Slack Declares War

Inside the brutal battle unfolding on your desktop

Typically, combat between companies is a subtle thing. There will be carefully barbed press releases or the headhunting of a rival’s key hire, but nothing all that overt. It's quiet building and long-term planning that wins the day in tech.

Sometimes though, when there are billions, maybe trillions on the line, there is no choice but to attack directly. The modus operandi is win at all costs.

That time is now in B2B software. Violence is nigh and Slack just struck first.

“Slack, an instant-messaging service popular with businesses, recently blocked other software firms from searching or storing Slack messages even if their customers permit them to do so, according to a public disclosure from Slack’s owner, Salesforce.”

Unless you happen to be one of those sickos who gets their jollies from building B2B SaaS companies, this may seem like a snoozefest of a quote. Data data data blah blah blah.

However, this is incredibly important—it means that Slack owner Salesforce is trying to own all of your data. You can’t do anything with it. No company and no customer is allowed to touch it outside of the ways they see fit.

As with everything else these days, this is about AI. All of your applications will be doing this soon, whether it’s your favorite social media site or Microsoft Word. Slack is just the first one to act. Everyone should care about this story, whether you’re concerned about never being able to download your Slack data and use it for an internal tool or you’re more worried that you’re missing the boat on the next AI millions. Beyond that, this sort of business strategy will dictate how all of society functions tomorrow.

Don’t worry, I’m here to help. I’m going to review the strategic implications of Slack’s moves against AI agents, moves that are vast, biggly dealio, pay-attention-cuz-your-career-will-wither-if-you-don’t-change level of importance—and then discuss how you should be adjusting your career or company accordingly.

Because like it or not, Slack just changed the world.

The battle for ambient data

Every company’s ideal customer is a rich ding-dong who gives you more money forever. These people are in tragically short supply. So, companies are forced to find other ways to make you pay more money for longer periods. In good, competitive markets, companies do so by providing more value than what they charge. In software, that only sorta happens. Companies get customers to stick around by making it really, really, really painful to swap to another provider. This move is called, in industry parlance, becoming a “system of record.”

A system of record is when a software application holds crucial data or workflows that a customer cannot function without. Slack is the system of record for your chat data, Stripe has your payments data, that sort of thing. Technically, you could migrate to another service, but for large companies, that can be a multi-year effort. Once you become the system of record, you then build products that take advantage of that data integration to go multi-product. Stripe has your payments data and can naturally extend into products around tax or billing.

Andddd that’s kinda it. The last two decades of software has been a pitched battle between tens of thousands of venture-backed companies all vying to be a system of record for either a specific market vertical (like Toast for restaurants) or for a broad horizontal use case (like Adobe for design and marketing).

AI changes this. To understand how, we’ll need to go over how information is stored at a company. To keep it simple, there are four sources of data in a company:

The Four Horsemen of Data, Author’s Illustration

1. Owned Databases: This data is produced and owned by a company, like a database of customer interactions with its product. It’s often too large or clunky to be wrangled into a highly polished application. Employees manipulate and query this data using technical skills like SQL.

2. SaaS Applications: Workflows too complicated for productivity applications are funneled into more specialized apps. For example, customer relationships are better in a fancy CRM than crowded together in an Excel sheet. Tools like these are simple to navigate and users move around by clicking a few buttons. Perfect for the MBA in your life.

3. Productivity Apps: These are the fundamental primitives of knowledge work like word documents and spreadsheets. Individual employees will make them and share them with relevant parties.

4. Ambient Data: You know the guy on the 4th floor who seems to be the only one able to fix the printer? Or how your best ideas come from discussions in a meeting? Ambient data is all the stuff contained in conversations or people’s heads that has never really made it into a computer before.

These categories are not mutually exclusive. They all kinda bleed into each other. However, splitting them up is useful because only the first two were easily accessible before LLMs. Owned databases could be accessed by employees skilled enough to work SQL, and SaaS applications (depending on their data policies and design) were navigable by mouse. You could technically search an individual productivity tool, but only through keywords, which would unearth exact phrases or words, or location, which might tell you which sheet contained the word “budget.” Search was unintelligent and you needed to know what you were looking for. And because ambient data had no capture mechanism, you could not search it at all.

Because LLMs are remarkably good at understanding the written word and spreadsheets, productivity apps are suddenly searchable. Transcription technology has benefited from AI as well, meaning that capturing ambient data, like your coworker chatting it up at the water cooler, is as easy as turning on your computer microphone. And that’s not all! It is trivial to use AI to write quality SQL queries, making it even easier to access owned databases than before or navigate APIs for software applications.

Combine all of that and a connected chatbot could answer any question that you have about the company you work for. This is a long-standing category of software called “enterprise search.” In my opinion, previous solutions like Microsoft SharePoint have been terrible, but LLMs are finally good enough to make it work. Notion (disclosure: they sponsor The Leverage) and Glean are two of the startup challengers going after this category.

So why is Slack panicking?

Still, this seems mostly additive for systems of record, right? Slack still holds the most important data about its customers. Why are they panicking?

The answer is simple: AI can generate, not just query.

Everything flows through the chatbot

These chatbots have a fourfold competitive threat:

1. They can generate new applications, productivity documents, and databases. For example, if your project needs tracking, you can avoid paying a monthly subscription to Asana and create a mini-app yourself. Over time, micro-applications and databases operated by employees will dwarf the number of custom applications they used to buy.

2. Until now, chatbots have been able to see (and potentially copy) any data contained within a system of record. It can access your customer data and also export it if it wants to.

3. The companies that are moving into enterprise search all build productivity suites. Notion started with productivity workflows, Glean brags about their AI agents on their website, OpenAI is reportedly building their own productivity suite and browser, and Google is connecting their applications together with their chatbot Gemini.

4. Models can potentially automate workflows that workers previously performed in specific SaaS apps. Instead of logging into Salesforce to change a customer’s address, you’ll just tell your chatbot to do it.

Again, this is a business strategy. It’ll take execution and persuasion to play out, and my estimate is that it’ll take at least a few years. It’s still a very real fear that all B2B SaaS are grappling with. From Slack’s point of view, it is better to nip this in the bud and shut down all data access before enterprise chatbots become the only category that matters. The last thing a system of record wants to be is “the dumb database” and nothing else.

It's why I imagine companies will greatly limit the APIs and MCPs that LLMs use to connect to their platforms. In the short-run, personal chatbot subscriptions for employees are no big deal. In the long-run, an enterprise-wide license and data pipeline might be a company-ending threat.

The next five years will be a furious storm of acquisitions, investments, and consolidation up and down the stack, where companies will be rapidly trying to corner key sources of data and workflows before an AI provider gets to it first . Your application environment will be totally different. In the medium-term—written for paid subscribers—here is how AI companies are going to respond to companies like Slack limiting data access and which of today’s software companies will survive the transition.

Change or die