The Bill Always Comes Due and Nobody Reads It

The Gartner report is sitting open in front of me and I've been staring at it long enough that the words have started to blur into the larger shape of what they're actually saying, which is: we bought the vision on credit and the statement just arrived.

The pitch was clean. AI-assisted development, agent-driven workflows, productivity gains that would justify everything. And the productivity gains are real, I'm not disputing that, I've seen them. But there's a thing that happens in every technology cycle where the people selling the shovels are very good at talking about gold and very quiet about the weight of the shovel, and we are deep in that moment right now. The pricing model underneath all of this is not like any tool economy we've built before. You used to buy the thing and then use it, and the marginal cost of using it was basically nothing. The hammer doesn't charge you per nail. But the new model charges you every time the tool thinks, and the more it thinks, the more it costs, and the more complex your workflow, the more it thinks, and nobody put that in the deck when they were selling the transformation. It's a tax on intelligence, billed monthly, and the meter was running before anyone built the governance layer to watch it.

That governance layer is the part that keeps me up. Not the cost itself, not even the technology, but the specific conversation that's coming in organizations everywhere, where someone who trusted the pitch has to sit across from someone holding a budget and explain that yes, the productivity gains are real, and yes, the tooling costs might exceed what you're paying the developers the tooling is supposedly helping, and no, there isn't a clean answer because nobody built the safety rails, because safety rails don't increase token consumption, and the vendors' incentive structure doesn't actually include your margin as a design consideration. The alarm is downstream from the sale. The panic is not a bug.

I've been in IT long enough to have watched this specific movie in several genres. The cloud promised to make everything cheaper and it did, eventually, after the wild years of untagged resources and surprise invoices and the guy whose Lambda function went recursive and cost seventeen thousand dollars over a weekend (we don't talk about that guy, but we all know that guy). The SaaS explosion promised simplicity and delivered a subscription stack that requires its own management layer. Every wave of genuinely useful technology arrives with an economics problem strapped to its back, and the economics problem is always the boring part, the part that doesn't get funded until the panic makes it unavoidable. Rate limits. Cost allocation tags. Governance policies that someone has to write and someone has to enforce and nobody wants to own because it's not in anyone's OKRs. The organizations that survive the next two years of this are not going to be the ones who deployed the most agents or bought the most tokens. They're going to be the ones who did the unsexy work first, who sat down with spreadsheets and built the operating model before the bill made thinking straight impossible.

The $20,000 developer month number in the report is going to do a lot of work in a lot of boardrooms, and I genuinely don't know if it's the norm or the outlier, but it doesn't matter, because someone will read it and make a decision about their team, and that decision will be made without the full picture, because the full picture requires understanding the cost-per-unit-of-value math that nobody has been incentivized to teach you. The vendors will keep selling the vision. The analysts will keep sounding the alarm in language careful enough not to lose the vendor relationships. And the rest of us will keep paying, which is fine, that's the cycle, I've made my peace with the cycle. What I haven't made peace with is the part where the developers whose salaries might be exceeded by their tooling costs have to explain the math to people who were promised magic and are now being handed a bill that looks nothing like the estimate.

The only move I know, the one that's worked every time I've watched an industry hit this particular wall, is to build the boring stuff before the panic. Document the cost model now, while you can still think. Build the governance layer before the agent calls are running wild across seventeen workflows nobody fully mapped. Treat every context window refresh like the meter it is. This is not glamorous work, it won't make it into the conference talk, and the vendors will not help you do it because it is not in their interest for you to do it, but it is the only thing standing between you and the conversation where you have to explain to someone that the future arrived and it's more expensive than the brochure suggested.

I still think the technology is real. I still think the productivity gains are real. I'm not here to tell you the whole thing is a scam, because it isn't, and I've seen enough genuine wins to know better. But I've also been around long enough to know that real technology with broken economics is just a faster way to the same bad place, and right now, everyone is paying for magic on credit, and the interest rate is whatever the vendor decides it is next quarter. Build the boring stuff. Read the bill before it reads you.