively trained to use and the context they inherit. In Salesforce’s world, “tools” mean governed actions inside Customer 360 (the verbs that change state) and data from Data Cloud that lets agents perceive the business or customer state accurately. The combination of trusted actions and contextual state is what gives one agent meaningful advantage over another.
Notably, the keynote spotlighted builder and scripting features but was lighter on the “Datadog-for-agents” side of the story. Our assessment is that Salesforce is folding those operations primitives into the System of Intelligence itself. That integration matters because by co-locating observability with the SoI, perception, decision, and action is captured once, governed once, and fed back into the learning loop. That is how fleets of agents avoid brittle prompt spaghetti and move from pilots to production.
We believe Agentforce 360’s edge will not come from yet another agent runtime but from scaffolding software – i.e. governed actions in Customer 360, Data 360’s business model of the customer, MuleSoft’s policy-aware integration to systems of record, and SoI-native observability and testing. That is the path to reliable, at-scale outcomes and the essence of Service as Software.
Salesforce Decision Intelligence
We believe a core enabler is a semantic data fabric in Salesforce 360 that converts enterprise data into dynamic knowledge graphs. In our view this provides crucial context the agents need. Our research indicates demand is real. For example, as shown above, in a 625-respondent, 13-industry survey, 73% plan to invest in reasoning/decision intelligence within 18 months; 36% call it strategic. Today, Einstein’s “what-if” simulations are largely probabilistic and correlation-based, which are useful, but brittle as conditions change. Service as Software requires more in our view. Specifically, we call for semantic grounding plus causal models that explain, justify, and predict consequences.
As shown below, around 45% of respondents in our survey indicate intent to invest in chain-of-thought reasoning, knowledge graphs, and causal AI. In our view, Salesforce has the pieces (agentic architecture, semantic layer, multi-model ecosystem) but must execute to turn them into trustworthy, repeatable decision intelligence at scale.
Systems of Agency Are More Than Agents
Returning to the emerging Service-as-Software stack, this section isolates the system of agency and clarifies what actually differentiates agents in production. The market is overflowing with agent runtimes and dev tool startups—there are, as we like to say, more of them than fleas on the average camel. Most look similar at the core.
We believe the meaningful differentiation is not the agent shell; it is the native tools and context the agent can wield with governance. In Salesforce’s world, “tools” are the actions inside Customer 360 that actually change business state – create, update, approve, fulfill, escalate – and the Data Cloud context that lets agents accurately perceive the state of the customer and the business. Agents trained to use these governed actions against high-fidelity context can operationalize decisions end to end; agents without them remain clever prompt chains.
Equally important is the operations layer. The keynote emphasized builder and scripting features, but the decisive capability is what we’d call Datadog for agents – observability, testing, reasoning traces, and policy controls – now being folded into the System of Intelligence. By embedding operations into the SoI, every perception, decision, and action is captured once, governed once, and fed back into the learning loop. That is how fleets of agents avoid brittle behavior and scale safely.
Our view is that production-grade agency equals agents + scaffolding. The winners will pair:
- Governed actions in Customer 360 (the verbs that change state)
- Rich context from Data Cloud/SoI (the map that orients decisions)
- SoI-native observability and testing (the feedback that hardens workflows)
That combination, not yet another agent runtime, is what turns pilots into reliable, scalable outcomes.
The Developer Perspective
The next section of this post drills into the developer angle. We’ll start with Paul Nashawaty who runs our AppDev practice and get his take on Dreamforce and what it means for developers. After that we’ll rifle through the other announcements with quick hits on the announcements – i.e. 1) What it is; 2) Why it matters; and 3) A bottom line takeaway for each.
The Developer Angle of Agentforce
[Watch Paul Nashawaty’s Complete Analysis]
Dreamforce 2025 marked a shift from app-centric SaaS to the agentic enterprise – i.e. Service as Software. We believe the center of gravity moves from building apps to operating systems of action where agents, grounded in Data Cloud (now called Data 360) and governed workflows, deliver outcomes and learn continuously. Einstein, Data Cloud, and Agentforce are converging into a unified code-and-data fabric; developers shift from writing features to architecting observable, policy-controlled agent workflows.
Heroku is evolving from a developer-friendly PaaS into the agentic stack with open development with enterprise governance, hosting microservices and models, and extending Agentforce. MuleSoft’s “intelligent integration fabric” becomes the nervous system, with a single control plane (e.g., Anypoint Flexible Gateway) for authenticated, observable APIs across systems.
Core benefits Salesforce has promised are as follows: Less glue code, faster time-to-value, hybrid low-code/pro-code velocity. In our view, the risks Salesforce needs to manage that customers will focus on include the following: Ecosystem lock-in, operational complexity, and the need for rigorous observability, security, and governance. In our view, performance, transparency, and repeatability are critical as Service as Software becomes the operating model for enterprise development.
Developer Feature: Agentforce Builder
- What it is: A simplified, low-code experience—plus conversational UI—for business analysts and AI engineers to assemble agents.
- Why it matters: We believe it shortens time-to-pilot and expands the builder pool without sacrificing governance when paired with Data 360 actions and policies.
- Bottom line: In our view, Builder accelerates creation; Day-2 value still depends on observability, testing, and governed actions inherited from the broader Agentforce fabric.
Developer Feature: Agent Script
What it is: A way to inject deterministic steps into otherwise probabilistic agents—at the task, process, or multi-agent orchestration level.
Why it matters: Our view is that guardrails, branching logic, and required checkpoints reduce variance, improve reliability, and make agents enterprise-safe.
Bottom line: Determinism layered onto adaptive agents is how teams move from clever demos to repeatable, auditable workflows.
Developer Feature: Agentforce Vibes
What it is: “Vibe coding” brought into Salesforce—using conversational prompts to generate and modify experiences within Customer 360 data, workflows, and agent tools.
Why it matters: Our view is that moving vibe coding from greenfield apps to governed enterprise context makes rapid UX/app iteration practical without abandoning control.
Bottom line: We believe Vibes turns prompt-first creation into production-shaped development by anchoring it to Salesforce data models, actions, and policies.
Developer Feature: Intelligent Context
What it is: An intent-aware context engine that automates chunking, embedding, and retrieval based on the question being asked.
Why it matters: We believe it reduces hallucinations and admin toil by selecting the right spans of text and signals for each query – rather than relying on brittle, manual chunking rules.
Bottom line: In our view, adaptive context construction is table stakes for reliable agents; it turns generic RAG into fit-for-purpose evidence at run time.
Developer Feature: Agentforce Voice
What it is: Voice-native interaction for Salesforce apps and agents—usable in contact centers and by employees to work with CRM hands-free.
Why it matters: We believe it turns after-hours form filling into real-time capture; reps can log activities, update records, and trigger actions by talking.
Bottom line: In our view, making voice a first-class modality boosts adoption and data quality while extending agents to where work actually happens.
Agentforce Service + Command Center
What it is: Salesforce’s flagship Customer Service app agent-enabled with a centralized command center to monitor, govern, and tune a fleet of agents across Systems of Record.
Why it matters: We believe this is the bridge from pilots to pervasive agents – giving ops teams levers for policy, routing, escalation, and performance while keeping actions grounded in Customer 360.
Bottom line: In our view, embedding agency into a core app plus fleet management is the Day-2 pattern enterprises need to scale outcomes safely.
Admin Feature – Observability: From POCs to Operating a Fleet
Last year was heavy on POCs and vision; this year Salesforce put Day-2 operations on the table. The first admin feature that matters is observability – because once agents are deployed, teams must see what they are doing, how well they are doing it, and why.
Traditional analytics focuses on business state – e.g. pipelines, CSAT, resolution times. We believe the agentic era adds a second analytics plane: metrics about the agents themselves. That includes health, policy adherence, tool usage, success/failure modes, latency, escalation patterns, and outcome quality. Crucially, this is not a sidecar; observability must be converged with Data 360 so agent telemetry is correlated with customer, case, and transaction context. With that integration, admins can move beyond “what happened” to “how the agent decided,” “which actions it took,” and “what should change.” The orchestration view that accompanies observability is equally important: it treats agents as a fleet (or even an army) that can be routed, throttled, or retrained based on live signals. That closes the learning loop—telemetry informs tuning, tuning improves outcomes, outcomes feed back into models and policies.
Our view is that observability fused with Data 360 is the Day-2 control surface. It upgrades BI from tracking the business to tracking the agents that run the business, enabling scalable governance, faster root cause analysis, and continuous improvement of agent behavior.
Observability, Part 2: Aggregate to Trace, Human to Agent
The second observability view matters because admins need two lenses at once: fleet-level aggregates across many sessions for a given agent type, and instance-level traces – including both human and agent reasoning traces – for a single session.
We believe this dual view is the difference between dashboards that describe and systems that improve. At the aggregate level, teams can spot cohort patterns (tool failure rates by segment, drift by region, latency hot spots, escalation frequency, containment rates), set baselines, and run A/B tests on scripts, prompts, or policies. At the instance level, reasoning traces expose how an outcome was reached: what the agent perceived from Data 360, which actions it invoked, why it branched, and where a human intervened. That combination enables:
- Rapid root cause analysis (policy mismatch vs. data quality vs. tool error)
- Compliance and auditability (who/what decided, with provenance)
- Targeted retraining (feed specific failure modes back into the SoI)
- Safety and quality gates (block risky action patterns before rollout)
Crucially, both lenses must be bound to customer and case context so “what happened” lines up with “how the agent decided.” Without that stitching, aggregates are blind and traces are anecdotes.
Our view is that observability only earns its keep when teams can zoom from fleet metrics to a single reasoning path in one motion. That is how enterprises tune agents continuously, prove governance, and scale from pilots to dependable, production outcomes.
How CXOs Achieve Strategic Outcomes Across
Our closing slide below synthesizes the stack and brings together harmonized data and governance; Systems of Engagement at the edges; a System of Intelligence inferring context; and, in yellow, a System of Agency taking action through an agent control layer. The aim is to have agents delivering outcomes across silos so CXOs can hit their most strategic objectives.
We believe the vision is credible because it defines who does what at each layer and measures success as outcomes, not app usage. The desiloed posture assumes data platforms participate, engagement systems provide signals, and governance is enforced end to end. Within this frame, competitive dynamics center on which vendor can make processes first-class and operate them with agents at scale:
- Out in front: Palantir and Celonis treat business processes as first-class citizens with a four-dimensional map of how work flows. They prove the category’s viability and set the production bar.
- Aspirants: Blue Yonder (building on RelationalAI with deep supply-chain domain), ServiceNow (extending a workflow data fabric toward the same goal), SAP Business Data Cloud (entity-centric today – customers, orders, POs – moving toward explicit processes).
- Open question: Oracle has hardware-scaled systems and separation benefits in owned stacks, but, in our view, has not yet shown a credible data-platform entry in this framing that rivals Databricks/Snowflake nor the pull of application logic into the data plane the way Salesforce is attempting via Data 360 + Agentforce.
For Salesforce specifically, our assessment remains constructive. The 2030 financial framework targets $60B organically (before any Informatica implications). Data Cloud 1.0 was rudimentary; this week showed real progress – misfit parts starting to cohere around a governed SoI and an agency layer. The principal risk in our view is organizational. Specifically, Salesforce’s muscle memory and street cred are as an application company. Winning the agentic era requires a hybrid identity – i.e. apps + platform – and a two-front go-to-market targetting board-level business value and credibility with the IT organization that builds cross-functional agent pilots and runs Day-2 operations. The new IT Service offering is strategic not just as product, but as a conduit to earn trust on the IT side of the house. Ironically, we noted during Marc Benioff’s keynote, the synergy between Michael Dell (IT cred) and Marc (line-of-business affinity).
We believe this will be enterprise tech’s greatest era as experience-curve economics move from consumer online to every company in every industry. The winners will unify apps, data, and engagement under a governed SoI, operate fleets of agents with observability and testing, and prove repeatable outcomes at scale before hyperscalers, SaaS peers, or open-source ecosystems commoditize the platform.
The straightforward mandate for CXOs and boards of directors is get on the curve, start learning and dominate.
