Rethinking Work Series: Part 1
There's no shortage of articles about how AI will transform work. What interests me more is how we actually do it. If I walked into an organization today as a leader tasked with transforming how it operates, I wouldn't start with an AI strategy, an organizational chart, or a transformation program. I'd start with the business. Where are we underperforming? What's preventing us from achieving our goals? And how could we organize work differently to get better results?
Please read through the end. The steps I'm about to describe may sound familiar, and perhaps a little exhausting. I've been through enough organizational transformations to know how much work they involve. But don't give up on me. The most interesting part comes at the end because AI may fundamentally change how much of this work we have to do ourselves.
So here is my operator approach:
1. Choose a Business Problem
I'd start with the CEO, CFO, and business leaders to understand the company's priorities and where performance needs to improve. Perhaps customer acquisition costs are rising, product releases are taking too long, or the company is struggling to grow without adding headcount. I'd choose one or two problems where changing how work gets done could make a meaningful difference.
For example, let's say enterprise contracts take 20 days to complete, slowing down sales. I'd establish the current turnaround time, agree on a target, and identify the executive accountable for improving it. That's our starting point. A business problem, not an AI project.
2. Understand and Redesign the Work
I'd trace a contract from the initial customer request through to signature, examining transactions, systems, handoffs, approvals, and exceptions across sales, legal, finance, and operations. Where does work get stuck? Which activities require expertise? Which are repetitive? Where are decisions delayed?
I'd also examine the behaviors behind the process. Are employees escalating decisions because they lack authority? Are managers reluctant to delegate? Are functional boundaries creating unnecessary delays? Perhaps legal review takes only a few hours, but contracts spend days waiting for approval. Automating legal review wouldn't necessarily solve that problem.
Once we've identified the constraints, I'd ask: If we were designing this process today, would we build it the same way?
I'd identify what we can eliminate, simplify, automate, or handle differently. Perhaps an AI-native contract platform could handle much of the routine work currently performed by legal operations and contract specialists: drafting standard agreements, reviewing terms against company policies, identifying deviations, and routing exceptions.
BUT that raises a bigger question. If the technology can perform much of this work, why preserve the existing process and staffing model?
Next and the most missed step in most process redesigns: I'd examine the decision-making framework alongside the workflow. Which contracts can move through automatically within approved parameters? Which decisions can be delegated to sales or operations? What contract values, terms, or risk thresholds require legal, finance, or executive approval? Who retains final signing authority? I'd establish clear rules and escalation paths so routine agreements can move quickly while higher-risk decisions receive appropriate oversight.
The objective isn't simply to accelerate contract review. It's to rethink the entire contracting process, including the work, the decisions, and the people required to manage it.
3. Redesign Roles and the Organization
Once we've redesigned the work, I'd examine what happens to the roles and people supporting it. In our contracting example, AI could take over much of the routine work performed by paralegals and contract specialists, reducing the need for some positions. Junior lawyers might shift from drafting and reviewing standard agreements toward negotiations and risk assessment, while experienced lawyers focus more on complex commercial decisions. Sales operations roles could also change as contract preparation, approval routing, and administrative tasks become automated.
But new opportunities emerge. Some employees could become builders, configuring AI-enabled solutions; others orchestrators, managing workflows between people and AI agents; and others strategists, applying expertise and judgment to more complex business problems.
I'd assess which roles disappear, which evolve, which new ones are needed, and where employees could be redeployed. I'd also reconsider team structures, management layers, and decision-making authority.
Not everyone will be able to transition, and some positions may be eliminated. But the goal isn't simply to reduce headcount. It's to design an organization around the work that remains, the new opportunities AI creates, and the people best equipped to deliver them.
4. Build the Talent and Capabilities We Need
Now that we understand what the future organization requires, I'd assess the talent we have against the capabilities we need.
In our legal example, commercial judgment, negotiation, risk assessment, the ability to evaluate AI-generated recommendations, and managing automated workflows may become more valuable than routine document processing. Who already has those capabilities? Who has the potential to develop them? Where do we need to hire, retrain, or redeploy?
This is where talent density becomes especially important. As routine work becomes automated, organizations will increasingly depend on people who can exercise sound judgment, adapt quickly, solve unfamiliar problems, and take accountability for outcomes. I'd want leaders who can recognize and develop those capabilities while maintaining high performance standards.
I'd also rethink how we develop junior talent. If AI performs much of the foundational work through which people traditionally gained experience, how will they develop the expertise and judgment required for more complex responsibilities? We can't eliminate the work that builds expertise without creating new ways for people to learn.
Workforce transformation isn't complete when we've redesigned the roles. We also need a credible plan for helping people succeed in them.
5. Test the New Operating Model
I'd start with a focused pilot rather than a company-wide rollout, introducing the redesigned contracting workflow with clear responsibilities, appropriate controls, and targeted training.
Then I'd observe how people actually work. Are managers still requiring approvals we've eliminated? Are employees creating workarounds? Are AI-generated recommendations reliable? Do people have the authority, skills, and confidence to make decisions differently?
These are the issues that can undermine an otherwise sound design. I'd address them quickly, adjust the model, and continue testing. The success of the pilot depends as much on leadership behavior and employee adoption as it does on the technology.
6. Scale What Works
Finally, I'd compare the results with the business outcome we established at the beginning. Have we reduced contract turnaround time? Maintained quality and managed risk? Improved the customer experience? What happened to the capacity we freed up, and are we using it productively?
If the model works, I'd look at where we can apply it elsewhere in the organization. If it doesn't, I'd change it or stop.
The objective isn't to deploy more AI. It's to improve business performance.
This Isn't One Function's Responsibility
Transforming how an organization operates isn't the responsibility of a single executive or function. Business leaders understand the work and the customer. Technology leaders understand what's possible. Finance understands the economics. People leaders bring expertise in organizational design, workforce planning, talent, leadership, and culture.
Each has an important role, but the transformation needs to be led around shared business outcomes rather than functional priorities. The challenge is connecting those perspectives and making decisions about how work, technology, and people come together.
But Haven't We Done This Before?
If you're reading this and thinking it sounds exhausting, I agree. And none of these steps are particularly new. We've spent decades on business process reengineering, organizational redesign, cross-functional transformation teams, and change initiatives. I've participated in these efforts and know how long they can take. Months of interviews, process mapping, skills assessments, workshops, and implementation planning. Sometimes, by the time we've diagnosed the problem, the business has moved on.
But here's what's different today: AI can help us transform the way we approach transformation itself.
Much of the information we previously spent months gathering already exists across business systems. With the right data, integrations, and tools, AI can help reconstruct workflows, identify bottlenecks, generate alternative operating models, and assess how changes affect roles and capabilities.
Instead of spending weeks documenting processes, we can use technology to surface how work actually moves through the organization, leaving people to validate the findings and make decisions. Rather than starting with broad skills assessments, we can identify the capabilities required by redesigned roles and focus our efforts there. And instead of lengthy planning cycles, we can test smaller changes, learn from the results, and adjust.
Of course, this depends on reliable data, appropriate safeguards, and human judgment. AI won't make the difficult leadership decisions for us, particularly those involving people, accountability, and organizational priorities. But it offers the possibility of making organizational transformation faster, more focused, and less resource-intensive. I don't want to spend six months studying an organization before we begin changing it. I want to identify a business problem, understand the work, redesign it, and test a better approach as quickly as possible.
We've known how to redesign organizations for a long time. AI gives us an opportunity to do it differently.
And perhaps that's where we should start.
