Share with your CTO
Gartner is warning that AI coding tool spending per developer is on track to exceed the average software developer’s salary by 2028, driven by token-consumption pricing models that scale directly with usage. Nearly 25% of tech leaders already spend $200 to $500 per developer monthly on AI coding tokens, and 6% exceed $2,000. Tools like Claude Code, Cursor, and OpenAI Codex bill per token rather than per seat, meaning every prompt, context window, and autonomous task run adds to the invoice in ways traditional software budgets weren’t built to absorb.
What this means for your business
The cost structure your engineering org inherited from SaaS doesn’t apply here. A seat license is a fixed cost you forecast once. Token consumption is a variable cost that scales with developer behavior, and developers, left ungoverned, optimize for speed rather than efficiency. That’s not a criticism of developers. It’s how incentives work. The problem is that the people generating the spend aren’t the people watching the bill.
The architectural mistake most teams are making right now is treating AI coding assistants as drop-in replacements for search or autocomplete, when they’re actually running compute jobs priced like cloud infrastructure. Oversized context windows, the practice of feeding an AI model far more code and documentation than a task requires, are the clearest example: developers paste entire codebases into prompts because it’s easier, and every unnecessary token costs real money. Gartner’s framing of “context engineering” as a discipline isn’t soft advice. It’s the same cost-hygiene logic that drove FinOps adoption after AWS bills spiraled.
The signal worth watching: vendor pricing transparency. Gartner specifically notes that many AI coding platforms provide limited visibility into how token usage is measured or billed. That opacity is a governance failure waiting to become a budget crisis. Any vendor unwilling to give your engineering org clear, queryable usage data at the developer and project level should be treated as a procurement risk, not a productivity partner. The teams that build metering and escalation policies into their AI tooling contracts now will have significantly more leverage when consumption scales.
Concept deep-dive: Token consumption pricing
Tokens are the units AI models use to process text: roughly 750 words equals about 1,000 tokens. Every input you send and every output you receive gets counted and billed. The model exists to serve the vendor: unlike a SaaS seat that costs the same whether you use it daily or once a quarter, token pricing means a hyperactive agent running overnight debug loops can generate hundreds of dollars in a single sprint. Think of it as cloud compute, not software licensing. The business implication is that usage governance isn’t an IT courtesy, it’s a cost control function.
Based on reporting from AI Coding Costs Could Exceed Developer Salaries, Gartner Warns, originally published 2026-06-30 09:48:00.
