Key Takeaways

  • Databricks builds a live organizational ontology graph that ingests Slack messages, emails, calendar schedules, and meeting transcripts to map company operations.
  • Tracking output across 2,000 Databricks engineers revealed an L-shaped productivity distribution, where top developers ship vastly more code per week through automated pipelines.
  • Jason Calacanis labeled the approach "management moneyball" because it pulls ground-truth data on operational friction before executives can spin their progress.
  • The primary operational metric Ghodsi tracks across the company graph is reducing human blockers across teams.

Replacing Executive Spin With Ground Truth

Every CEO spends hours playing detective. An executive misses a deadline, points fingers at product or sales, and produces a narrative designed to avoid blame. In a live demonstration with Jason Calacanis, Databricks CEO Ali Ghodsi showed a different way to run an executive team: query an automated graph instead of listening to a pitch deck.

Databricks built Genie to ingest internal communication streams into an ongoing knowledge graph. As Ghodsi put it, “The key thing is that it has an ontology. The ontology is a graph that it constantly is constructing for your company. That graph understands all of the people, who reports to who, what are they working on, what are the emails that are coming in, what meetings have happened, and what's the transcripts of all those meetings and all the decisions that are being made there, what's on Slack, what's being communicated.” (play from 23:11)

Calacanis immediately recognized the shift, comparing it to Moneyball for management. When operational facts sit in a queryable database, status updates stop being political exercises. “Here you take the people out of that discussion,” (play from 32:06) Calacanis noted. “You say here's first principles transparently what happened. Now we have ground truth and we can start the discussion without you being the CEO having to be Mr. Detective.” (play from 32:09)

The L-Shaped Reality of 2,000 Engineers

Ghodsi demonstrated this visibility by opening internal metrics for Databricks' engineering organization. The data did not show a standard bell curve. Instead, it formed a severe cliff.

“This is our 2,000 engineers and this is how many pull requests or how much code they've actually shipped into production,” (play from 29:46) Ghodsi explained during the demo. “You look at the middle here, it's maybe per week roughly. But it has this L shape, and you look at this L shape. The top people are actually committing way more.” (play from 29:52)

The divergence comes down to workflow adoption. The engineers at the tall end of the L do not simply type faster. They integrate automated coding pipelines into their daily routine, allowing them to ship multiples of what an average engineer completes. The engineers at the flat bottom of the L continue working manually. In a modern software company, the gap between tool-augmented engineers and traditional coders is no longer ten percent; it is an order of magnitude.

The Single Metric That Matters: Human Blockers

Most companies measure activity: lines of code, tickets closed, or hours spent in reviews. Ghodsi focuses on friction points. When asked what he targets inside the graph, his answer was direct: “So one of our main goals is reduce number of human blockers. So we measure that throughout the organization.” (play from 33:08)

A human blocker occurs whenever work stops because an engineer or project manager is waiting on an approval, a code review, an access request, or an email reply from another person. By analyzing cross-team chatter, calendars, and pull requests, Databricks identifies where projects sit idle waiting for manual sign-offs. You fix the delay not by telling engineers to work harder, but by removing the human who sits between the engineer and production.

What to Do With This

Audit your engineering team's pull request cycle times this week to find your human blockers. Open GitHub or GitLab, filter your last 50 merged pull requests, and calculate the exact hours between "ready for review" and "merged." If your median delay exceeds 24 hours, identify the two reviewers who hold up reviews the longest, and replace their manual gatekeeping with automated test suites or peer-rotation rules.