Agents Are Eating the System of Record. But the Real Feast Is in the Field.

I read Jason Lemkin's latest piece on The Agents and couldn't shake one thought: everyone is watching agents eat the system of record, but the real feast is somewhere else entirely.

Lemkin's team saw their Salesforce storage balloon from 5GB to 40GB in 30 days. Roughly 21 million records. And here's the kicker — they didn't write any of it. Their agents did. Salesforce flagged them for overages. They hadn't logged in for a week.

That's not just an amusing anecdote. It's evidence that the economics of the CRM are fundamentally broken in the agentic era. When an AI agent writes records constantly — task updates, email opens, call metadata, enrichment — the storage meter spins like a slot machine. And the per-record costs we tolerated when humans clicked a few dozen times a day become absurd when software is doing thousands of writes per hour.

Lemkin points out that the same 40GB is a rounding error in Postgres. He's right. The pricing model for systems of record wasn't built for this. And that's creating an opening for a whole new class of infrastructure — and a whole new class of opportunity for builders who can see past the CRM.

But here's what caught my attention: the entire conversation — Lemkin's, and most of the industry's — is happening inside the digital layer. Salesforce, Postgres, Clay, MCP. These are all software tools eating software data. And yet, when I look at the data we track at PainSignal, the loudest pain signals are coming from the physical world.

We track 24,609 problems across 88 industries. That's not a marketing stat; it's a map of where operators are actually hurting. And the industries at the top of that list aren't SaaS companies. They're construction, trucking, HVAC, plumbing.

Take construction: 733 distinct problems tracked, 445 app ideas generated just from those pain points. Trucking and logistics: 929 problems, 290 app ideas. These aren't abstract complaints. They're specific, recurring, expensive frictions — document chaos on job sites, driver fatigue reporting, compliance paperwork that never ends, dispatch coordination that still runs on phone calls and whiteboards.

Lemkin's renewal agent builds custom decks for sponsors. That's impressive. But the same agentic architecture — pull data from multiple systems, reason about the right narrative, generate a tailored output, deliver it through the right channel — could just as easily build a daily safety briefing for a construction crew, or a compliance packet for a trucking company's DOT audit, or an automated follow-up for an HVAC company's service calls.

And here's the kicker: the physical world doesn't have a Salesforce. It has a patchwork of point solutions, spreadsheets, and gut feel. The system of record for a plumbing company is often a whiteboard in the shop and a bunch of text messages. That's not a data storage problem. It's a data absence problem. And agents thrive where data is missing but workflows are predictable.

The opportunity for indie hackers and seed investors isn't to build another CRM plugin. It's to build the agent that becomes the system of record for a field service business — before that business even knows it needs one. The agent that answers the phone, schedules the job, logs the hours, files the paperwork, and sends the invoice. The agent that sits on top of QuickBooks and ServiceTitan and Google Calendar and just quietly runs the back office.

That's the flip side of Lemkin's ServiceTitan-Podium story. Both companies are fighting over who owns the record when an agent does the work. But in plenty of field service niches, nobody owns the record yet. That's greenfield territory for a builder who understands the domain.

We've seen app ideas like BuildSync Pro emerge directly from construction pain points — project management that actually understands how a job site operates, not just another Gantt chart. These ideas come from the data because the pain is so severe and so consistent. And that's before anyone even tries to apply agentic workflows to them.

The AI agent revolution isn't going to be won in the CRM. It's going to be won in the field. The builders who understand that — who go talk to a roofing company owner or a fleet manager and ask what they actually spend their day doing — are going to find problems that make Salesforce overages look trivial.

Lemkin's article is a great read for what it says about the system of record. But the real lesson is between the lines: the agent doesn't care where it lives. It will go wherever the work is. And right now, a lot of the work is happening outside the dashboard.