Disclaimer

This is a working thesis meant to encourage debate, not a definitive claim or investment advice. It reflects my personal views and I’m sharing it to invite discussion. It’s not exhaustive, and I could be wrong. I genuinely welcome feedback, critiques, and counterarguments. I hope this will make all of us a bit smarter and more thoughtful about AI and its impact on businesses.

The Bears

In early 2026, a series of announcements from the foundation model labs related to coding, enterprise software, and white-collar work helped catalyze a massive sell-off in publicly traded software companies. This is now known as SaaSapocalypse. Over the last 12 months, the Emerging Cloud Index (a collection of SaaS and tech public companies) has returned -30% compared to the NASDAQ of +13% and S&P 500 of +12% over the same period (as of 2/13/2026).

The current bear narrative has several themes:

1. Software creation is collapsing as the marginal time and cost to create code approaches zero, making any SaaS company easy to replicate.

2. Seat-based pricing is structurally challenged if AI replaces white-collar roles.

3. SaaS multiples have been inflated due to high stock-based compensation, which is a real expense to shareholders.

4. Value will now accrue to an agentic orchestration layer that sits on top of systems of record and commoditizes them.

In light of the indiscriminate selling across public software names, I’m here to offer a counterargument and a point of view on how a business may be positioned to combat disruption, accelerate growth, and even widen its business moats leveraging points 1 and 4: advancements in coding and creating an agentic orchestration layer. Points 2 and 3 deserve articles of their own which I may cover at a later date.

A business I will use for my argument is Microsoft (focusing on its productivity platform, not Azure which is a more obvious beneficiary of AI). I believe Microsoft is unfairly caught in the SaaSapocalypse crossfire and may come out an even stronger business than before.

My thesis is that coding acceleration will benefit Microsoft, allowing it to capture new and existing software markets faster, and if an agentic orchestration layer emerges, Microsoft is structurally advantaged to own it, all while strengthening its business moats.

Software Creation

AI agents will inherently make software development faster. If you can code and iterate faster, you can ship more — and higher-quality — software products. AI is also strong at researching, determining, and validating the specs needed to replicate existing software. AI agents excel at traversing databases and applications to determine how to copy them — you can envision a future where you give AI agents access to one of your software tools, and it replicates it and ports the data.

Software replicability does not automatically equal business moat erosion (that topic alone requires its own essay), but the marginal cost and time to create software has declined significantly — and will continue to fall.

The bears argue that if everyone can ship more software products, existing software companies lose their product advantage. However, if an incumbent’s business moats remain intact, I believe faster coding can disproportionately benefit companies that can ship features into an existing workflow surface.

Agentic Orchestration Layer

The agentic orchestration layer vision: a chat or voice AI interface acts as an orchestration layer between the user and their applications. A user chats with this AI agent, which does work on their behalf — leveraging email, chat, productivity tools, notebooks, and documents. The user no longer needs to directly use these applications. The AI agent has the context and connectivity to complete work across apps and report back when finished.

The bear camp’s major argument: it no longer matters which software applications you use because AI agents can navigate and interact with anything on your computer — or even create their own applications. Unlike humans, AI agents have no workflow preferences. They don’t care whether they use Edge or Chrome, Teams or Slack, OneNote or Notion. They’ll use whatever works, or build something new. In this world, user interfaces for these applications become irrelevant. The only interface that matters is the AI agent’s. Everything else on the computer becomes a commodity, including all the SaaS vendors.

Many bottlenecks and friction points could prevent this vision from becoming reality. However, if it does emerge, Microsoft has unique structural advantages to build and own this agentic orchestration layer.

The Microsoft Bull Case

Now that we’ve established that coding productivity is advancing and an agentic orchestration layer is a compelling possibility, the question is: who will benefit?

My thesis: Microsoft is one of the best-positioned companies — if not the best — to capitalize on both faster software development and the potential emergence of an agentic orchestration layer.

The Microsoft bull case rests on two cornerstone points:

1. It is supply-constrained, not demand-constrained, on its ability to produce and ship products within its productivity suite.

2. It owns the operating system (OS) level, which could be the most important layer for AI agents.

Microsoft’s Best of Suite and Network Effects

One way to frame the AI coding debate is through supply and demand for software. (Note: coding is only one piece of the software development life cycle, which itself is a small part of what’s required to build and ship a product). Advances in agentic coding create a positive supply shock — the cost per unit of code drops for software suppliers.

For most SaaS, AI creates a glut of competition which can cause a race to the bottom on pricing. But for Microsoft, AI clears the backlog of features its users are already asking for.

Microsoft has historically been constrained from building applications due to the cost and time required to produce and maintain code. Even at Microsoft scale, the suite’s roadmap is a queue: every new app or feature competes with debugging, patching, maintenance, and integration work across a massive installed base. That’s partly why they created an application marketplace and support both locally installed and web-based apps. The app store and web app ecosystem essentially outsources the capex of software creation to third parties.

However, when an application is deemed important enough to be part of the Microsoft suite, they productize the category, bundle it, and integrate it tightly. This strategy has performed extremely well. Examples include Microsoft Teams (vs. Zoom and Slack), OneNote (vs. Notion), OneDrive + SharePoint (vs. Dropbox / Box), and Microsoft Edge (vs. Chrome and Firefox). They succeed because these disparate tools work better when offered by the same vendor and natively integrated.

Having Outlook, Word, Excel, Teams, and SharePoint interconnected creates productivity synergies compared with using separate tools. This creates a compounding advantage: individual users work faster, teams collaborate more seamlessly on shared documents, and when external partners also use Microsoft, cross-company collaboration becomes frictionless. These last two points enable Microsoft’s network effect: each additional Microsoft user increases the value of the same shared files and workspaces for everyone else — more teammates and partners can open, co-edit, permission, and share the same documents with consistent links, version history, and access controls, without format conversion or access pain points — so the platform gets more useful as adoption spreads. Even if AI agents are doing the work instead of humans, that friction is still lowest when they can rely on a common identity/permissions model and a single canonical file protocol.

The more software Microsoft provides, the more the entire enterprise ecosystem benefits. As a result, there is inherently more demand for Microsoft productivity tools than it can supply.

As AI coding advances and software creation costs fall, supply-constrained companies will benefit most. This points to distribution power, product suite integration, and network effects as powerful economic moats and sources of growth in an era increasingly defined by AI coding productivity.

Microsoft’s Agentic Orchestration Layer

Microsoft has strong solutions across nearly every major layer of an enterprise tech stack: Cloud (Azure), Platform (Windows), and apps (productivity suite). The only major layer it has historically lacked is the foundation model layer, and public market investors have penalized them for this. The bear argument: if LLMs sit on top of all apps, then those apps — and the rest of the tech stack — lose its value. However, Microsoft is well-positioned to become the agentic layer for three reasons: 1) it owns the front door of the computer, 2) no other vendor in your tech stack knows your business like Microsoft does when you use its productivity suite, and 3) Microsoft has proven it can replicate applications and bundle them into its suite well — and it can foreseeably do this with an LLM application layer.

Model providers have the intelligence layer, but Microsoft owns the front door: the OS login and the default workflow surfaces. Even as some work shifted from locally hosted to web and mobile, identity and policy are still centralized — and Microsoft is the control plane for a huge share of enterprise access. The OS and the web have been competing as the primary ‘workflow surface’ for decades, but in many enterprises the OS still sits one layer above the web: it’s where users authenticate, where permissions are enforced, and where work is orchestrated across web-based applications.

If you keep rolling forward ‘who layers on top of whom,’ the top layer — the one users start from, and the one best positioned to ingest context and dispatch actions — looks like Microsoft. Imagine logging into your computer: the first thing you interact with is Microsoft’s agentic layer (this could be Microsoft Copilot or something new). In this tech stack, LLM providers become one of many applications running on top of Microsoft’s operating system — tools rather than the orchestrating platform itself.

The more context an AI agent has about your business, the more effective it will be at the task. Microsoft contains the most context about your business and how you work. This information is stored in productivity apps (Outlook, Teams, Excel, Word, OneNote) and in the folder architecture (local drive and SharePoint). Many model providers are pursuing interoperability via ‘connectors’ and MCP-style integrations. Today, those integrations can be higher-friction than native wiring: they take effort to set up, and data sharing is often less seamless. If the goal is real automation across a business, agents need that end-to-end context and permissioning — making Microsoft the vendor of choice.

Finally, it’s too early to rule out Microsoft building a strong foundation model, leveraging open-source LLMs, or forming partnerships that eliminate the need to own one. If any vendor excels at waiting for a market to mature, replicating the leading application, and integrating it into its ecosystem, it’s Microsoft.

Value Capture

The ‘agentic orchestration layer’ thesis doesn’t automatically imply that all value migrates to standalone model labs; it implies that value migrates to whoever can combine models with context and workflow tools. Microsoft already sits at the choke points that matter: the OS surface, the required enterprise context and the surface on which organizations do work. As software creation gets cheaper, the advantage shifts away from who can build the most features and toward who can ship them into the default workflow, securely, with minimal friction — then compound adoption through ecosystem network effects. Model providers may supply the raw intelligence, but durable economic value will also accrue to whoever owns the front door, the enterprise data, and the primary surfaces where work gets done.

You can apply this framework not just to Microsoft, but other software businesses who have similar characteristics in their domains.

The bear case assumes Microsoft gets flattened by the agentic layer; the bull case is that Microsoft becomes it.