AI token costs are breaking engineering budgets. Over the past 2-3 months, spending on AI agents has exploded across the industry, and The Pragmatic Engineer sourced details from 15 companies to document exactly how bad it has gotten and what teams are doing about it. This is not a rounding error. It is a structural cost problem that engineering leaders did not see coming.
The supply side is cracking under the same pressure. GitHub Copilot and Anthropic are actively throttling individual users to protect capacity for business customers, whose spend has increased roughly 10x in recent months. OpenAI and its Codex product are the notable exceptions so far. The original piece maps which vendors are pulling back and what that means for teams dependent on them.
The third thread in this piece is Meta: business metrics are strong, but internal morale is at a reported low. Developers are facing looming layoffs alongside a new employee tracking program rolled out to all US staff. Read the full piece for the vendor-by-vendor breakdown, the 15-company cost data, and the specific coping strategies teams are testing right now.
[READ ORIGINAL →]