Ramp dropped a study this week showing firms that adopted AI grew headcount 10.2% in the two years after. It's already making the rounds as proof that AI hires instead of fires. Open the actual paper and the story turns: low-intensity adopters saw no change at all. The entire gain came from heavy spenders, around $30 per employee each month. So the founder takeaway isn't "use AI." It's that half-measures move nothing. Two things to hold onto. The adopters were already larger and growing faster before they spent a dollar, so this is correlation wearing a causation costume. And Ramp sells the card that tracks the spend, worth remembering when a vendor's data tells you to buy more. The line I'd underline: small firms adopt AI the least, but when they actually commit, the return per dollar is the highest.

Read the study →

Per-seat SaaS pricing is breaking. AI agents don't log in or take a seat, so charging per human user stops mapping to value the moment one agent does the work of ten. By one widely cited estimate, pure seat-based pricing fell from 21% to 15% of SaaS companies in a single year, with hybrid models now the default. This hits you twice. As a buyer, your vendors' price increases are getting repackaged as "AI features" at renewal. As a founder, your own pricing model is now a live strategic question, not a settled one. Read more →

Defense is becoming one of AI's clearest budget lines. Dominion Dynamics, founded by a former Anduril executive, just raised $100M USD, the largest Series A in Canadian defense history, to build command-and-control software and a robotic wingman drone for the Arctic. Ugly but obvious truth: the easier the buyer can name the mission, the faster the budget appears. That rule applies to your buyers too. Read more →

Enterprise AI has crossed into the opex column. West Monroe found 85% of executives plan to raise IT budgets next year and 91% say AI is driving tech spend up. But the same reporting shows those budgets now face line-item scrutiny they didn't a year ago. Rising budget and rising scrutiny are not a contradiction. They are what happens when AI graduates from experiment to expense. Read more →

Why you should care: "We use AI" is no longer a strategy. Buyers are asking where it saves money, where it makes money, and what it costs per workflow. The founders who cannot answer those three questions are about to get filtered out, on both sides of the table: when they buy tools and when they sell theirs.

As of April 2026, HubSpot charges $0.50 per resolved conversation for its AI support agent. Intercom charges $0.99. Both took the seat out of the equation and bill by outcome. Read more →

The optimistic read: pricing is finally honest. You pay when the software actually works. But billing by outcome means the outcome gets measured, and once it is measured, every workflow has to justify its cost per use. The pricing page just became a P&L.

"AI spend is moving from possibility to accountability."

Try this Monday: Pick one AI feature, workflow, or internal tool you are excited about. Write down three numbers: cost per use, time saved per use, and the revenue or expense line it moves. If you cannot fill in all three, you do not have a business case yet. You have a demo.

Langfuse: tracks LLM usage, prompts, traces, evaluations, latency, and cost. Useful once your AI product crosses from prototype to production and you need to know what each workflow actually costs to run, not what you hope it costs. Read more →

Reply and tell me one thing in this issue worth keeping and one worth cutting. I read every response, and it is how the Brief gets sharper.

— PingMunk