“AI will eat software” has become the reigning market narrative. It’s a provocative narrative, and it may be directionally correct, but it’s about to mislead a lot of investors and operators. The reality from inside the software space is more nuanced — and the implications for who wins a nd loses are more profound.   A new AI-driven computing architecture is splitting the enterprise software landscape into two camps: vendors whose competitive moat is evaporating before their eyes, and those whose value has just multiplied. This restructuring isn’t arbitrary, nor is it a question of execution or branding. It follows a clear architectural logic most market observers haven’t yet articulated.  The interface layer — the polished UI moat most SaaS vendors spent the past two decades perfecting — is being commoditized by large language models (LLMs). Any vendor whose core value is “we make data beautiful and easy to query” is now competing with LLMs that do the same thing for free, via natural language.   But LLMs are probabilistic. They can’t produce the verifiable, auditable computations enterprise decisions require. For that, the enterprise is relying on a different kind of system entirely – one many executives and investors haven’t yet learned to recognize, and one that’s about to become the most valuable layer in enterprise software: the Deterministic Domain Authority (DDA).   Transparently, I have a horse in this race. At Anaplan, I’ve spent the last several years building the kind of deterministic planning infrastructure I’m describing here: pairing the reasoning power of large language models with a computational engine designed to meet the enterprise’s accuracy imperative. That certainly informs this argument, alongside my experience working with customers navigating these decisions in real time. The new stack The traditional, monolithic SaaS application is being disassembled into a stack of three specialized layers, each with a distinct nature, a distinct role, and a distinct relationship to the other two. The layers are not interchangeable. Each does something the other two cannot. At the top sits the LLM: the universal conversational AI assistant that recognizes and processes text, images, audio, and video. The LLM is the user’s new front end to everything. It understands natural language in all its nuance and ambiguity, allowing users to state their objective as they would to a colleague. Its role is extracting intent from natural language and orchestrating a response. Below it sits the DDA: a governed, auditable, 100% accurate computational engine that owns authoritative truth within a specific domain such as finance, supply chain, or workforce. A deterministic system produces the exact same output every time it receives the same input. LLMs offer varying outputs based on statistical pattern matching, but DDAs rely on fixed, rule-based logic. The result is precision, compliance, and auditability — non-negotiable requirements for the decisions enterprises actually run on. The bottom layer is the Model Context Protocol (MCP), the secure execution layer that translates the computed truth from the DDA into native commands for other enterprise systems, ensuring actions are carried out accurately and with governance. The user experience is now owned by the LLM, and everything else becomes AI infrastructure. Who’s at risk and who’s amplified For a wide swath of SaaS companies, this architecture is an existential event. Any vendor whose primary value proposition was making data easy to see, interact with, or visualize is in immediate danger. The competitive moat of the experience wrapper is being disintermediated. The casualties are obvious once you know what to look for. Business Intelligence (BI) and dashboard tools, whose core function was making data beautiful and navigable, are directly superseded by an LLM that synthesizes and presents the same information in a sentence. Lightweight analytics products built around the slogan “ask questions of your data in plain language” are redundant as the LLM is now the plain-language interface. Workflow automation tools that simply move data between systems without owning any authoritative computation are next; an LLM paired with MCPs can orchestrate those workflows natively. Same for collaboration layers built on top of commodity data: that conversation now happens inside the LLM. The survivors, however, share a common trait: they own a domain of deterministic, authoritative, scalable computation the LLM can’t replicate. Their value is structural. LLMs cannot produce identical computational outcomes given the same inputs. They fail at multi-step algebraic manipulation, complex modeling, and any task requiring precise step-by-step reasoning. They have no choice but to delegate to DDAs. The vendors poised to win include enterprise planning engines, which compute the definitive impact of a real-time scenario on a financial plan. They include core systems of record like Human Capital Management (HCM) and Customer Relationship Management (CRM), where elements including hire date, comp number, and closed-won deal amount are governed facts, not suggestions. And they include specialized scientific and regulatory databases — drug interaction systems, tax code engines, orbital mechanics simulators — where approximation is unacceptable. Despite this sorting of SaaS, DDA owners aren’t fully immune to commoditization. If multiple DDAs exist for the same domain, the LLM will be indifferent. It will choose the one that provides the necessary determinism at scale and at the lowest cost.   Therefore, the surviving moat in this new world is the customer’s unique model running on the DDA: a complex, computational digital representation of how a specific organization plans, allocates, and governs. Recreating that model in a competing system goes beyond data migration; it’s a re-encoding of the company’s institutional intelligence, a genuinely painful process that creates a powerful new form of competitive differentiation. The conclusion for every enterprise software vendor is stark and clarifying. The future belongs to the vendors that provide undeniable truth. The pretty window into data is no longer a defensible business.  The strategic imperative is to lean into invisibility. Every dollar spent polishing a UI is a dollar wasted defending a moat that’s already gone. Investment must flow in the opposite direction: toward deeper computation, stronger governance, and richer modeling.   The opinions expressed in Fortune.com commentary pieces are solely the views of their authors and do not necessarily reflect the opinions and beliefs of Fortune. This story was originally featured on Fortune.com ...read more read less