The CRM Was Always Just a Table

Salesforce has 3,000 engineers and charges $300 a seat and what you get, at the bottom of it, is a table with rows and columns — company name, contact name, deal stage, last activity — wrapped in enough chrome and workflow configurability that you forget you're looking at a spreadsheet someone hotglued a phone dialer to.

DataGrid and OttoGrid remember.

Both products are, structurally, tables. You put rows in. The rows are companies, leads, contacts, whatever. Then you describe what you want to know about each row — find the LinkedIn profile, check if they're hiring, pull the tech stack from their jobs page, summarize recent news — and the AI works down the column. Every cell in that column is a small research job that used to belong to an SDR at 9am with a coffee going cold next to their laptop.

The CRM industry spent thirty years building abstractions on top of this. Pipeline stages. Activity feeds. Opportunity records. Cadences. Sequences. The entire cathedral of enterprise sales software is a very elaborate way of not saying: we have a list of people we want to talk to, and we need to figure out things about them, and then someone has to send an email.

These tools say it.

What's interesting is the specific abstraction they land on — not "AI does outreach" or "AI writes your emails" but AI as a column-filling engine against a table you control. The human decides what the table is. The human decides what the columns mean. The AI fills them. Then the human decides what to do with the output. That's the loop, and it's the right loop — it keeps the judgment where judgment belongs and automates the part that was always just tedious pattern-matching anyway.

The pattern-matching part is: go to a website, read some things, return a structured answer. An SDR doing this for six hours a day isn't doing something a person is uniquely suited to do. They're doing something a person was available to do.

The CRM platforms will respond to this by adding AI features to their existing products, which is roughly equivalent to a taxi company adding an app. The interface is not the problem. The interface was load-bearing in the sense that it generated billable complexity — implementation fees, training, custom workflows, integrations — but it was never the thing that mattered. The thing that matters is: who are we trying to reach, what do we know about them, and what are we going to say.

A table, some columns, an AI that fills the columns in, a human who reads the output and decides what happens next.

Three thousand engineers and thirty years to not figure that out.