Microsoft Can’t Add a Table to Word

Daniel Lemire — computer science professor, ranked in the top 2% of scientists globally — asked Copilot for a table inside a Word document. It opened a new document in the cloud. The AI couldn’t modify the file the user was sitting in.

This is Microsoft’s flagship AI product. $30 per user per month. Two years on the market. 446 million potential customers across Microsoft 365.

After all of that: 3.3% converted. 15 million paid users. And of those, only 35.8% actively use the thing. That’s roughly 5.4 million people actually getting value from a product Microsoft positioned as the future of productivity.

As Aakash Gupta documented, Satya Nadella has been personally serving as PM for Copilot since December. Four months of CEO-level intervention to fix a product that can’t insert a table into an open document. Microsoft’s stock just posted its worst quarter since 2008.

And this is the company with the deepest enterprise distribution on the planet.

That’s not a product problem. It’s a structural one. When your entire business model is built around a complex interface — menus, ribbons, dialogs, settings nested three levels deep — grafting an AI on top doesn’t simplify the product. It exposes how much of the product existed only because humans needed the complexity to operate.

Jensen Described the Replacement

While Microsoft was debugging its table insertion, Jensen Huang explained why the patch won’t work.

“There will be no software in the future that’s not agentic.”

He didn’t describe AI as a feature you add to existing software. He described it as a labor class. A new type of worker that doesn’t interact with software the way a human does.

“Our job is not to do the job. Our job is to have the job be done.”

Everything SaaS built over the last two decades — the dashboards for interpreting data visually, the guided workflows so humans wouldn’t skip steps, the integration layers because nobody could keep track of twelve systems at once — all of it assumed a person sitting at a screen. An agent doesn’t sit at a screen. It reads the API response, figures out the sequence on its own, and calls downstream systems directly. The entire presentation layer that SaaS companies spent a decade perfecting becomes optional the moment an agent gets API access.

A five-year product roadmap becomes skippable inside a single model update. Not because the product is bad. Because the human bottleneck the product was designed around stopped being the bottleneck.

Microsoft added Copilot to Word. Jensen described a world where nobody opens Word. That gap is where the entire SaaS industry is about to get stuck.

Chamath Priced the Funeral

Chamath Palihapitiya picked up the financial side of the same argument.

“Nobody in the history of the world has ever seen two businesses like this at this scale.”

He was talking about OpenAI and Anthropic. Two companies that went from zero to trillion-dollar trajectories in under a decade. Anthropic specifically went from a $61 billion valuation to $380 billion in under a year. And on April 4th, they announced that Microsoft 365 connectors for Claude are now available on every plan, including Free. Outlook, OneDrive, SharePoint — all accessible from Claude without manually uploading files. Microsoft charges $30/month for a Copilot that can’t edit your Word document. Anthropic just gave away read access to the entire M365 ecosystem for zero dollars. That kind of capital concentration, combined with that kind of competitive aggression, changes how every other company in tech gets valued.

Chamath used three verbs for what AI does to SaaS moats: eliminate, cannibalize, erode.

Not one of those verbs means “compete.” They all mean “make irrelevant.”

His investment thesis: “I’ll buy the first five or six years of this story, but I’m not buying year 15 anymore.” Translation: the long-duration SaaS premium — the idea that a software company at 20x revenue will keep compounding for 15 years — is over. When OpenAI and Anthropic go public, institutional capital will consolidate around them. The premium multiples SaaS companies have commanded since Salesforce’s IPO will compress to match the rest of the market.

That’s not a market correction. It’s the end of the software premium. Every SaaS company trading at high multiples just got their runway cut by two-thirds, according to the person managing $3 billion in capital who is actively placing those bets.

And Chamath wasn’t bearish on tech overall. He said he’d buy the first five or six years. But the specific SaaS model — per-seat pricing, UI moats, feature accumulation — doesn’t survive a world where agents bypass the UI and vendors compete on API surface alone.

The Products Already Moved

This isn’t just theory from investors and GPU executives. The products themselves already shifted.

Linear. PostHog. Attio. Three different SaaS companies in three different categories. All shipped the same change in the last few weeks: the homepage is a text field. Not a dashboard. Not a feature tour. A prompt.

When your interface collapses to a text input, the only differentiator left is what sits behind it. API surface. Data model depth. Integration quality. The presentation layer — the thing SaaS companies used as their moat for 20 years — just became a commodity.

Think about what this means for switching costs. Salesforce has 300+ tabs. ServiceNow requires a six-week training course before new users can operate the system. That complexity was expensive to learn. Which meant it was expensive to leave. The complexity wasn’t a bug. It was the moat.

A chat bar has zero switching costs. If Linear’s prompt and Jira’s prompt both accept “show me overdue tasks assigned to the backend team,” the only thing that matters is which one returns a better answer faster. The training investment your company made in Salesforce, the custom workflows your admin team spent years configuring — none of that matters when the user’s interface is a text box.

Snowflake, Databricks, and Stripe have been building for this world without knowing it. They always treated the UI as a thin layer over a deep data engine. Their value was never in the dashboard — it was in the computation layer, the data model, the API surface. An agent-first world validates the bet they made years ago.

The companies in trouble are the ones where the interface IS the product. If you strip the UI and what’s left is a CRUD layer with nice fonts, there’s nothing for an agent to call that justifies the price tag. And their customers are about to figure that out.

Three Questions for Your Next Vendor Review

Here’s what I’d bring to the table if I were evaluating my SaaS stack this quarter.

Is your SaaS stack built for agents or for humans clicking dashboards? Check whether your critical vendors have public APIs that an agent can consume without going through the UI. If the only way to pull data from your CRM is exporting a CSV from a browser window, that’s a vendor shaped for a world that’s already ending.

If you build SaaS, is your moat your UI or your API? Linear bet that their data model and API would survive the interface collapse, and they were right. The Innovator’s AI Dilemma is already playing out in SaaS — Cuban’s warning about incumbents getting displaced by AI-native startups doesn’t just apply to your competitors. It applies to your vendors. Strip the UI from your product and honestly evaluate what’s left. If the answer makes you uncomfortable, at least you know before your customers do.

Are you pricing on seats or outcomes? This one is less about strategy and more about arithmetic. Per-seat pricing assumes a human sits in front of a screen. When an agent replaces that human, your client paying $50K a year for 100 seats discovers they need three seats and an API key. Ask your vendors what their pricing model looks like when agents do the work instead of people. Most won’t have an answer yet, which tells you everything about where they are in this transition.

These questions aren’t academic. The execution bottleneck already moved from typing code to directing agents. The same shift is hitting SaaS: the bottleneck is moving from operating dashboards to choosing which APIs to point your agents at.

The bill just arrived

It’s rare to see a GPU CEO, a $3B fund manager, and the products themselves all converge on the same diagnosis within weeks. Jensen says all software will be agentic. Chamath says the valuations already priced in the answer. Linear, PostHog, and Attio shipped the replacement without waiting for anyone’s permission. And Microsoft has been showing, in real time, what happens when you have every distribution advantage on the planet but your AI product can’t modify the file the user is sitting in.

Somewhere in your company there’s a SaaS vendor with a 300-tab interface that your team spent two years learning. That complexity used to be a moat. Now it’s a wall between your data and the agent that could be using it.

If you want to evaluate whether your SaaS stack survives an agent-first world — which vendors are assets, where your data is trapped behind UIs that agents can’t reach — that’s a conversation we have at IQ Source. One session to map where your interfaces are real moats and where they’re just walls. Start here.