Yamini Rangan stood at a podium in early May and named a price: fifty cents per resolved conversation. HubSpot's AI agent would now do the work that thousands of partner agencies bill for, at a rate no human can match. The stock fell 19% in a single session. But the operators who should have been paying attention were not the ones watching the ticker.
They were the ones sending invoices for CRM configuration.
"Build your business on the platform your clients use." If you run an agency or a consulting practice tied to a software vendor, you have heard this line. It sounds right. For years, it was profitable. You set up workflows, trained client teams, built automation sequences. The platform grew. Your billable hours grew with it.
Then the platform priced that work at fifty cents and called it a feature.
The Pattern Has a Name
The evidence suggests something most people find uncomfortable: this is not new. The pattern has fired before, and it left a trail you can trace.
Mark Pincus built Zynga on Facebook. Every game, every user, every dollar ran through Facebook's rails. For two years, it looked like genius. Then Facebook forced all game developers onto its own payment system and took a 30% cut of every transaction. Zynga's stock went from ten dollars at IPO to $2.32 within a year. A 76% fall. Wharton published a one-line verdict: the business had one fatal flaw, its total dependence on Facebook's platform and user base.
Clayton Christensen named the law behind this pattern in The Innovator's Solution. He called it the Law of Conservation of Attractive Profits. The law states that when one layer in a value chain becomes a commodity, the profit does not vanish. It migrates to an adjacent layer the commodity cannot reach.
The platform eats the task. The margin moves to the judgment the platform cannot perform.
The platitude does not fail people. It fails the system they are trying to run. It places their billable hours inside the exact layer that the platform is working to compress.
Where the Flaw Sits
The Zynga story and the HubSpot shift share the same structural flaw. Both rewarded builders for going deeper into the platform layer instead of building adjacent to it.
HubSpot's AI agent now resolves 65 to 70 percent of customer conversations. It cuts resolution time by 39 percent. The old price was one dollar per conversation, win or lose. The new price: fifty cents, charged only when the AI solves the problem. For comparison, Salesforce charges two dollars per conversation for its own AI agent. The compression is not one vendor's choice. It is an industry-wide pattern.
MarTech put the result in plain terms: the decline of the generalist implementation partner. The platforms are pushing service partners away from setup and configuration work, toward the judgment work that the AI cannot do on its own.
The problem is not that you built on a platform. The problem is that building on the platform put your hours inside the layer the platform just learned to eat.
The Five-Column Audit
The better principle: map every billable hour against the law. Which hours sit in the layer the platform just consumed? Which sit in the layer it cannot reach?
Once that is clear, three moves follow from it.
Move 1: List Every Task You Bill For
Open a blank sheet with five columns. Column one: the task name. Column two: what you charge. Column three: can the platform's AI do this now? Column four: can it do this within twelve months? Column five: does the task require your judgment about a specific client outcome, or does it follow a set of repeatable steps?
This is the move that costs something. Not money. Honesty. Most of what fills a service menu started as judgment work and quietly became configuration work over the years. The audit forces you to see which column each hour actually sits in.
Move 2: Mark the Dead Column
Any task where column three or four says yes, and column five says repeatable, sits in the dying layer. You can still bill for it. But the price ceiling is falling, and it will keep falling. Only 9,000 of HubSpot's 299,000 customers use the AI agent so far. That is 3 percent adoption. The window to rebuild your service menu is open. It is not permanent.
Move 3: Name the Adjacent Layer
For every dead-column task, ask one question: what judgment call does the client still need a human to make after the AI runs? That is the live layer. Outcome design. Performance interpretation. The decision about what the data means for this specific client. The AI resolves the ticket. You design the system that decides which tickets matter and what to do with the ones the AI cannot close.
What the Audit Shows
Running this for one billing cycle does something the platitude never did.
It shows which hours are already priced at fifty cents, even if your current invoice says more. It shows which clients pay for your judgment and which pay for your keystrokes. It shows where the margin moved, not where it died. And it shows the gap between what you sell and what actually requires you to be in the room.
The Feedback Loop
At the end of one billing cycle, ask three things.
→ Which hours moved a client outcome that the AI could not have reached on its own?
→ Which hours looked like progress but left no trace once the platform caught up?
→ Which friction showed up more than once, pointing to a judgment gap the AI will not close?
That is the difference between advice that sounds right and a system that proves itself.
Attio - the AI CRM for modern businesses.
Attio is the AI CRM that keeps you ten steps ahead.
Ask Attio anything. Where should I focus? What deals are at risk? Search, update, and create across your customer data.
Ask more from CRM. Ask Attio.
Where You Stand
Rangan set a price in early May. The price that matters is not what happened to the stock. It is the price the platform just placed on the hours you bill. Christensen's law says the margin did not die. It moved. The audit shows which side of that line your hours sit on.
