George Fraser | AI + a16z Summary

AI + a16z | George Fraser (CEO, Fivetran) & Martin Casado (a16z)

1. Key Themes

Data Centralization Is Now an AI Imperative, Not Just a BI Nicety

What was once a business intelligence best practice has become a foundational requirement for AI agents to function. The same infrastructure built for reporting now powers AI context — but only if vendors allow access.

"There is a new reason to have all your data in one place, which is if you want to use AI agents in business, AI agents need context... if you don't do that, then it's using ChatGPT from before ChatGPT was connected to the internet." — George Fraser [00:02:43]

SaaS Vendors Are Locking Down Data Access — and It's a Mistake

A fear-driven, defensive reaction from SaaS incumbents to the AI wave is causing some to restrict API access. Both speakers agree this strategy is self-defeating and historically familiar.

"SAP announced a new API policy that literally said all AI agent access was banned, except in a way specifically approved by SAP... it just shows how extreme the reaction of some of these companies has been." — George Fraser [00:04:33]

"I am hopeful that this is merely a brief flirtation with closed data by most of these vendors. And they will realize this is not a good idea, that they are at war with their own customers, and it's not even going to work anyway." — George Fraser [00:13:46]

AI Agents Are Best Treated as Human-Like Employees, Not Software

The most practical near-term architecture for enterprise agents is giving them their own identities, credentials, and workflows — slotting them into existing human-designed systems rather than rebuilding everything.

"We'll actually have an HR. HR, and that HR team will onboard AIs as they come. They will train them. They will show them the access to the documents that you need. They'll be part of teams. They'll join the Slack, just like humans do. And in that world, these aren't software. There's actually more seats, more consumption of software." — Martin Casado [00:21:00]

"The reason this works well is because you can slot it into the existing workflows without having to refactor the whole universe." — George Fraser [00:21:28]

2. Contrarian Perspectives

SaaSocalypse Is Overblown — The Real Threat Is New Entrants, Not Agent Replacement

Against the dominant narrative that AI agents will destroy SaaS categories, Fraser argues the actual threat is fast-moving AI-native startups outcompeting incumbents — not wholesale replacement of software with agents.

"I don't really buy into this reason that all the SaaS categories are going to disappear and be replaced with vibe-coded software. I think there will be some, but I think the bigger threat is simply new companies coming along. It is just so much easier to write software now that AI-native companies will just zoom and catch up to the established incumbents and maybe be better in some ways." — George Fraser [00:29:50]

Data Gravity Is Completely Fake

While nearly every enterprise data vendor and analyst treats data gravity as a given constraint, Fraser argues it's a myth created by poorly designed data pipelines — not a real physical or economic constraint.

"I think data gravity is completely fake... You can have a huge data set, but if you just replicate the changes, the changes are always much smaller than people think. A lot of this idea of data gravity came from dumb data pipelines that people wrote where they would copy their entire company's data sets out of their database every day, once a day at midnight." — George Fraser [00:15:39]

Postgres Is Bad — An Undergrad Writes Better Databases

Against the near-universal reverence for Postgres as the gold standard of open-source databases, Fraser argues it's technically inferior legacy software that the industry is simply stuck with.

"Postgres, contrary to popular belief, is very old technology. It is not a good database. Undergraduates writing class projects write better databases than Postgres. Not because the people who built Postgres were not smart, but simply because it was written a long time ago. It has a lot of technical debt." — George Fraser [00:45:11]

The Consumption/UI Layer of Infrastructure Is the Only Layer AI Actually Threatens

Rather than AI commoditizing all infrastructure, Fraser makes a surgical argument: only the most user-friendly abstraction layer is under threat — the core infrastructure actually benefits from AI-driven demand.

"AI is quite good at navigating slightly more complicated infrastructure. So if you have an AI agent, maybe you don't really need that very most user-friendly layer. You can drop down to the next one and use that." — George Fraser [00:38:12]

AI Coding Agents Won't Kill Fivetran's Core — The Long Tail of Complexity Is Underappreciated

The assumption that AI can vibe-code data connectors to replace Fivetran underestimates the discovered, not designed, complexity of real-world data replication.

"They continue to improve in terms of what they can put out. They still do not discover this long tail of complexity. It is really surprising how difficult it is just to make an accurate copy of a system and keep it up to date." — George Fraser [00:31:49]

3. Companies Identified

Fivetran Data integration company founded in 2013, helping enterprises centralize data from all SaaS systems into a single data lake. Mentioned as the subject company — growing despite AI disruption fears, with both OpenAI and Anthropic as customers.

"We replicate lots of data from these very SaaS tools on their behalf into their data lakes. So if they're still using them, do we really think the company of the future is not going to be?" — George Fraser [00:10:09]

dbt Labs Data transformation tool that models and organizes raw data into business-logic-ready structures. Highlighted as a major beneficiary of AI coding agents and a natural complement to Fivetran.

"I think coding agents are going to write tons of dbt models... it's a great way to express the rules of data at my company. And even if it's being written by AIs, you still want to have that artifact that is an executable documentation of how your business works." — George Fraser [00:41:42]

Anthropic AI safety and research company. Specifically called out because their internal data infrastructure looks identical to traditional enterprises — validating that modern data stacks are sufficient for AI-native companies.

"The systems at Anthropic, one of the people who helped set them up was a consultant who had set up Fivetran and DBT at many other companies. So their data platforms look very typical." — George Fraser [00:35:44]

Snowflake / Databricks / BigQuery Cloud data platforms. Mentioned as the right foundational infrastructure for AI context — not exotic new systems.

"If you have a reasonably modern data platform, something like Snowflake, Databricks, or BigQuery... that is a great foundation for your context for AI as well." — George Fraser [00:36:14]

4. People Identified

George Fraser Co-founder and CEO of Fivetran. Unusually technical for a CEO of a company at this scale — actively prototyping experimental databases, AI agents, and internal tools.

"I'm working on a from-scratch classic OLTP SQL database... what it attempts to do is to be like SQLite except S3 is the backing store... I think there's actually a real opportunity right at this moment." — George Fraser [00:44:11]

Sridhar Ramaswamy (Sirdar) CEO of Snowflake. Used by Fraser as a mental model for evaluating bold strategic decisions — asking "what should Sirdar do?" as a way to bypass personal risk aversion.

"In the DBT merger, when I was reflecting on that, one of the tricks I use is I ask myself, what should Sirdar do?... And that was like a clear answer in my mind was like merge with DBT, absolutely." — George Fraser [00:48:12]

5. Operating Insights

The "What Would Another CEO Do?" Decision Framework for Big, Scary Moves

Fraser uses a powerful cognitive trick to bypass founder risk aversion: when facing a large strategic decision, ask what a respected peer CEO would obviously do in your position, then simply execute that answer.

"One of the tricks I use is I ask myself, what should Sirdar do?... And then I just go do that. And that was like a clear answer in my mind... even though it seems very big and scary when you imagine, is this a good idea for someone else? Then you can kind of get there." — George Fraser [00:48:12]

Write Data Access Rights Into Your MSAs Before AI Locks You Out

CIOs have significant contract leverage right now that most aren't using. Fraser recommends proactively writing data access guarantees into vendor contracts — and notes that even asking for it sends a market signal.

"If you're signing $500,000,000 contracts, insist on data access in your MSA. And you will find surprisingly often that you get it." — George Fraser [00:19:01]

AI Coding Agents as an Infinite Junior Engineering Supply for Maintenance-Heavy Work

The insight isn't just "use AI to write code" — it's that AI agents are specifically valuable for the kind of high-volume, low-glamour bug-fixing and maintenance work that scales poorly with human headcount.

"You can imagine how you can use AI coding agents, which are basically an infinite supply of junior engineers — that is a particularly valuable tool for this kind of problem... we've really especially the last couple months started to see it work and started to see improvements at scale." — George Fraser [00:34:17]

The "Pretend You're a New Incoming CEO" Self-Audit

Fraser regularly runs a mental exercise imagining he was just brought in by the board to fix Fivetran — forcing himself to identify what an outsider would immediately dismantle, which breaks founder attachment bias.

"One of the mental exercises you've been doing as long as I've known you was pretending if you were a new CEO brought in by the board to fix Fivetran, like what would you immediately unwind?" — Martin Casado [00:47:13] "Many things. That's an exercise I do regularly." — George Fraser [00:47:42]

6. Overlooked Insights

SQLite-on-S3: A Real Product Opportunity for AI Workflow State Management

Fraser briefly mentions building a proof-of-concept database — SQLite-equivalent with S3 as the backing store — specifically to address the need for "zillions of tiny databases" in AI workflows. This is not a hobby project throwaway. It identifies a genuine architectural gap: AI pipelines need lightweight, persistent, per-workflow state storage that existing tools (Postgres, SQLite, DynamoDB) each fail at for different reasons. No major product currently solves this well at scale.

"What it attempts to do is to be like SQLite except S3 is the backing store... when you build AI workflows, you have this need for like zillions of tiny databases... I think there's actually a real opportunity right at this moment. I really lament that we're sort of stuck with Postgres forever." — George Fraser [00:44:41]

This is a greenfield infrastructure opportunity that Fraser is essentially flagging publicly while simultaneously saying he doesn't have the bandwidth to pursue it — an unusual and direct invitation.

dbt as the "Executable Documentation" Layer That AI Governance Requires

Almost in passing, Fraser cites Dijkstra to make a point that goes well beyond the dbt merger rationale. As AI agents write increasing amounts of data transformation logic, enterprises will need an auditable, human-readable artifact that encodes business rules. dbt's SQL models are uniquely positioned to serve this function — not just as a data tool, but as a compliance and governance layer for AI-generated data logic. Nobody in the AI governance conversation is talking about this.

"There is this great quote from Dijkstra... computer code should be seen as a means of communication between humans and only incidentally as an execution format for computers. And nowhere is that more true than in SQL queries in dbt projects... you still want to have that artifact that is an executable documentation of how your business works." — George Fraser [00:42:12]