Part IV — The New 'No Software' Moment: Salesforce and the Birth of an Agentic Era Species
In this latest SaaSpocalypse analysis, Salesforce’s TDX 2026 reveals production-scale proof of a shift from SaaS to orchestration, evidence that demands a new framework for the Agentic Era.
The market is miscategorizing Salesforce. Investors are pricing it as a victim of the agentic transition, sending its stock down ~30% year-to-date. That bundles it with the workflow wrappers being commoditized by the AI labs, transforming into platform plays. And yet at Salesforce’s TDX 2026, the company presented production-scale evidence of something structurally different: an incumbent destroying its own application layer, open-sourcing the governance grammar that sits above MCP’s connectivity layer, and building the orchestration infrastructure the Agentic Era demands. And doing it all at a velocity on a context substrate that the labs cannot replicate. Anthropic escapes model commoditization by orchestrating on frontier capability. Salesforce is mitigating application-bypass risk by orchestrating in the enterprise context. The market recognizes the value of orchestration when labs build it, but applies a blanket SaaS discount when an incumbent builds it. That asymmetry demands a more sophisticated framework.
The market believes Salesforce is a casualty of the agentic transition.
Its stock is down about 30% year-to-date, putting it in the same category as workflow wrappers and thin AI applications, which are being commoditized by model companies moving up the stack. From Wall Street’s perch, the company looks like another SaaSpocaplypse victim. It’s the classic, simplistic narrative being applied repeatedly as investors try to come to terms with the complexity of a massive valuation shift.
However, Salesforce’s recent TDX developer conference in San Francisco offered ample evidence for a different conclusion about the company’s position in the Agentic Era.
Salesforce is not being displaced. Instead, it is actively dismantling its own application layer and rebuilding itself as orchestration infrastructure for the Agentic Era.
Consider this statement from Parker Harris, the company’s co-founder. He framed the logic driving the company’s emerging strategy with a question that would have been unthinkable from a SaaS chief executive five years ago: “Why should you ever log into Salesforce again?“
At TDX on April 15, Salesforce unveiled Headless 360, an initiative that exposes every capability in its platform as an API, MCP tool, or CLI command, ships more than 100 new tools and skills, open-sources a new language for defining agent behavior, launches a multi-vendor governance control plane, and unifies over 13,600 apps, agents, and MCP servers into a single marketplace.
Salesforce is not defending its browser-based CRM. It is destroying the browser dependency, open-sourcing the governance grammar, and reconstructing its value in the context and orchestration layers beneath the application. These are not marginal product shifts. It is a category transition.
Of course, TDX 2026 did not occur in a vacuum. It unfolded against a backdrop of weekly detonations from the AI labs that have been systematically ascending the enterprise stack. And so, the market, viewing these announcements through traditional lenses, heard a series of product announcements that caused it to price it with the wrong framework.
The historic parallels of this moment should not be lost. Twenty-six years ago, Salesforce proclaimed, “No Software” and redefined enterprise computing by replacing on-premises applications with a cloud-delivered service. Now the company is no longer the rebel at the gate, but it still must find a way to lead the revolution. TDX demonstrated the company understands what must be done, though it still must execute.
Then, as now, there is a parallel problem at such transitory moments: the lag in markets' ability to recognize and properly value these new paradigms.
And so, using Salesforce as a case study, my primary goal is to ask: which frameworks should investors use to properly assess whether a company is successfully transitioning to the Agentic Era or at risk of becoming SaaSpocalypse roadkill?
SaaSpocalypse decoded
Over the past two months, I’ve written a series of “SaaSpocalypse” articles trying to apply work I’ve done to move beyond the binary debate: Is AI going to kill SaaS?
Part II examined the AI labs’ transition from model companies to orchestration platforms.
Over the course of these articles, this has been one of my core arguments: The SaaSpocalypse is a repricing event, not an extinction event.
As we move into the Agentic Era, durable value shifts from systems of record to systems of action. The real question that needs to be asked: do incumbents have a credible path to make that transition?
Salesforce embarked on this transformation well over a year ago with the launch of Agentforce. It was early to embrace agents, and it clearly understood, as I have long argued, that the real value would come from being the orchestrator.
Since then, the company’s numbers have been as significant as the architecture.
Agentforce’s ARR reached $800M, up 169% year-over-year. Combined with Data 360 and Informatica, the agentic platform now runs at $2.9B in ARR, up over 200%. The platform has processed 19 trillion tokens cumulatively and delivered 2.4 billion agentic work units. Production accounts increased 50% in a single quarter. Over 75% of the top 100 deals included both Agentforce and Data 360. Over 60% of fourth-quarter bookings came from existing customer expansion.
In that context, TDX 2026 was not a radical departure. But rather an acceleration toward the company’s agentic future.
The agentic inflection point
Let’s break down what Salesforce announced. Salesforce unveiled a set of moves that, taken together, redefine the company’s position in the stack.
I. Headless 360: Eliminating the bypass threat
Browser-bypass risk is the vulnerability that allows agents to operate your software through its UI without your mediation. This undermines the structural leverage of a system of record through controlled API access. As UI agents mature, they will not need the incumbent’s permission. They will simply open the application and work.
Headless 360 collapses this risk by eliminating the browser as a mandatory interface. More than 60 new MCP tools and 30 preconfigured coding skills give external agents such as ClaudeCode, Cursor, Codex, and Windsurf complete, live access to the full platform’s data, workflows, and business logic. The platform is restructured around four explicit layers: a system of context (Data 360), a system of work (Customer 360 apps), a system of agency (Agentforce), and a system of engagement (Slack and other surfaces).
Each layer is programmable. None requires a browser.
Our proprietary Architectural Resilience Assessment Framework (ARAF) evaluates how platform companies build sustainable competitive advantages by orchestrating complex ecosystems. In the Agentic Era, orchestration capabilities become the primary differentiator between companies that capture platform value and those that become commoditized service providers.
ARAF scores companies on the spread between Opportunity, their strategic positioning for orchestration, and Risk, their vulnerability to bypass, displacement, and commoditization. The transition from system of record to system of action requires raising technical readiness (Opportunity) and collapsing bypass vulnerability (Risk).
In terms of ARAF scoring, the release of Headless 360 improves both sides of the framework’s core equation simultaneously. Most incumbents can only work on one at a time. Headless 360 does both. It raises Salesforce’s technical readiness, historically the company’s defining weakness, by making the platform API-first, while eliminating the browser bypass threat by removing the browser as a requirement.
And so, rather than waiting for the inevitable browser bypass, Salesforce has overhauled its platform architecture to reduce its dependence on the browser interface. You no longer face the risk of interface bypass if the interface no longer matters.
II. Agent Fabric and Agent Script: The governance grammar
While MCP solves connectivity. It does not solve governance. This is how enterprises control probabilistic agent behavior within deterministic business constraints. That governance gap is precisely where Salesforce’s context advantage becomes structural.
The gap matters because of a hard truth that the industry is confronting in production. Joe Inzerillo, President of Enterprise and AI Technology at Salesforce, framed it this way:
Traditional software is deterministic: one plus one always equals two. AI agents are fundamentally different. They produce different outputs from the same input. Salesforce learned this with painful specificity. They deployed an unconstrained service agent that started recommending competitor software to their own customers. Their agents degraded beyond eight instructions per topic.
Gartner documents a 1,445% surge in inquiries about multi-agent systems, accompanied by warnings about attack surfaces and compounded error rates. Nearly two-thirds of enterprise buyers surveyed by G2 were surprised by the oversight required.
The challenge is not intelligence. The challenge is governance.
Agent Script is the company’s architectural answer. It is a strongly-typed language that defines when agents should use LLM reasoning and when they must follow deterministic logic driven by sub-agents, actions, guardrails, transitions, all specified in structured files. The hybrid engine combines the creative capacity of probabilistic models with the precision of deterministic workflows.
As Salesforce framed it internally: “If you hire a college intern, you do not just point them toward a library and say, ‘Help our customers.’ You give them a script, a rubric, and a set of guardrails.“
Agent Fabric is the runtime. It provides a control plane for managing agents from any vendor with deterministic orchestration via Agent Broker, centralized LLM governance through AI Gateway, and agent discovery through scanners that register agents across platforms, including Amazon Bedrock and Microsoft Foundry.
In the framework I use to assess hub position, whether a platform hosts and coordinates agents or is called by agents, this positions Salesforce at the highest level: an agent marketplace host that governs multi-vendor ecosystems.
The open-sourcing of Agent Script is the strategic move that requires the closest scrutiny.
The logic mirrors Anthropic’s decision to open-source MCP, which did not dilute its strategic position but rather reinforced its advantages and leverage. By establishing the connectivity standard that the ecosystem adopted, Anthropic accumulated structural influence despite the protocol being free.
The lock-in was organizational, not technical. Last summer, I wrote about this pattern as the “11% switching” paradox. Despite model performance convergence and unprecedented ease of technical substitution, only 11% of enterprise builders switched AI providers in the past year. The reason: the orchestration of the tools and intelligence across an organization’s workflow was far more difficult to simply rip out and move to another provider.
Agent Script follows the same logic at a higher layer. It standardizes not how agents connect to tools, but how enterprises define agent behavior, including which decisions are deterministic, which are probabilistic, and what guardrails constrain each step.
Salesforce draws the SQL analogy: the core language is standard, platform-specific dialects extend it. If coding agents learn to generate Agent Script natively, it becomes the default grammar for enterprise agent governance.
But the parallel also carries MCP’s risks.
By early 2026, MCP had come under sustained criticism: security vulnerabilities, concerns about architectural maturity, and the argument that as agents grow more capable, MCP’s intermediary architecture becomes less necessary.
Agent Script could follow a similar trajectory: adoption momentum followed by fragmentation as competing governance languages emerge from Microsoft, Google, or the open-source community. If the language fragments rather than achieving universality, the governance grammar advantage dissolves.
The counter-argument is structural. Agent Script addresses a problem that MCP does not touch. Governance, unlike connectivity, requires an institutional context to function. A generic governance framework that does not know your approval chains, exception patterns, or escalation logic is just another configuration file.
Agent Script populated with Salesforce’s accumulated institutional memory becomes an enterprise nervous system. The context makes the governance layer structural rather than substitutable.
III. Expanded AgentExchange: the marketplace gravity
Salesforce also introduced an overhauled AgentExchange that unifies its various app stores: AppExchange, Slack Marketplace, and the Agentforce ecosystem. It brings together more than 10,000 Salesforce apps, 2,600 Slack apps, and 1,000-plus agents and MCP servers into an embedded, in-context discovery surface that surfaces agents and tools at the moment teams are composing solutions, compressing the lifecycle from discovery to activation.
Now it potentially has a single unified agentic catalog from which it can attempt to convert this governance position into marketplace gravity.
In addition, the company announced a $50 million Builders Initiative to incentivize third parties to build within the ecosystem. Each addition extends the governance surface.
What matters for defensibility in the Agentic Era is whether a company’s data and workflow position creates semantic density that agents must route through, and whether that density compounds as usage increases. AgentExchange is the mechanism by which Salesforce converts its semantic density into marketplace economics.
The context substrate
Beneath these moves lies the asset that makes them structural rather than incremental: decades of accumulated enterprise context that the labs cannot replicate at speed.
Context is the accumulated understanding of how work gets done inside a specific organization. This is the scarcer resource in the Agentic Era. Capabilities improve over time, measured in months. Context accumulates on a curve measured in years. In Part II of this series, I examined the model labs’ attempts to become platforms and build enterprise context from scratch. I concluded they were racing against the clock because their core business, frontier models, are becoming commodities.
What the market sees right now appears to be a relentless assault that incumbents are powerless to stop.
Consider just Anthropic. Its progression has been methodical, and each step has sent shockwaves through software markets. Claude Code proved that coding agents could become an enterprise beachhead, reaching $2.5B in ARR. Cowork extended orchestration from developers to knowledge workers. Vertical plugins for legal, finance, and sales triggered the SaaSpocalypse, $285B evaporated from global software stocks in 48 hours, with Goldman Sachs recording the largest single-day decline in a US software basket since April 2025. (The inspiration, of course, for this series.)
Two days after TDX 2026, Anthropic launched Claude Design, a product that turns text prompts into interactive prototypes and pitch decks. Anthropic’s chief product officer had quietly resigned from Figma’s board three days prior. Figma’s stock fell 7% within hours.
It all begins to take on an air of inevitability. Except that it shouldn’t.
Instead, as I have argued, the question is whether the context advantage is real or merely a legacy asset being romanticized. The financial evidence suggests it is real and that the market has not absorbed its implications. This is what specifically demands a more sophisticated, nuanced analysis to properly sort companies into those that have the opportunity to transform into Agentic Era winners and those that are at risk of being massively demoted and devalued.
If we turn this lens back on Salesforce, let’s consider first the depth of the context layer.
The Informatica acquisition was not a product-line addition. It extended the context moat from CRM data to the entire enterprise data estate. In FY26, Data 360 ingested 112 trillion records, including 53 trillion via Zero Copy, up 310% year-over-year, and processed 18 terabytes of unstructured data.
This means that when an agent needs to understand a customer’s full relationship, the open escalation, the renewal due in 30 days, the breached SLA, the relationship owner with a personal connection to their CFO, that understanding now draws on operational data flowing from ERP, HR, supply chain, and financial systems, not just the CRM fields. The context estate has expanded by an order of magnitude.
Consider next how the context is being consumed.
Token consumption leaped from 3.2 trillion in Q3 to 11.1 trillion in Q4, a 3.5x increase in a single quarter. Features get trialed, plateau, and settle. Infrastructure gets consumed at accelerating rates as production workloads scale. That trajectory is an infrastructure consumption curve, not a feature adoption curve. The 2.4 billion agentic work units delivered are output metrics, moments where AI was not reasoning abstractly but operating on institutional context to produce business outcomes.
Then consider how the context is being purchased.
Over 75% of the top 100 deals in Q4 included both Agentforce and Data 360. This cross-attachment rate is the financial signal that conventional SaaS frameworks are not equipped to interpret. When Microsoft sells Copilot alongside Office, the product works with or without it. The attach rate reflects a feature upsell. A 75% cross-attachment rate for two products that did not exist in their current form eighteen months ago reflects something different: enterprise buyers recognizing that the orchestration layer requires the context layer to function and purchasing accordingly.
Over 60% of fourth-quarter Agentforce and Data 360 bookings came from existing customer expansion, indicating that the installed base is the distribution channel for the new architecture. Meanwhile, custom AI agents on Slack have grown 300% since January 2026, confirming that the intent surface is being activated alongside the context and orchestration layers.
This is not a bolt-on. Slack captures what should happen. Data 360 knows what has happened. Agentforce directs what happens next. Remove any one, and the system is structurally incomplete.
OpenAI’s Andrej Karpathy recently argued that “context engineering” is “highly non-trivial” and the defining art of producing AI systems. He went further: the role of application companies is to animate agents into deployed professionals “by supplying private data, sensors, actuators, and feedback loops.”
Headless 360 now brings all of it into the open: private data, conversational sensors, and orchestration actuators. They are exposed as APIs, MCP, and CLI, and are accessible to any agent through any protocol.
The context remains proprietary. The access becomes universal. You can vibe-code an interface. You cannot vibe-code fifteen years of institutional memory.
The risk that remains
This is where the analysis must be unflinching because the risks are not marginal. They are the reason the market’s discount exists in the short term. But in the longer term, they will determine whether the architectural transformation translates into shareholder value.
The most significant risk is seat compression, the threat that agents automate what human CRM users do, reducing the number of people who need licenses.
The more successful Salesforce becomes as the orchestration substrate that agents depend on, the more it undermines the per-seat revenue model that currently generates the vast majority of its $41.5B in annual revenue. As one skeptic tweeted, “Agents don’t log in. They make API calls. So what happens to per-seat pricing?”
Salesforce has begun building the replacement model. Agentforce’s token-based and outcome-based pricing, including the “agentic work unit” as a new metric, represents the economic architecture of the system of action. But $800 million in Agentforce ARR against $41.5B in total revenue means the transition is early. The crossover point, where orchestration revenue growth outpaces seat revenue compression, has not been demonstrated. However, as stated above, Agentforce should not be interpreted as a stand-alone revenue line, but rather in consideration of the whole ensemble.
Still, there is some early validation of this approach. As Box CEO Aaron Levie wrote in response to the Headless 360 announcement: “Agents are going to use software 100X more than people will in the future. As a result, enterprise platforms will become headless and be able to work with any agent on or off the platform. If you don’t do that, you’re DOA. What some have missed is that this creates vastly more use cases for these platforms than existed pre-AI. This isn’t zero-sum. Software value props have traditionally been capped by the number of users in a company. Agents have no upper limit.”
This is the central tension for public-market investors. The architectural transformation is visible and credible. The revenue model transformation is not yet proven. The market is pricing the latter, and in its absence, applying the SaaS-era discount to a company that is ceasing to be a SaaS business in the traditional sense.
The question is not whether Salesforce can build the system of action. The question is whether it can price the system of action in a way that grows total revenue while the per-seat model compresses.
There is also execution risk.
The Testing Center and certain Agent Fabric capabilities are not yet generally available. Agent Broker’s deterministic orchestration is in beta until June 2026. The gap between announcement and production deployment is where transformations compound or stall.
The verdict
In this article, I’ve used Salesforce as a use case, but in a larger sense, this is not just about Salesforce. The most fundamental problem remains with the outdated frameworks that are causing potentially massive mispricing.
Those could be occurring on both the up and downside, and on both incumbents like Salesforce and insurgents like Anthropic.
At the moment, Anthropic is valued at $380 billion, with reports of preemptive offers at $800B, a valuation predicated on orchestration value rather than model value. Salesforce trades at a 30% discount to its own historical peak, a valuation predicated on SaaS-era assumptions about application-layer value.
Let me be clear: No one can guarantee that Salesforce will execute the strategy it has announced. But Salesforce has provided ample evidence at production scale that it has a coherent plan and that its transformation into a company built around capturing value around the orchestration layer is already underway.
To ignore that completely requires a level of skepticism the data no longer supports. And to believe that an LLM such as Anthropic or OpenAI will replicate these functions with ease represents a serious misunderstanding of the core dynamics at play.
The architectural transition is visible. The financial transition is not yet complete. That gap poses a genuine risk to a company like Salesforce.
And yet, capabilities scale in months. Context compounds over years. The current framework prices the former but has no way to evaluate the implications of the latter. So, the market effectively prices the probability that Salesforce successfully navigates the transition to the Agentic Era at zero.
Is that reasonable?
Here is what we can say with confidence. TDX 2026 should serve as the clearest signal yet that it is time to update the framework for answering that question.
Disclaimer: This post reflects the author’s opinions and analysis based on publicly available information. It does not constitute investment advice. The author may hold positions in the companies mentioned.