AI coding bills hit $920 monthly per engineer, 10x spread between low and high users
## The Numbers Larridin, an a16z-backed AI measurement platform, published benchmark data tracking what enterprises actually spend on AI coding tools per engineer. The sample: engineers who merged code and drew billed AI spend across four weeks ending August 2025. Median weekly spend: $213 ($920 monthly). 90th percentile: $911 weekly. That is a 10x spread within the same organisation, using the same tools, at the same prices. Annualise the high end and you are near $47,000 per engineer in tokens alone. For a 100-person engineering team, the gap between your 25th percentile and 90th percentile users is a seven-figure line item that most finance teams cannot track today, because it is scattered across provider invoices, corporate cards, and personal subscriptions. ## Same Spend, Different Output The second finding matters more. Larridin split engineers by how much of their shipped code was AI-attributed. Both cohorts came from the same company, same tools, same baseline spend of roughly $170 weekly. The gap between their curves is fluency, not budget. Deeply AI-native engineers (79% AI-attributed output): reached 11.8x productivity at around $1,300 weekly spend and had not hit a ceiling. At equal spend, this group shipped roughly 2x what partial adopters shipped. Partially AI-native (37% AI-attributed): saw real returns, but marginal payoff dropped by half once spend passed $600 weekly. Low-AI engineers (15% or less AI-attributed): topped out around 1.9x output and stayed flat across a 20x spend range, from $21 to $421 weekly. Extra dollars bought activity, not output. Practical read for 2027 budgets: more budget converts to output only where fluency already exists. There is no universal ideal cap. Track your own team's ROI curve and set review triggers where it levels off. ## What This Means for Sales Orgs If your engineering team shows this kind of spend variation and output dispersion, your revenue org probably does too. AI SDR tools, email assistants, conversation intelligence platforms: same pattern. High-fluency reps extract more from the same tooling budget. The comp conversation shifts when you can connect AI spend to output per rep. If your top 10% of AEs are drawing $500 monthly in AI tool costs and closing 40% more pipeline than the median, that is not wasteful spend. If your bottom quartile is paying for seats they barely use, that is budget you can reallocate or kill. Larridin sells four products around one question: what is your AI spend producing? Spend Intelligence tracks token usage and agent costs. AI Impact connects team spend to output. Developer Intelligence ties engineering performance to AI tool use. Workflow Intelligence maps repeated work for automation candidates. The platform launched in 2024, raised $17 million seed from a16z, Bloomberg Beta, Gradient Ventures, and others. Founding team is experienced: CEO Russell Fradin previously built Dynamic Signal, Adify, and comScore. Current team sits around 20 people, roughly four sellers, tens of customers. Early enterprise motion, US-focused, no visible ANZ presence yet. Gainsight used Larridin to map internal AI adoption before buying its first enterprise LLM contract. The insight: you cannot procurement-plan what you cannot measure. Worth noting: output here is not lines of code. Each merged pull request is scored by model-assessed complexity, discounted for low quality and missing tests, scaled by code churn. Larridin flags its own limits, which is rare in this category. The relationships are associational: high-output engineers may spend more because they ship more, not the other way around. ## The Budget Reality If you are setting 2027 AI budgets without usage and output data, you are guessing. The median enterprise cannot tell you which teams use which AI tools, what those tools cost per person, or whether the spend correlates with any business outcome. Larridin is built to answer those questions before your next board deck. Agent spend is the fastest-growing and least-understood line in most AI budgets. It does not map to a seat or a person, it runs without oversight, and it scales independently of headcount. Finance teams are flying blind. This is the category that fixes that.