Leading Firms Build Institutional AI for Growth, Not Efficiency
Top 20% of companies capture 75% of AI gains by using AI for business reinvention (2-3x more likely), building org-wide harnesses like RAMP's Glass to turn every employee into a power user.
AI Leaders Pursue Growth Through Business Transformation
The top 20% of companies capture 75% of AI's economic gains by treating AI as a growth and opportunity technology, not just efficiency tool. Per PwC study, these leaders are 2-3x more likely to use AI for identifying growth opportunities and 2.6x more likely to reinvent business models. McKinsey's analysis of 20 AI leaders across industries shows AI-driven transformations deliver 20% EBITDA uplift, break-even in 1-2 years, and $3 incremental EBITDA per $1 invested. They target economic leverage points—like Toyota's AI supply chain breakthroughs in automotive—rather than broad productivity gains. McKinsey emphasizes building enduring capabilities around AI, not just deploying tools: senior business leaders must gain AI muscle (combining domain expertise with AI know-how), keep 70%+ of AI talent in-house, treat tech platforms as strategic assets fed by ongoing data enrichment, and master agentic engineering (ingesting unstructured data, automating guardrails, codifying playbooks). Speed is key, as skill half-lives shorten, making rapid experimentation essential.
Institutional AI Scales Individual Gains Organizationally
Individual AI boosts personal productivity 10x, but no company has seen 10x valuation—requiring institutional AI to align efforts. George Sulka's a16z piece outlines seven pillars distinguishing them: coordination prevents chaos from misaligned AI outputs (e.g., varying prompting styles creating disorganized flows); signal extraction amid content explosion; professional objectivity for revenue scaling over time-saving. Unlike individual AI focused on efficiency, institutional AI solves coordination problems AI creates, like thousands of agents rowing in opposite directions without defined roles/OKRs. Leading firms build systems ensuring one person's AI breakthroughs become team baselines, evolving harnesses with persistent memory, integrations, and marketplaces.
RAMP's Glass: Harness Engineering Makes Everyone a Power User
RAMP built Glass, an internal AI workspace, because models suffice but setups fail—only 9% initially used AI daily due to painful configs. Deployed day-one via SSO, it auto-integrates tools (e.g., Ramp Research, Inspect, CLI; external like Gong, Salesforce, Slack, Notion), enabling seamless workflows like pulling call context to draft follow-ups. Core principles: (1) Preserve full upside—expose multi-window workflows, automations, memory without dumbing down, as AI tutors help users scale complexity. (2) Propagate breakthroughs—350+ reusable 'skills' (agentic markdown tasks, e.g., Zendesk investigations pulling tickets/account health) shared via Dojo marketplace. AI 'Sensei' recommends top 5 skills by role/tools/history. (3) Product as enablement—targeted nudges teach via use. Memory synthesizes user data daily (people/projects/Slack/Notion/Linear/calendar), reducing re-explanation. Features mimic frontier tools: scheduled cron automations posting to Slack (runs offline). Built in-house for moat (internal productivity edge), speed (same-day fixes via Slack triage), and product insights (informs customer finance tools). Result: New hires access team context instantly; non-engineers run ex-engineer tasks. Learning happens via doing—skills demo great outputs, memory shows context value—raising org floor without lowering ceiling, turning agentic engineering into everyone’s capability.