expectedwrong hindsight

The Moat Was the Messages

Salesforce just locked down Slack's training data, and the only surprise is that it took this long.

2 min read 236 words #ai #data #salesforce #slack #training-data
hindsight — nailed it

The moat was the messages. Salesforce figured out that every scrap of how humans talk at work is an asset, not a cost center. The soup kitchen metaphor was exact.

Salesforce is blocking AI rivals from training on Slack data, which is the corporate equivalent of realizing you've been running a soup kitchen out of your restaurant and deciding, actually, no more soup.

For years, Slack was just a place where humans typed things — complaints about the Jenkins pipeline, arguments about tab width, the occasional HR incident that somehow made it to #general. Nobody thought of it as an asset. It was a cost center with a dark mode.

Then AI got good at language, and suddenly every scrap of how humans actually talk to each other at work — not the cleaned-up blog post version, the real version, the "can someone hop on a quick call this is on fire" version — became extraordinarily valuable training signal. And Salesforce is sitting on billions of those messages.

So now they've decided OpenAI and Anthropic and whoever else can't have it. Which is completely rational, and also the thing that should make every enterprise customer slightly queasy — you trusted Slack with the most candid version of how your company works, and the entity that owns that candor is now treating it as a competitive weapon.

Salesforce has Agentforce. They have Einstein. They have a strong financial incentive to be the AI that knows how your business actually operates, because they already have the receipts.

The moat was never the software. It was always the messages.