How the Best Companies Really Use AI (And Why Everyone Can Be a Power User)

Some companies are quietly pulling far ahead with AI. They’re not just saving a few hours with chatbots—they’re redesigning how their entire business works. The gap between these AI leaders and everyone else is getting wider every quarter.

This article breaks down what those leading companies are doing differently, why individual AI use isn’t enough, and how one standout example, Ramp’s internal AI platform "Glass," shows what the future of AI-native organizations looks like.

AI Leaders vs. Laggards: It’s About Growth, Not Just Efficiency

Recent research from firms like PwC and McKinsey shows a clear split: a small group of companies is capturing most of AI’s economic gains. PwC found that about 20% of companies are capturing roughly three-quarters of the value from AI.

What separates these leaders isn’t just that they “use more AI.” It’s how they think about it:

1. AI as a growth engine, not just a cost-cutter. Leading companies use AI to find and pursue new revenue opportunities, not just to automate existing tasks. They’re 2–3x more likely to say AI helps them identify growth opportunities and reinvent their business models.

2. Real business impact, not experiments that go nowhere. McKinsey’s analysis of top AI performers found that AI-driven transformations delivered, on average, a 20% uplift in EBITDA, broke even in 1–2 years, and generated about $3 in incremental EBITDA for every $1 invested.

3. Systems, not scattered tools. Technology alone doesn’t create advantage. The leaders build enduring capabilities: processes, platforms, and habits that let them repeatedly turn new AI capabilities into business value.

In other words, the winners treat AI as a core part of their business model, not a side project or a set of disconnected tools.

From Individual AI to Institutional AI

Most people now have access to powerful AI tools. Individually, that’s a huge productivity boost. But that doesn’t automatically translate into a 10x better company.

George Zarkadakis, writing about “institutional AI,” makes a key point: simply adding up everyone’s personal AI usage does not create an intelligent organization. In fact, it can create chaos.

Imagine doubling your headcount overnight with clones of your best employees. If no one coordinates them—no clear roles, goals, or communication—you don’t get twice the output. You get confusion.

The same thing is happening with AI today:

• Every employee has their own ChatGPT or Claude habits.
• Everyone has different prompting styles and workflows.
• Outputs live in different tools and don’t connect to each other.

On paper, individual productivity goes up. In practice, the organization can stall because there’s no coordination layer that aligns all this AI-generated work toward shared goals.

Institutional AI is about building that layer. It means:

• Coordinating human and AI agents so they’re not “rowing in opposite directions.”
• Filtering signal from the noise of all the new content AI generates.
• Focusing on outcomes that matter to the business—like revenue and customer value—rather than just time saved.

This is why the best companies are investing in systems that organize, share, and standardize how AI is used across the org.

What McKinsey Says AI Leaders Do Differently

McKinsey’s AI transformation manifesto highlights several patterns that show up again and again in AI-leading organizations. A few stand out:

1. They build capabilities, not one-off projects. Leaders don’t just “deploy a model.” They build repeatable ways to turn new AI capabilities into products, workflows, and decisions. AI is treated as a strategic asset, not a gadget.

2. They focus on economic leverage points. Instead of sprinkling AI everywhere, they ask: “Where in our business model does improvement matter most?” In automotive, for example, supply chain integration is a huge leverage point, so that’s where AI goes first.

3. Senior leaders learn AI, not just IT. It’s not enough for the IT team to understand AI. Business leaders who own P&Ls and functions need to understand what AI can do so they can redesign processes, products, and even pricing around it.

4. Talent is mostly in-house. McKinsey argues that more than 70% of AI talent should be internal. Consultants can help, but transformation is ultimately a people change inside the company.

5. Data becomes an ongoing discipline. For many lagging organizations, data is the bottleneck. Leaders treat data as a product: they invest in cleaning it, enriching it, and maintaining it as a continuous operational practice, not a one-time project.

6. Agentic engineering is the next big capability. McKinsey highlights “agentic engineering” as a frontier skill: building AI agents that can ingest unstructured data, act across tools, follow guardrails, and be reused through playbooks. This connects directly to how companies like Ramp are working.

If you’re interested in the hands-on side of agentic workflows and coding assistants, guides like how to use Gemma 4 and Ollama for local AI coding in VS Code show what this looks like at the developer level.

Ramp’s Glass: Turning Everyone Into an AI Power User

Ramp, a finance automation company, offers one of the clearest examples of institutional AI in action. Internally, they noticed something important: 99% of employees were using AI, but many were stuck. Not because the models were weak—but because the setup was painful and everyone was figuring it out alone.

So they built Glass, an internal AI workspace that every employee gets on day one. It’s essentially an AI operating system for the company.

Principles Behind Glass

Ramp designed Glass around three core principles:

1. Don’t limit anyone’s upside. The default approach for “non-technical” users is often to simplify and lock things down. Ramp did the opposite. Power users at Ramp rely on:

• Multi-window workflows
• Deep integrations across tools
• Scheduled automations
• Persistent memory
• Reusable skills

The goal wasn’t to remove complexity, but to make it invisible while keeping full capability available to everyone. The assumption is that any employee, with the right environment and AI support, can become a power user.

2. One person’s breakthrough becomes everyone’s baseline. Before Glass, if one employee discovered a great AI workflow, it stayed with them. With Glass, workflows can be turned into reusable “skills” that others can install in one click.

Ramp built an internal marketplace called Dojo, where these skills live. For example, a CX engineer might create a Zendesk investigation skill that:

• Pulls ticket history
• Checks account health
• Suggests a resolution path

Once published in Dojo, the entire support team can use it immediately.

3. The product is the enablement. Ramp found that workshops and training sessions weren’t what drove the most value. The biggest gains came when people installed a skill on day one and immediately got a result.

Glass is designed so that every feature teaches by doing:

• Skills show what “great” AI output looks like.
• Memory shows how context changes the quality of answers.
• Self-healing integrations show that errors are handled by the system, not the user.

Instead of formal training, people learn AI by using AI in their real work.

Everything Connects on Day One

Glass comes pre-configured. When an employee signs in, all of Ramp’s core tools are already connected via SSO and one-click setup—things like:

• Internal products (Ramp Research, Ramp Inspect, Ramp CLI)
• Sales tools like Gong and Salesforce
• Collaboration tools like Slack and Notion

So when a sales rep asks Glass to pull context from a Gong call, enrich it with Salesforce data, and draft a follow-up, it just works. This is organizational-level context engineering: the system is designed so that AI can see and use the full relevant context of the company by default.

Skills, Dojo, and the Sensei

With more than 350 internal skills shared, Glass needed a way to help people find what matters to them. Ramp built an AI guide called Sensei that:

• Looks at the tools a user has connected
• Understands their role
• Observes what they’re working on

Then it recommends the most relevant skills. A new account manager doesn’t have to browse a catalog of hundreds of skills—the Sensei surfaces the handful that matter most on day one.

Memory and Continuous Context

Glass also includes a robust memory system. When a user first opens it, Glass builds a memory layer based on the tools they’ve authenticated. Each chat session can access:

• Who they work with
• Active projects
• Relevant Slack channels
• Notion docs
• Linear tickets
• Calendar events

Under the hood, Glass runs a daily synthesis and cleanup pipeline that:

• Scans previous sessions
• Pulls updates from tools like Slack, Notion, and Calendar
• Refreshes the user’s context automatically

This means users don’t have to re-explain everything in every session. The agent adapts to their world over time. Technically, this is hard to do well—but it’s also where a lot of the real value of AI agents comes from.

Automations and Scheduled Work

Glass supports scheduled automations that can run daily, weekly, or on custom schedules (like cron jobs). These automations can:

• Run complex workflows
• Post directly into Slack
• Act across tools without the user being online

This mirrors a broader trend across AI platforms: moving from “chat with a model” to “set up an ongoing code factory or workflow engine.” If you’re exploring this direction, you might also find it useful to look at practical skill-building content like the Claude co‑work skills people use daily, which shows how reusable skills can compound over time.

Why Ramp Built Glass In‑House

Ramp could have tried to buy something like Glass from a vendor. Instead, they chose to build it themselves, for three main reasons:

1. Internal productivity is a moat. Using AI well is now a core business capability. If you can make every employee more effective with AI, you move faster, serve customers better, and build an advantage that’s hard to copy. Ramp sees internal AI infrastructure as part of its moat—something too important to fully outsource.

2. Speed of iteration. When you own the tool, you can see exactly where users get stuck and ship fixes the same day. Ramp has a Slack channel where people report issues; those get turned into tickets automatically and are often resolved within hours. That kind of tight feedback loop is hard to get if you’re waiting on a vendor’s roadmap.

3. It shapes the external product. Ramp builds AI-first products for finance teams. The problems they solve internally—like building useful memory, distributing skills, and surfacing the right functionality at the right time—are the same problems their customers face. Glass gives them “reps” on the hardest AI product problems before those solutions ever reach customers.

The deeper point: AI use is becoming a core primitive of how organizations operate. It’s not something you can fully externalize. Building internal systems is both capability-building and product R&D at the same time.

Raising the Floor for Everyone

One of the most powerful ideas in Ramp’s approach is this: don’t lower the ceiling, raise the floor.

Instead of assuming only a small group of “AI experts” will use advanced tools, they assume everyone can become a power user if:

• The environment is well configured
• Context is wired in by default
• Great workflows are easy to share and reuse
• The product teaches as people work

When a new hire’s first AI session already knows their team, projects, and tools, and when someone who’s never opened a terminal can schedule automations that used to require an engineer, the entire organization levels up at once.

That’s what the best companies are doing with AI today:

• They use AI to drive growth and new business, not just efficiency.
• They build systems that coordinate and amplify individual AI use.
• They treat agentic engineering and internal AI platforms as core capabilities, not nice-to-haves.
• They design for a world where everyone can be an AI superuser, not just a select few.

As more companies share their internal playbooks and as open-source reference platforms for AI agents emerge, it will get easier to follow this path. But the mindset shift—treating AI as a structural, organizational capability—is the real unlock.