Cloudflare

Ramp benchmark: Leadership framed AI leverage as core to operating at Ramp and made adoption part of how the company works.

Diligence question: Has management made AI adoption an explicit company-wide expectation with visible follow-through?

Activist read-through: If leadership still frames AI as experimentation instead of expectation, the company is more likely to drift while AI-native peers compound.

Ramp benchmark: Ramp defined L0-L3 proficiency levels and used them to move employees from chat usage to production builders.

Diligence question: Is there a clear ladder from casual AI usage to system-building, with expectations rising as tools mature?

Activist read-through: Without an explicit proficiency model, investors should assume adoption is shallow and concentrated in a small technical minority.

Ramp benchmark: Ramp built Glass so people installed once, authenticated once, and started with a ready-made AI coworker rather than a pile of configuration steps.

Diligence question: Has the company eliminated setup friction so a non-technical employee can become productive immediately?

Activist read-through: If employees still need to build their own setup, the organization is capping AI leverage at the technical elite instead of the full workforce.

Ramp benchmark: Glass connected users to company tools on day one so Gong, Salesforce, Slack, Snowflake, and internal products were usable immediately.

Diligence question: Can non-engineers use AI against internal systems immediately through pre-connected tools and SSO?

Activist read-through: If access is gated by procurement, tickets, or connector work, the company is throttling internal compounding before it starts.

Ramp benchmark: Ramp’s Dojo turned discovered workflows into shared, Git-backed skills so one person’s win became everyone’s starting point.

Diligence question: Can one team’s AI workflow become every team’s baseline through an internal skills or workflow marketplace?

Activist read-through: Without shared skill distribution, learning stays local, reuse stays low, and the organization never compounds.

Ramp benchmark: Glass builds memory from connected tools and refreshes it continuously so users start each session with real context.

Diligence question: Does the company provide persistent memory and contextual awareness so employees get useful output without re-explaining everything every session?

Activist read-through: If context resets every session, output quality stays generic, trust stays low, and employees revert to old workflows.

Ramp benchmark: Glass supports scheduled automations, headless mode, and Slack-native assistants that run with the user’s memory, tools, and skills.

Diligence question: Can employees run AI workflows on schedules, in the background, or inside collaboration tools rather than only in live chats?

Activist read-through: If AI stops when the chat ends, the company has an assistant feature, not an operating layer.

Ramp benchmark: Glass uses split panes, persistent layouts, and inline rendering for files and outputs so the workspace matches real work.

Diligence question: Can employees work across files, documents, data, and multiple sessions without losing context or switching tools constantly?

Activist read-through: Weak UX keeps AI stuck in demo mode and prevents employees from incorporating it into complex, multi-step workflows.

Ramp benchmark: Ramp built Glass, Dojo, Ramp Research, and Ramp Inspect into a shared AI productivity layer rather than relying only on third-party chat tabs.

Diligence question: Has the company built or standardized an internal AI platform that knows company workflows, data, and systems?

Activist read-through: A company relying only on generic external chat tools is more exposed to slow diffusion and weak institutional learning.

Ramp benchmark: Ramp paired a small platform team with business-side builders across finance, ops, risk, CX, and sales.

Diligence question: Is there a central enablement layer plus distributed builders inside functions, or is AI still stuck in a centralized queue?

Activist read-through: Pure centralization creates bottlenecks; pure decentralization creates duplicated waste. Both create activist attack surface.

Ramp benchmark: Ramp used hackathons, Slack channels, office hours, all-hands demos, and visible builders to make AI building contagious.

Diligence question: Are hackathons, office hours, demo moments, and internal sharing channels driving visible cross-functional building?

Activist read-through: If the company cannot make AI usage socially visible, management is likely overstating internal adoption and pace.

Ramp benchmark: Ramp used company-wide leaderboards, team rankings, and usage visibility to accelerate adoption.

Diligence question: Does the company measure AI adoption and make the results visible enough to influence manager behavior?

Activist read-through: No instrumentation usually means no management system. Investors should discount broad AI claims when usage is not measured.

Ramp benchmark: Ramp added AI proficiency to hiring screens and talent management, including practical build exercises.

Diligence question: Has AI proficiency changed recruiting screens, onboarding expectations, and performance language?

Activist read-through: If the talent bar is unchanged, the company is importing yesterday’s operating model into tomorrow’s market.

Ramp benchmark: Ramp treated tokens as a learning budget and removed slow connector queues and role-based access ceilings.

Diligence question: Has management removed budget, token, and connector bottlenecks that block the first real AI win?

Activist read-through: A company that optimizes token spend before it learns to build is usually underinvesting relative to the leverage at stake.

Ramp benchmark: Ramp argues that internal AI productivity infrastructure is a moat and built Glass in-house to iterate quickly and learn faster.

Diligence question: Does management own enough of the AI harness to ship fixes quickly and turn internal usage into product and operating insight?

Activist read-through: If the company hands the harness entirely to vendors, iteration speed slows and internal AI leverage becomes less differentiated.

Ramp benchmark: Ramp expects internal AI tools to become obsolete quickly and treats that as a sign of progress rather than chaos.

Diligence question: Does the company sunset internal AI tools aggressively when better approaches appear, or defend stale tooling?

Activist read-through: If internal tools stay untouched for quarters, management is probably moving too slowly to defend margins against AI-native competitors.