App Software - Not All Tokens are Created Equal: A Framework for Token Intensity & The Evolving Inference Ecosystem

研报英文原文证据摘录

App Software - Not All Tokens are Created Equal: A Framework for Token Intensity & The Evolving Inference Ecosystem

Truist Securities

Equity Research Report July 5, 2026

TMT: Application Software and

Agentic AI App Software - Not All Tokens are Created

Equal: A Framework for Token Intensity &

Terry Tillman The Evolving Inference Ecosystem

404-926-5362 In this report, we introduce a framework for evaluating AI workloads based on token

Terry.Tillman@truist.com

intensity, categorizing workflows as token light, token moderate, and token heavy,

while examining how the broader AI inference ecosystem is evolving to improve the

Connor Passarella economics of AI. We believe that agentic workflows, particularly software development,

212-326-6153 represent a fundamentally different class of inference demand, with autonomous tasks Connor.Passarella@truist.com

capable of consuming materially more (~1,000x) tokens than traditional AI assistants.

We believe value will increasingly accrue to software vendors that maximize business

Giancarlo Valle outcomes while minimizing inference costs.

212-537-9006

Giancarlo.Valle@Research.truist.com

Across our public company coverage, we see AI and agent adoption measured today

Luc Radice in a variety of ways; however, often a blend of metrics converge on the same idea:

212-537-9032 some kind of unit of AI work. For example, Salesforce (CRM, Buy) talks about agentic

Luc.Radice@truist.com work units (AWUs), Microsoft (MSFT, Buy) talks about Copilot users, agents deployed/

managed, and first-party agent monthly active usage, and ServiceNow (NOW, Buy,

7 Page Document Jump) talks about “assists” when measuring AI consumption across its customer base.

These metrics and measures represent an attempt to quantify the value from AI and