Most will agree that AI transformation is the foremost challenge of modern enterprises. It will determine the winners and the losers of the coming decades.
But there isn’t a clear consensus on how to successfully transform with AI.
Broadly speaking, Enterprises are approaching AI transformation in two camps. One follows the traditional transformation playbook: map the processes, build a roadmap, deploy in phases. The other hands tools to employees and lets a thousand flowers bloom.
Neither works on its own. Both implicitly depend on something enterprises have struggled to capture: how work actually happens, down to the exceptions, edge cases, and judgment calls that agents must understand to be useful.
The Traditional Transformation Playbook
The first camp is using a tried-and-tested approach. Lengthy consulting engagements, process mapping, phased rollouts. It worked for ERP, for cloud, and for RPA. Why not use it for AI?
Let’s start with scope. AI can change hundreds of processes across an organization, each with its own variations and dependencies. Understanding how much work that entails takes time. By the time you map one part of the business, the underlying reality has changed. The ground is constantly shifting.
Then there is depth. An agent making a judgment call needs to understand why a rule exists, when an exception applies, and who has the authority to decide. A process map that says, “review contract, flag terms, route for approval” leaves out crucial context: the “Why.”
And the technology landscape is evolving faster by the day. As AI capabilities change, so does the set of work worth automating. A roadmap built around last year’s assumptions need to be revisited. The organization needs a way to keep its understanding current without starting discovery over again and again.
Let a Thousand Flowers Bloom
The second camp starts with a different premise. The people closest to the work know where the problems are. Give them the tools and let them build.
The premise makes sense. A finance analyst may understand a reconciliation better than anyone on a central transformation team. But knowing how to do the work is different from designing an agent to do it.
The analyst remembers an exception when it appears. When asked to describe the process in advance, they may never mention it. They may also have little visibility into the work upstream that creates the exception, or what another team does to resolve it. The agent can look convincing in a demo but struggles as soon as those exceptions appear.
The problem starts before anyone builds an agent. An enthusiastic team can choose a workflow with little economic value, too many unresolved dependencies, or complexity that makes it a poor first deployment. Local improvements can accumulate without moving the enterprise toward a shared objective.
A thousand siloed experiments burn plenty of tokens, but they don’t add up to a cohesive enterprise AI strategy and effective agents in production.
The Knowledge That Never Gets Written Down
The two camps meet in the same gap from opposite directions. A central team struggles to gather enough detail across the organization. Distributed teams struggle to connect their detailed knowledge to the wider business.
Much of that knowledge never makes it into a system of record. A discount is approved; a ticket is escalated; a contract is flagged. The system records the action, but it doesn’t capture the reasoning behind the decision.
This gap shows up in Within’s Transformation Readiness Index (TRI), a survey of operators on how prepared their organizations are for transformation. Only 40% of respondents say their documentation reflects reality very accurately, and 41% say losing their three most knowledgeable operators would make that knowledge difficult or impossible to recover.
Capturing it requires more than yet another questionnaire. It requires watching people do the work. Asking them why they took a particular action. Comparing what happened with the documented procedure, then checking the differences with the people responsible.
The result should be a shared understanding that connects individual decisions to the processes, systems, and people around them. Leaders need that understanding to set priorities. Builders need it to design, create, test, and ship agents.
What Leading Organizations Do Differently
Understanding the work gives organizations a basis for deciding which agents to build and how to make them useful. Three practices turn that understanding into better deployments.
Use AI to map the work before automating it. Start with a business objective and understand the relevant work well enough to identify where an agent can make a difference. That includes the exceptions, dependencies, and effort required to handle them. Among organizations with agents in production, TRI finds that 48% with full operational visibility report strong value 6–12 months after launch, compared with 20% of those with low visibility.
Use that understanding to redesign the process. Consider a senior contract reviewer who knows which clauses need attention for a particular deal and which variations require escalation. Capturing that judgment helps a team define how routine cases should be handled and what information an exception needs before it reaches a human. That knowledge gives the team a basis for changing the workflow, including the handoffs and approval steps that slow it down.
McKinsey’s 2025 State of AI report supports the importance of workflow redesign. Among 25 attributes examined, it had the strongest association with reported EBIT impact from generative AI. Agentification and process redesign must go hand in hand; adding agents to an unchanged workflow risk preserving the very inefficiencies they could eliminate.
Commit to production. Organize teams with the expertise, funding, and mandate to move agents into production. We’ve observed three recipes for success:
- a business owner paired with an AI center of excellence;
- tech-savvy business owners building directly on an enterprise agent platform with governance and guardrails built in;
- and business teams partnering with external consultants and forward-deployed engineers.
- In every case, a named owner must be accountable for and knowledgeable about the core business outcome, and someone must have the technical acumen to ship an agent into production within the company’s constraints. The best companies also understand that agents require maintenance – and design and funding for that from the start.
Why Context Matters
That is the foundation we are building at Within: a living, continuously updated model of how an organization operates, connecting observed work with the reasoning behind decisions. A system of record for human work that people and AI systems can query to identify opportunities, design agents, and understand what has changed. As deployment expands, changes in policies, systems, and responsibilities can leave agents working against outdated assumptions. Capturing those changes gives their owners a basis for deciding what to update and retest.
A central team can set direction, and employees can build close to the work. What connects those efforts is a shared, current understanding of the business. It gives leaders a basis for choosing where to invest, and builders the context to deliver against those choices. Each deployment adds to what the enterprise knows, so the next team can build upon it without having to rediscover it.
