系统之记录仅保留存，增长需靠数据层与AI代理策略

The other day ServiceTitan shut off Podium’s integration for roughly 1,000 shared customers after a nine-year partnership. The coverage mostly framed it as a breakup. It works better as a case study in what owning the record can and can’t do for you.

ServiceTitan delisted a partner with $100M in AI agent ARR, cut off 1,000 shared accounts, and kept essentially all of those accounts. The contractors stayed because their jobs, invoices, customer history, and technician schedules all live in ServiceTitan. Podium was the removable piece. That is what being the System of Record buys you.

What it didn’t buy was a change in the growth rate. ServiceTitan grew 25% last quarter, which is a very good number and roughly the best case for a vertical System of Record in 2026. Snowflake grew product revenue 34% with 126% net revenue retention. Databricks crossed $7B ARR growing over 80%.

Owning the record is a retention asset. The growth has to come from somewhere else, and the gap between the two is widening every quarter.

Why Being a System of Record Alone … Isn’t Enough:

• Systems of Record convert into retention, not acceleration. ServiceTitan can cut off a partner, lose nobody, and still grow 25%. Salesforce’s applications business, the actual System of Record, grew 7% in constant currency last quarter.
• Systems of Record are the most expensive databases most companies own. Salesforce’s list price for additional data storage is $125 per month per 500MB, or $3,000 per GB per year. The same gigabyte on S3 costs a fraction of a cent per month.
• Agents create, consume, and touch far more data than humans ever did. Salesforce processed 28.6 trillion tokens for customers, up 152% quarter over quarter. Data 360 ingested 52 trillion records in a single quarter, up 136% year over year. None of that is human-scale volume, and none of it is priced for.
• Salesforce already conceded the point. Of those 52 trillion records, 35 trillion came in via Zero Copy, meaning they never physically moved into Salesforce. Zero Copy volume grew 277% year over year. The fastest-growing data motion at the largest System of Record is the one where it doesn’t hold the data.
• Owning the record doesn’t win you the agent sale. Podium built $100M in AI agent ARR in under 24 months, much of it inside ServiceTitan’s own customer base. Open platforms lose agent deals to specialists. Closed ones eventually lose the customer.
• API limits bite before storage costs do. Salesforce allocates API calls per user per day. Agents don’t have users. ServiceTitan’s April 2026 API terms go further and bar AI systems from independently choosing endpoints.

#1. What the Podium fight actually proved

ServiceTitan’s Marketplace policy, published in June 2026, welcomes competitive partners as long as they aren’t using the partnership and the support that comes with it to gradually displace parts of ServiceTitan. Its API terms from April 2026 require calls to stay inside a predefined certified scope and bar AI systems from choosing endpoints on their own.

Together those two documents describe a perimeter. ServiceTitan is defending the ground around the record, and it has the leverage to do it, because the alternative for a contractor is ripping out the thing that runs their business.

That leverage is worth a lot. It shows up as high renewal rates, as pricing power, as net revenue retention above 110%. It is also almost entirely defensive. Cutting off Podium doesn’t add a dollar of new revenue to ServiceTitan. It prevents the loss of some future dollars. Those are different businesses.

Every System of Record carries the same asymmetry. The switching cost that keeps customers in is a separate asset from the thing that makes them spend more.

ServiceTitan Just Shut Off Podium’s Integration for ~1,000 Shared Customers. Why? Agents Turned a 9-Year Partner Into a Direct Competitor.

#2. Retention is the floor, not the engine

Look at what the Systems of Record are printing.

Salesforce reported $11.13B in Q1 FY27, up 13%. Strip out the Informatica acquisition and organic growth lands closer to 8-9%. More telling is how the company now segments itself. Agentforce Apps, which is sales, service, marketing, commerce, and Slack, came in at $6.91B and grew 7% year over year in constant currency. Data 360, Headless Platform, and Other went from $2.95B to $3.68B.

The record layer grew 7%. The data layer grew 25%. Same company, same customers, same quarter.

Veeva, one of the most locked-in Systems of Record in existence, grew 16% last quarter while running a full CRM migration underneath its customers. Salesforce management flagged lower attrition as a driver of net new order value. Retention is holding up fine.

Now the layer underneath. Snowflake: product revenue up 34%, NRR 126%, RPO up 38%, 779 customers over $1M. Databricks: over 80% growth at $7B ARR.

Customers aren’t leaving the System of Record. They’re spending their incremental dollars somewhere else.

#3. Systems of Record are extremely expensive databases

Salesforce charges roughly $125 per month for an additional 500MB of data storage. That works out to $250 per GB per month, or $3,000 per GB per year. File storage is cheaper at about $5 per GB per month, which is still $60 per GB per year.

One 2026 analysis put it at $15,000 a year for 5GB of Salesforce data storage against under $30 a year for the same 5GB on S3.

Those aren’t the same product, and Salesforce would rightly say you’re paying for a governed, permissioned, audited, workflow-attached record rather than for bytes. That’s a fair defense of the price. It stops being a defense of the architecture once agents show up.

The price was set for a world where data volume is bounded by how fast humans can type. A rep creates a few dozen records a day. A support agent logs a case. A field tech closes a job. At that volume nobody notices they’re paying four orders of magnitude over commodity storage, because the absolute number stays small.

Agents are not bounded that way.

#4. Agents produce an absurd amount of data, and much of it doesn’t have to be in a System of Record

Run a real agent workload and look at what it generates. Tool call traces. Intermediate reasoning. Embeddings. Retrieval logs. Scoring runs. Evaluation results. Failed attempts. Retries. Every one of those is data the agent creates, keeps, and reads back.

In our own Connect build, a single scoring pass writes 164,523 rows. That’s one job, on 14.5k profiles. We’re importing 250k profiles next, roughly 17x the scale. That data has to live somewhere cheap enough that I can re-run the whole thing when a scoring rule turns out to be wrong, which happens regularly.

Almost none of it is a business object. It’s the exhaust of the agent doing its job: enormous, mostly disposable, read far more often than written.

Salesforce’s own disclosures show the scale. 28.6 trillion tokens processed to date, up 152% quarter over quarter. 3.8 billion Agentic Work Units delivered, up 111% quarter over quarter. Data 360 ingesting 52 trillion records in a single quarter, up 136% year over year.

Now set that against a 7% growth rate in the applications business. Data volume moving through the platform grew 136%. System of Record revenue attached to it grew 7%.

#5. The API limit hurts before the storage bill does

Storage you can at least pay for. API access is often capped regardless.

Salesforce editions come with API call thresholds allocated per user per day. That made complete sense when API calls were a proxy for integrated human work. It breaks immediately with agents, because agents don’t have seats. The more agentic the workload, the faster the ceiling arrives, and the only lever is buying seats nobody sits in.