KorvaJoin the waitlist
Tech

AI Cannot Help With Company Work Because It Cannot See Meetings And Decisions

To do core work, software needs meeting notes, decisions, and org charts, not just smarter models.

AI image
All-In with Chamath, Jason, Sacks & Friedberg · Databricks CEO: AGI Is Already Here, Why Companies Are Behind, and the Superintelligence Debate

Ali Ghodsi, the chief executive of data company Databricks, says the enterprises he meets still use artificial intelligence in only two basic ways.

They use chatbots as a glorified Google search, getting an answer instead of ten blue links, and as coding helpers. For coding, Ghodsi says customers tell him they see only about 20 to 30 percent speedup, which he describes as not enough.

He says he asks customers how many have large groups of agents doing core work, and the answer is that it just has not happened yet.

The reason, Ghodsi argues, is that AI cannot help with work it cannot see.

It does not matter how smart the model is, he says, using the example of a great mathematician. If that genius is kept out of the meeting, and does not know what was decided by hundreds of thousands of employees, it cannot do the job.

Without that record, Ghodsi says, the system misses what happened in the room, so it cannot help with that work.

One chief executive of a large bank told him his company would never record meetings.

The blindness matters more as companies grow. Up to about 150 people, a number researchers call Dunbar's number, everyone can still know everyone else.

Beyond that size, a company becomes a coordination problem. Formal charts show decisions moving up and down, but real products move sideways.

Customer feedback goes to product managers, who hand it to engineers, who build something that marketing explains and sales sells. Each handoff can stall.

To move faster, Ghodsi says, AI needs to understand the whole business, including its private know-how, its organization chart, and who is who.

That is what Databricks is trying to build with a product called Genie. Its core is what engineers call an ontology, a graph of the company that updates constantly and links people, reporting lines, assignments, emails, meeting transcripts and decisions, chat messages, and documents such as spreadsheets and presentations.

Ghodsi says newer, digital native firms, meaning companies that have existed only 10 to 15 years, are likely ahead at building that model, while older large employers face more old software, regulation, and security rules that make change harder.

Questions people ask

What is an organizational ontology?

An ontology is a structured map of how things relate. In this case, it links people to their managers, their projects, their meetings, their messages, and their documents, so software can trace who decided what and why.

Why would a company refuse to record meetings?

Companies worry the recordings could be used in lawsuits, regulatory investigations, or leaks. A bank handles sensitive client and trading matters daily, so executives often decide the legal risk outweighs the potential productivity gain.

What is Dunbar's number?

It is a proposed limit of about 150 stable social relationships, named for anthropologist Robin Dunbar. Ghodsi uses it to mark the point where informal, everyone-knows-everyone coordination stops working and formal processes take over.

Why might younger companies adopt this kind of AI faster?

They started with cloud software, fewer legacy systems, and fewer regulated processes. That makes it easier to connect email, chat, code, and documents into one governed place, while older firms must untangle decades of separate systems and approvals.

More in Tech