Closing the AI transformation gap: How to prepare your business for agentic AI

Amidst all the hype surrounding AI, there’s a key topic missing from discussion: the AI transformation gap.

By now, organizations of every size are feeling the pressure to deploy AI. And some individuals are becoming genuinely more productive—writing, coding, and analyzing faster.

But have companies as a whole become significantly more productive because their employees are using LLMs? Not yet. A PwC study shows that only 12% of CEOs report meaningful ROI from generative AI investments, and 95% of pilots deliver zero measurable ROI, according to an MIT study.

Despite the growing AI budgets and individual productivity gains, most orgs just aren’t seeing the returns they hoped for, and the reason is actually quite simple: Productive individuals don’t automatically equate to a productive organization.

So, how do you experience the organizational benefits that AI transformation promises?
As it turns out, closing the AI transformation gap isn't about how powerful your AI tools are or how many people are using AI. It’s about reimagining how your workflows actually function.

What is AI transformation?

AI transformation is a comprehensive process in which an organization integrates artificial intelligence (AI) into all aspects of its operations, products, and services. AI transformation can involve implementing multiple types of AI, such as generative AI, used to create text, images, and other content, and agentic AI, which refers to AI applications that perform tasks autonomously. 

Most companies plan to use AI to automate tasks, synthesize information, accelerate decision-making, improve customer experience, and create products or services. 

AI transformation goes beyond simply adopting a few AI tools; it represents a fundamental shift in business strategy, operational processes, and organizational culture. And this is precisely why it’s so difficult. To adapt and shift aspects of the business, you need to understand them properly—but most companies don’t. 

Let me elaborate.

What are the greatest challenges in AI transformation?

The greatest challenges in AI transformation occur in scaling AI beyond individuals or teams. People may be coding, brainstorming, and researching faster than ever, but that value is trapped in silos. It doesn’t compound or connect.

The good news is that we’ve seen this challenge arise from technological advancement in the past, and we can apply those lessons to today’s challenges. 

In the 1890s, factories ran on steam, with a complex web of belts and pulleys. When electricity arrived, factory owners swapped out the steam engine for an electric motor but kept the same floor plan. Despite new technology making individual machines faster, factory productivity barely increased for 30 years.

It wasn’t until someone questioned how the work should actually flow that the factory experienced any real productivity gains. Someone finally understood the work deeply enough to redesign it in a way that maximized the benefits of the electric motor. 

We're in that exact moment with AI: Companies have swapped the motor, but they haven’t yet redesigned the factory. 

To redesign the business for AI, especially agentic AI, you need to understand it clearly: where the handoffs are, where the bottlenecks are, and where decisions are made. Only with an understanding of the business can you examine, challenge, and redesign it to maximize the value of AI. 

But according to a recent Lucid survey, most businesses lack the documentation needed to understand the business properly:

• 49% of knowledge workers say that their organization's current operational workflows are somewhat or hardly well-documented. 
• 60% say that half or more of their team’s workflows rely on informal or person-dependent knowledge.
• 80% rely on tribal or institutional knowledge to complete work.

Bottom line: AI transformation will never reach its full potential until current enterprise processes, workflows, architectures, data flows, and collaboration practices are understood and documented.

The three phases of AI transformation

A recent a16z article describes the gap between individual AI and institutional AI in detail. What I’ve noticed, though, is that most organizations struggle with the transition from one to the next. Why? Because there’s a critical phase missing in between the current state and the end goal. Let’s take a closer look.

Phase one: Individual AI 

The individual AI phase is where most companies are today. During this phase, people are using ChatGPT, Copilot, Claude, and other AI tools individually to accelerate tasks. Many are more productive at their workstations, but nothing else in the organization has changed.

Individual AI is an important phase that involves experimentation and learning, but there’s a ceiling on how much value it can bring to organizations as a whole.

Phase two: Visibility and mapping

As much as organizations may try, it simply isn’t possible to leap from individual AI to institutional AI. There’s a key phase in between that most companies miss: making the invisible, visible.

We know that the end goal is a redesigned org that’s optimized for AI. Before you can redesign anything, though, you have to know the current state. That’s why phase two is all about documentation—mapping processes, systems, dependencies. Once you get institutional knowledge out of people's heads and into a shared visual model, it’s much easier to challenge the existing design and identify where AI would provide the most value. 

But it’s not just people who need documentation on how the business works. If AI agents are to interact with various business systems and processes, they also require significant context to work effectively. Forrester states, “AI agents need step-by-step instructions on how to execute tasks. For most businesses today, this know-how lives in fragmented workflows, undocumented data, and unofficial processes.” (Forrester blog, “Autonomy Is The Future, But AI Agents Still Deliver Value Today,” July 2025).

Phase three: Institutional AI 

Institutional AI is the destination: redesigned, AI-first workflows that drive productivity across the entire organization. At this phase, core processes have been rebuilt from the ground up to fully incorporate AI agents. AI is no longer an individual productivity tool but rather a governed, trusted capability that’s aligned with the broader organization’s goals. 

How to close the AI transformation gap

Within phase two, the bridge between individual and institutional AI, there are a few distinct steps I recommend orgs take: Visualize your current reality, align on a strategy, and deploy with context.

See the reality

The first step in bridging the AI transformation gap is to get visibility into how work is actually done, not just the processes outlined in an outdated company wiki. Think about the judgment calls, tradeoffs, and unwritten rules that exist within each process. Capturing all this context in a shared visual space creates the foundation for AI readiness.

From there, you can evaluate your current workflows, processes, and systems to diagnose how well equipped they are for AI enhancement. After all, adding AI to an inefficient operation would only magnify the problem, not solve it. It’s important to carefully consider which parts of the business are ready for AI and which parts need to be optimized first.

Key activities at this step include:

• Capturing tacit knowledge in visual process maps and system diagrams
• Evaluating the current state to identify priority use cases for agentic AI
• Surfacing system dependencies and integration points within workflows

Align the organization

Next, you’ll want to align teams around an AI strategy that clarifies a shared vision, defines new workflows to achieve that vision, and establishes governance guardrails to safeguard the rollout. Consider: How will processes be redesigned? Where will humans and AI agents work together? What systems will need to be accessible within each workflow? How will the organization ensure everyone has access to the right information?

Developing an enterprise AI strategy is similar to the process teams use to develop products. Leaders need a space to come together, in real time or asynchronously, to brainstorm, prioritize, and plan. And teams need access to documentation that’s centralized, version-controlled, and always approved and up to date.

This process typically involves:

• Facilitating strategic planning to prioritize AI investments based on goals, risks, and projected returns 
• Re-architecting workflows and defining roles between humans and AI agents via RACI charts and swimlane diagrams
• Instituting governance, approval flows, and compliance boundaries for AI initiatives 

Build the future 

With an understanding of your current state and plans for your future state, you can effectively deploy and scale AI. This step involves coordinating execution tasks, deploying agents against redesigned workflows, and refining and scaling processes across teams. 

Notably, it’s also the point at which your documentation becomes infrastructure. While documentation historically functioned primarily as a compliance artifact or presentation visual, it now serves as the critical context layer that guides AI agent decision-making. For example, I’ve seen teams use Lucid to document their operational blueprint and connect their Lucid visuals directly to their AI tools (including ChatGPT, Claude, and Microsoft Copilot) via the Lucid MCP server for seamless execution. 

Deploying AI at scale looks like:

• Creating flexible implementation timelines with clear milestones, dependencies, and task owners
• Feeding approved process documentation directly into an AI execution environment for agents to follow 
• Monitoring results and updating documentation within a governed single source of truth (such as the Process Accelerator in Lucid) 

AI transformation examples

As more organizations look to move from individual AI to institutional AI, a common thread is emerging: The most successful implementations hinge on clearly documented visuals. Here are a couple of examples of orgs leading the charge. 

Carta: Mapping agentic workflows

When Carta set out to deploy autonomous AI agents to handle complex accounting workflows—ultimately saving over 3,500 hours per month—they discovered that the secret to intelligent automation hinged on building a flawless visual blueprint. 

Before an AI agent could step into the cash reconciliation process, execute tool-based tasks, and make micro-decisions, the human operators needed to deeply understand the existing process down to every edge case. 

The Carta team chose to visualize their process in Lucid. With the Lucid diagram as the contextual layer, they translated it into an agent that now performs the vast majority of the manual work, with occasional human-in-the-loop checkpoints. As the agent or the team observed departures from the officially documented process or decisions that fell outside the guardrails, they refined the process in Lucid and updated the agent accordingly.

AI is already transforming how we work at Carta, and Lucid is central to that shift,” said Henry Ward, CEO at Carta. “Lucid has been invaluable for mapping and iterating on AI workflows with our business teams early in the process, giving us the clarity we need before codifying them in other systems. That early-stage collaboration has been critical in helping us turn ambitious ideas into practical, scalable AI solutions.”

Uber: Creating blueprints for AI decision-making

One of the world's largest rideshare companies, Uber Technologies, has already recognized the value of a visual blueprint to accelerate a key AI initiative and save millions of dollars. 

For an organization of this size, responding to millions of support tickets per week is a significant expense. They endeavored to automate this process using agentic AI, but to build this automation, they had to provide the right context so the agent could respond appropriately to support requests.

Uber already had decision-tree-style documentation in Lucid that guided support reps in addressing tickets. Using these visuals as a base, they created comprehensive, multi-layered support documentation in Lucid to serve as a blueprint for the support agent to follow.

The project is on track to automate 70% of response tickets by the end of 2026, with a projected annual savings of $500 million.

Take the next steps in your AI transformation journey

Bridging the gap between individual AI and institutional AI requires visibility, alignment, and governance. Some tools may facilitate basic diagramming but lack governance, while others provide standardization but lack collaboration. Lucid provides the best of both worlds, balancing flexibility with intelligence and structure to help you see your current state, redesign your org, and manage the deployment. 

Lucid supports AI transformation with:

• 100+ integrations with leading apps, including an MCP server to connect to leading LLMs
• Intuitive and intelligent diagramming features to accelerate the visualization of processes, systems, and data
• Advanced add-ons to streamline documentation standardization and governance
• A professional services team for providing tailored solutions