Anthropic, Atlassian and Scale VP Converge on 'Build the Skills Layer, Not a New Stack' — With $688B Capex Against $110B Revenue as the Backdrop

SaaStr·high signal

In a July 25 SaaStr writeup of SaaStr AI 2026, Anthropic's Eleanor Dorfman ('not a new stack' — be 'intentional about the existing one'), Atlassian's Sharif Mansour, and Scale Venture Partners' Rory O'Driscoll all landed on the harness/skills layer as the actual moat rather than the model. Hard numbers: O'Driscoll put hyperscaler AI capex at $688B against $110B of revenue with no capex breakeven until 2031–2032, and said foundation models need 15–17% of knowledge-worker wages (25%+ for developers) to pencil; Anthropic reported 54% of new enterprise logos arriving self-serve in 2026 with AEs previously burning 70% of their time on internal process; Atlassian reported 5M+ users on AI features and that juniors are 19–30% more likely to use AI while seniors are twice as likely to refine output. Anthropic threads Claude through Clay, LeanData, Salesforce, Gong, Slack and Intercom — an AI lab keeping its SaaS stack, which is the counter-evidence to the rip-and-replace thesis.