AI Agents and the Fight for Customer Data
- 01The Data Centralization Imperative Is Now Existential for AI
- 02SaaS Incumbents Are Irrationally Locking Down Data Access
- 03AI Agents Are Being Treated Like Human Employees
The a16z Show | George Fraser (CEO, Fivetran) & Martin Casado (a16z)
1. Key Themes
The Data Centralization Imperative Is Now Existential for AI
The old reason to centralize data was business intelligence and reporting. The new reason is AI agent context — and the stakes are dramatically higher. Without centralized, continuously updated data, AI agents are blind to your business reality.
"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 [00:02:21]
SaaS Incumbents Are Irrationally Locking Down Data Access
A wave of defensive data restriction is sweeping legacy SaaS vendors (SAP, Salesforce/Slack) in response to AI disruption fears. Both speakers argue this is strategically self-defeating, historically repeatable, and ultimately temporary — but the damage to customers in the interim is real.
"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 [00:04:37]
"Anytime vendors put up walls and try to regulate data access... if you don't do that, then it's using ChatGPT from before ChatGPT was connected to the internet." — George Fraser [00:10:27]
AI Agents Are Being Treated Like Human Employees — And That's a Feature, Not a Bug
Rather than building exotic new agentic architectures, the most practical path is giving AI agents their own identities (email, phone, Slack access) and slotting them into existing human workflows. This sidesteps the complexity of rewriting legacy systems while unlocking immediate productivity.
"We'll actually have an 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." — Martin Casado [00:21:19]
"The reason this works well is because you can slot it in to the existing workflows without having to refactor the whole universe." — George Fraser [00:21:28]
2. Contrarian Perspectives
Data Gravity Is Completely Fake
The widely accepted belief that data is too large and expensive to move — forcing companies to co-locate all services in one cloud region — is, per George Fraser, based on a fundamental misunderstanding. Change data capture means only deltas move, not full datasets, making egress costs trivial.
"I think data gravity is completely fake. I am the only person who thinks this... You will be astonished despite replicating huge data sets for thousands of companies. We have 7,000 customers of size... The amount of data being moved at any given time is tiny. And the reason is that we're doing change data capture." — George Fraser [00:15:37]
"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... They were just repeatedly copying the same data over and over." — George Fraser [00:17:13]
The SaaSpocalypse Threat Is Misdiagnosed — The Real Danger Is Greenfield AI-Native Competitors
Everyone is focused on agents replacing SaaS. Fraser argues the more dangerous threat is that AI makes it so easy to build software that new companies will simply outcompete incumbents from scratch, not that agents will replace existing workflows.
"I don't really buy into this reason that all the SaaS categories are going to disappear and be replaced with vibe-coded software... 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 [00:29:52]
Postgres Is a Bad Database and the World Needs to Replace It
Contrary to the near-religious reverence for Postgres in the developer community, Fraser calls it technically obsolete with a storage engine that even undergraduates write better alternatives to. He sees a real market opportunity for a from-scratch modern operational database.
"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 [00:45:18]
"The storage engine that you would write as an undergraduate in a database course is better than Postgres' heap storage engine." — George Fraser [00:46:03]
Software Costs Are Too Small to Be a Real AI Optimization Target
The argument that enterprises will use AI to eliminate SaaS seats and cut software spend ignores how insignificant software already is as a cost center. Companies will use AI to grow revenue, not to shave a rounding error.
"If you look at the budgets of real companies that are heavy consumers of software, they spend 5% to 10% of headcount on software. Software costs are immaterial in the grand scheme of things... The idea that they're going to use AI to value engineer the number of seats they have on Slack or something is ridiculous." — George Fraser [00:09:17]
3. Companies Identified
Fivetran Data integration and pipeline company founded in 2013. Mentioned as a thriving incumbent that is accelerating (not declining) through the AI wave, with both OpenAI and Anthropic as customers. Actively using AI coding agents internally to improve connector quality at scale.
"We are actually starting to see new capabilities inside Fivetran trying to push the bounds of quality... you'll see the quality and reliability of Fivetran take yet another leap this year because of that." — George Fraser [00:32:46]
dbt Labs Data transformation tool that has merged with Fivetran. Identified as a major beneficiary of AI coding agents because coding agents are already writing large volumes of dbt models.
"I think coding agents are going to write tons of dbt models... There is this great quote from Dijkstra which is that 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." — George Fraser [00:41:38]
Anthropic AI lab and Fivetran customer. Noted because despite being at the AI frontier, their internal data infrastructure looks entirely conventional — a powerful signal that exotic new data architectures are unnecessary.
"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 [00:35:41]
OpenAI AI lab and Fivetran customer. Cited alongside Anthropic as evidence that even the most AI-native organizations still depend on traditional SaaS and conventional data infrastructure.
"As do we, as do OpenAI and Anthropic... who are both Fivetran customers. And we replicate lots of data from these very SaaS tools on their behalf into their data lakes." — George Fraser [00:09:51]
Nanobot Personal AI agent platform that George Fraser personally uses and prefers over OpenClaw and Nanoclaw. Highlighted for its clean separation of concerns — stable agent core with skills written on top.
"I actually churned off Nanoclaw onto Nanobot, which is what I've stuck with. But I use it to manage my tennis team... It has its own email. It's got its own WhatsApp number, its own email address." — Martin Casado [00:20:33]
4. People Identified
George Fraser Co-founder and CEO of Fivetran. Highlighted throughout for contrarian intellectual rigor, technical depth despite being a CEO, and disciplined M&A strategy without a formal corp dev function.
"One thing I do really appreciate about you as an executive is you as the founders, you're actually quite reflective. And one of the mental exercises that you've been doing as long as I've known you was like pretending if you were a new CEO brought in by the board to fix Fivetran, like what would you immediately unwind?" — Martin Casado [00:46:46]
Sirdar (CEO of Snowflake) Referenced as a benchmark for strategic decision-making. Fraser uses the mental model of "what would Sirdar do?" to pressure-test big, scary strategic moves like the dbt merger.
"One of the tricks I use is I ask myself, what should Sirdar do? Sirdar is the CEO of Snowflake, who we've worked with for a long time. It's a good mental exercise. And then I just go do that. And that was like a clear answer in my mind — merge with DBT, absolutely." — George Fraser [00:47:45]
5. Operating Insights
Use "What Would Another CEO Do?" to Overcome Strategic Paralysis
When facing a big, scary decision, Fraser reframes the question away from himself to remove personal fear and organizational inertia. By asking "what would Sirdar do?" he was able to clearly see the dbt merger as obvious — then execute it.
"When I was reflecting on that, on whether that was a thing we should seriously consider, one of the tricks I use is I ask myself, what should Sirdar do?... And then I just go do that." — George Fraser [00:47:45]
Treat Acquisitions as One-Off Events, Never Build a Corp Dev Machine
Fraser's counterintuitive approach: no corporate development function, every deal must feel like a "last one ever" with uniquely compelling reasons. This prevents deal-making for its own sake and ensures only high-conviction moves happen.
"Any acquisition or merger... should feel like it's for these unique reasons and it feels like it's the last one you're ever going to do. And it's not going to be the last one you ever do, but the reasons to do it should be really, really strong. You shouldn't go looking for this." — George Fraser [00:40:23]
Write Data Access Rights Into Vendor MSAs Before You Need Them
CIOs have significant contract leverage to guarantee data portability and access — but almost nobody uses it. Simply asking for data access language sends a market signal that shifts vendor behavior.
"Write language guaranteeing your own data access into those MSAs... 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 [00:18:44]
AI Coding Agents Are Best Applied to Long-Tail Maintenance, Not Net-New Features
Fivetran's most non-obvious AI use case is deploying coding agents as an "infinite supply of junior engineers" for continuous bug remediation across 750 connectors — a breadth of maintenance work impossible to scale with humans alone.
"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. Many, many, many small improvements start, we've seen the flood start to come." — George Fraser [00:34:15]
6. Overlooked Insights
A Greenfield SQLite-for-S3 Database Is a Real Commercial Opportunity Right Now
Buried in a casual aside about personal projects, Fraser describes building a from-scratch OLTP database where S3 is the backing store — essentially SQLite but cloud-native. He explicitly calls it a "real opportunity right at this moment" and invites database experts to come build it at Fivetran. Given AI workflows' need for vast numbers of lightweight, ephemeral databases, and his simultaneous indictment of Postgres as technically obsolete, this is a concrete product gap with a potential acqui-hire or venture opportunity attached.
"What it attempts to do is to be like SQLite except S3 is the backing store... Because when you build AI workflows, you have this need for like zillions of tiny databases... If you're hearing this and you want to work on this, and you are an expert in databases, come talk to me and maybe you can come do this at Fivetran. I think there's actually a real opportunity right at this moment." — George Fraser [00:44:23]
The Consumption Layer of Infrastructure Is the Only Layer AI Actually Threatens
In a brief but precise observation, Fraser draws a critical line: AI does not threaten data centers, cloud vendors, or core databases. It only threatens the highest-level "user-friendly" abstraction layer — the tools built to make infrastructure accessible to humans. This has enormous implications for where infrastructure investment is safe versus exposed.
"I think that last layer is the one that is threatened by AI. 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 [00:38:16]