Databricks' Matei Zaharia and Reynold Xin argue enterprise agent clouds require an open ecosystem and debate database versus OS expertise

Highly recommended if you want to understand how why Omnigent and LTAP are important!

@latentspacepod @databricks @rxin Thanks for having us on, really fun conversation!

@swyx @neondatabase @nikitabase @ankrgyl Fun conversation, thanks for coming to @Data_AI_Summit!

Why the Frontier Ecosystem must be Open — Matei Zaharia and Reynold Xin, Databricks https://www.latent.space/p/databricks

@databricks cofounders @matei_zaharia and @rxin explain why Databricks is moving into the infrastructure layer for enterprise agents, how Omnigent creates a shared harness for coding agents and custom agents, why LTAP and Lakebase rethink the split between operational and analytical databases, why agent security needs contextual policies and spend controls, and why the future of software may be as simple as getting the right data in place and putting agents on top.

@swyx @neondatabase @nikitabase @ankrgyl loved this episode. contextual policies for agent security is exactly where the puck is going. static rbac is basically useless when an agent needs to reason about whether an api call is safe. 🫶

@swyx @neondatabase @nikitabase @ankrgyl lot of heat but that Databricks vs Snowflake answer needs to stand on its own without the pod aura

did they actually give receipts or just assert?

@swyx @neondatabase @nikitabase @ankrgyl A lot of alpha, yes, but also a lot of new jargon for people to pretend they understood on first listen.

@swyx @neondatabase @nikitabase @ankrgyl the neon talk is always solid, love that for them.

@swyx @neondatabase @nikitabase @ankrgyl LTAP completely redefines real-time architectures. Merging analytical efficiency directly into transactional systems is a massive win.

@swyx @neondatabase @nikitabase @ankrgyl That metaharness pivot is wild. Everyone’s racing for the same dev infra stack now, but who actually owns the developer experience when the novelty fades? That's where Snowflake stumbled.

@latentspacepod @databricks @matei_zaharia @rxin Everyone says agents. The moat keeps moving toward the harness that decides which ones deserve to act.

@swyx @neondatabase @nikitabase @ankrgyl I was hoping Latent Space would be the next Semianalysis.

@swyx @neondatabase @nikitabase @ankrgyl ltap is the real story here. the shift from human-in-the-loop to agent-first infrastructure is finally forcing the merge of oltp and olap. databricks moving to postgres-compatible lakebase makes a lot more sense now that agents are the primary users.

@alighodsi This was awesome

@swyx @neondatabase @nikitabase @ankrgyl such a fun one!! thanks for having us

@swyx @neondatabase @nikitabase @ankrgyl Databricks vs Snowflake is just compute vs data gravity strategies converging—the real alpha is who owns the metadata layer long-term.

@swyx @neondatabase @nikitabase @ankrgyl The metaharness idea is the sleeper. The interesting version isn't better eval tooling — it's that the hard part is now deciding which evals mean anything in the first place. That's a judgment problem, not a tooling one.