Breaking

Cognition Grades Engineers on AI Efficiency; Devin Spending Limits Rise With Performance

October 19, 2026 at 12:49 EDT

Foundation Models
Sales & Business
Software Dev & Coding

Cognition AI has introduced an efficiency-grading system that ties Devin usage allowances to engineer productivity, rewarding top performers with automatic spending increases while pairing the tool with Devin Coach, an AI assistant designed to help teams optimize their AI workflows. The approach, which went live with enterprise customers in mid-2026 and was publicly detailed in October 2026, marks a shift away from blanket usage caps—which the company characterizes as hidden hiring constraints—toward performance-based resource allocation.

AI NEWS • OCTOBER 2026

Cognition Rewards Efficient Engineers With Auto-Scaling Devin Budgets

Performance-based spending limits replace blanket caps, tying AI compute to measured engineering output.

How Efficiency Scoring Works

Each Devin session is converted to estimated human-engineering hours, then divided by token and compute costs. Engineers scoring "good" or "satisfactory" see automatic increases to their monthly ACU allowance—those below threshold maintain current limits. This reverses traditional "tokenmaxxing" where high token use was the success metric, now treating it as inefficient without proportional output.

Validation & Precision

Tested on: 126 engineers across 258 sessions

Correlation: 0.74 log score vs. user estimates—reasonably strong but individual sessions often miss by 2–3×

Limitation: Estimates measure task completion, not business impact, which varies by project context

Upside: Dynamic Allocation

Productive engineers can scale their AI investment without arbitrary caps. Enterprise administrators set per-engineer monthly limits tied to measurable output, not raw token volume.

Challenge: Precision & Adoption

Individual accuracy varies (2–3× margin), and estimated hours don't reflect actual business value. Risk-averse teams may hesitate to request higher limits even if they qualify.

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.