Here's What Over 200 Executives Taught Me About What Comes Next.
By David Nilssen, CEO, DOXA Talent
AI does not fail companies. Leaders fail to lead it.
That may sound harsh, but after spending several weeks working directly with more than 200 mid-market executives across five cities, I believe it is true.
This July, Dave Hajdu, CEO of the AI Officer Institute, and I took a one-day executive workshop called AI in Action to Philadelphia, Denver, Dallas, Seattle, and Walnut Creek. The rooms were filled with owners and executives from professional services, real estate, construction, technology, distribution, food and beverage, and other industries. These were not professional technologists. They were real operators running real companies, responsible for revenue, margins, customers, employees, and increasingly, for deciding what their organizations should do about AI.
We did not ask them to sit through another presentation about the future. We asked them to bring a real business problem. Over the course of the day, they documented an existing workflow, challenged whether it was still the right way to work, identified where AI could help, and began building a new approach. Most of them had never written a line of code in their lives.
The results from the events were encouraging:
But the event scores were not the most valuable result. The more important output was what these leaders taught us about the actual state of AI inside the mid-market.
The problem is not access to technology. The problem is not awareness. The problem is not even willingness. The problem is leadership. And I believe the answer requires a new kind of workforce built around leaders, AI engineers, intelligent systems, and trained global talent.
Seven in ten executives at our AI in Action events said their companies did not have a clear AI strategy, even though the average attendee already used two and a half AI platforms.
Their biggest challenge was not learning that AI exists. It was moving from personal experimentation to operational implementation.
Leaders do not need to become outstanding AI engineers. They do need enough AI fluency to define outcomes, redesign workflows, establish guardrails, ask intelligent questions, and evaluate results.
The goal is to delegate AI without abdicating responsibility for it.
The engineered workforce connects AI-fluent leaders with technical builders and trained global professionals who operate, monitor, and continuously improve AI-enabled workflows.
Finding one: the strategy gap is enormous.
Before each event, we asked attendees a straightforward question: does your company have a clear AI strategy?
Seventy percent said no.
Sit with that for a moment. These were not skeptics who had ignored AI. They were motivated enough to give up a full day, bring a real business problem, and work hands-on with the technology. Yet seven out of ten walked into the room without a clear organizational plan.
That does not mean they had never used AI. Most had. It means their experimentation had not yet become strategy. A few employees were using ChatGPT. Someone had tried Claude. The marketing team was generating content. A manager was summarizing meetings. IT had approved Microsoft Copilot. There might even have been a pilot somewhere inside the company. But isolated usage is not an AI strategy.
A strategy connects technology to business outcomes. It answers questions such as:
What problem are we trying to solve? Which workflow should change? What would a better outcome look like? What information will the system need? Where should a human remain involved? What risks must be managed? Who will own the workflow after it launches? How will we know whether it is working?
Most companies have not answered those questions. Instead, they have collected tools, experiments, subscriptions, and disconnected use cases. The resulting activity can look like progress. But activity without organizational direction rarely produces meaningful returns.
Companies are not struggling because AI is incapable. They are struggling because no one has done the leadership work of connecting AI to outcomes that matter.
Finding two: everyone has the tools, but almost nobody has the results
The executives in our rooms used an average of two and a half AI platforms each, and more than eight in ten used at least two. Only one person among the pre-event respondents said they were using no AI at all.
Adoption is not the problem. The tool era is effectively over.
Executives do not need another list of the best AI applications. They do not need another breathless presentation about what the newest model can do. Most have already purchased access to more capability than their companies know how to use. What they lack is the ability to translate that capability into changed workflows and measurable results.
That distinction matters. Buying software is easy. Redesigning work is difficult. A subscription can be approved in a few minutes. But determining how information should move through a company, which decisions should be automated, which exceptions require judgment, and who remains accountable for the outcome is serious management work. The technology can be acquired almost instantly. The organizational capability cannot.
One additional issue surfaced repeatedly in our conversations. Microsoft Copilot appeared in roughly a third of the responses, but many of those users told us they were using it because their company’s security or data policies required it, not because it was the tool they would naturally choose. That reveals another unresolved leadership issue: many organizations have created a gap between the tools IT permits and the tools employees actually prefer. When that gap is not addressed honestly, people either avoid the approved tools or use unapproved ones in the shadows.
Good governance cannot simply mean saying no. It must create a safe and practical path for people to do better work.
Finding three: the pain is implementation, not awareness
We also asked attendees to describe their biggest AI challenge in their own words. The largest category, representing nearly a third of the responses, was some version of “I do not know how to actually start.” Leaders wanted to know how to connect AI to a real workflow. They wanted to move beyond individual experimentation. They wanted to understand how to build something their team could use repeatedly and safely.
Keeping up with the pace of change, the problem that most AI newsletters and conferences are designed to solve, came up only about a third as often as implementation.
Every week brings another model, product launch, benchmark, prediction, course, and collection of prompts. Information is abundant. A practical method is not. The executive does not need to understand every new feature released by every model provider. The executive needs a repeatable way to select a valuable business problem, examine the existing workflow, redesign it, build the right solution, and assign clear ownership.
AI literacy without implementation discipline creates sophisticated spectators. The market will reward operators.
Finding four: leaders want to build, not listen
Before the events, we asked attendees what would make the day valuable. Their answers clustered around playbooks, automation, agents, workflows, and hands-on building. They did not ask for more inspiration. They did not ask for another trend presentation. They wanted something they could take back to their company and use again.
That is why we structured the day around actual work. Each participant brought a business problem. They mapped what happens today, identified pain points, challenged assumptions, and began redesigning the workflow.
That hands-on approach created an important realization. Most leaders do not need more convincing that AI will matter. They need help developing the capability to lead it.
What this means: leadership just doubled
For most of our careers, leadership was primarily about people. Hire well. Set direction. Build a healthy culture. Allocate resources. Create accountability. Coach the team. Make difficult decisions. None of that goes away.
But a second half of the leadership job has arrived. Leaders must now direct work performed by both people and intelligent systems.
When I say the non-technical leader is dead, I do not mean every executive should learn to code or become an AI engineer. That would be unrealistic and, for most leaders, a poor use of time. I mean leaders can no longer treat technology, data, automation, and AI-enabled workflow design as someone else’s responsibility.
You can delegate technical execution. You cannot abdicate the business outcome. That is the distinction many companies are missing.
A leader cannot simply tell the team to “use AI,” assign the initiative to IT, and assume the responsibility has been handled. That is not delegation. It is abdication disguised as empowerment. The leader must still define why the work matters, what a good result looks like, which tradeoffs are acceptable, where judgment belongs, and how success will be measured.
The engineer may own the build. The operational team may own daily execution. Technology may own security, architecture, and scalability. But the business leader must own the outcome. That requires a new level of managerial fluency.
Leaders need enough fluency to ask the next question
Executives regularly lead disciplines they have not personally mastered. A CEO does not need to be an accountant to lead a CFO. A president does not need to be an attorney to work effectively with legal counsel. A chief operating officer does not need to be a software developer to sponsor a technology implementation. But those leaders must understand the fundamentals well enough to ask intelligent questions, assess tradeoffs, recognize risk, and evaluate whether the work supports the business.
AI is no different. The goal is not mastery. It is managerial fluency. Leaders need enough understanding to ask:
Why are we automating this workflow? Is the current workflow even worth preserving? What information is the AI using? What happens when the system is uncertain? Which decisions can it make, and which must remain with a person? How will we evaluate quality? How will we detect drift? Who will manage exceptions? What happens to the process when the business changes?
That is what it means to delegate without abdicating. You do not need enough AI fluency to build every system. You need enough to lead the people who do.
The three new leadership skills
After working with leaders across the five cities, I believe directing AI comes down to three foundational and learnable skills.
1. Workflow design
Every business is a collection of workflows. A customer requests something. Information enters the organization. Someone reviews it. A decision is made. Work is completed. An output is delivered. An exception occurs. Another person gets involved.
Most leaders know the major functions of their business, but many cannot describe precisely how work moves from input to outcome. That was already a management weakness. AI makes it a direct constraint on growth. You cannot intelligently automate a workflow you do not understand. And you should never assume the existing workflow is the right one simply because it is familiar.
The first step is not asking where AI fits into the current process. The first step is asking whether the current process should exist in its current form at all. Automating chaos only creates faster chaos.
Workflow design requires leaders to examine the intended business outcome, the sequence of steps required to produce it, the decisions made along the way, the information required for each decision, the delays and handoffs and failure points, the work that requires human judgment, the work that is repetitive or rules-based, and the exceptions that break the standard process. Once the workflow is visible, leaders can decide which responsibilities belong to a person, which belong to a machine, and which require both working together.
2. Information architecture
AI is only as useful as the information it can access and understand. If the relevant knowledge is scattered across inboxes, buried in meeting notes, stored in inconsistent folders, or trapped in an experienced employee’s head, the AI cannot reliably use it.
Tribal knowledge used to be inconvenient. In an AI-enabled organization, it becomes a direct constraint on scale.
This is why information architecture is becoming a leadership issue. Companies must become deliberate about how operating knowledge is captured, organized, governed, updated, and made available. That includes policies, procedures, customer history, product information, decision criteria, approved language, examples of high-quality work, escalation rules, compliance requirements, and lessons learned from previous exceptions.
AI cannot compensate for organizational confusion. If the information is contradictory, outdated, incomplete, or inaccessible, the resulting system will reproduce those weaknesses at speed. The companies that organize their knowledge well will have an enormous advantage over those that simply buy better software.
3. Writing instructions
People often refer to this skill as prompting. That description is too small. Writing instructions means translating human judgment into language a machine can interpret and execute. A strong instruction explains the objective, context, standards, constraints, available information, expected format, and circumstances that require escalation.
That is not a trick for getting a better chatbot answer. It is a management discipline. For decades, leaders have relied on conversations, proximity, experience, intuition, and shared context to communicate what good work looks like. Machines do not benefit from that context unless we provide it. They force us to make our thinking explicit. That is uncomfortable, but valuable.
Poor instructions expose fuzzy management. Clear instructions force leaders to define what they actually want. Prompts may not be software code in the traditional sense, but they increasingly serve a similar organizational function: they convert intention into repeatable action. Clarity is becoming a competitive advantage.
AI is raising the standard for delegation
AI is not eliminating the need for delegation. It is raising the standard for it.
Historically, a leader could delegate a function to a capable employee and allow that person to determine much of how the work should be completed. AI-enabled work is different because the leader is no longer delegating only to a person. The leader is helping create a system composed of people, models, data, rules, integrations, exceptions, and feedback loops.
That system needs to be designed. The decision rights must be clear. The information must be governed. The human role must be intentional. The outcome must remain owned.
This is why saying “let the technical team handle AI” is insufficient. The technical team can determine how to build a system. It cannot independently determine which business outcomes matter most, which customer tradeoffs are acceptable, or where the organization is willing to accept risk. Those are leadership decisions.
The same is true of outsourcing the work. Global talent cannot compensate for an executive team that has failed to define the outcome. Global talent is not a substitute for leadership. It becomes far more valuable when leadership provides clarity.
Fluency is necessary, but it is not enough
A leader can understand workflow design, information architecture, and instruction writing and still fail to produce a meaningful result. Skills alone do not create ROI. Structure does.
Companies need a repeatable delivery model that brings the right people together, redesigns the work before automating it, creates the solution responsibly, and establishes ownership after launch. The model we have landed on is the AI Design Sprint.
The AI Design Sprint
The sprint brings together four essential capabilities: an executive sponsor who connects the project to a business outcome, a subject-matter expert who understands how the work happens today, an AI engineer who can design and build the technical solution, and a technology or security partner who advises on data, architecture, governance, and scalability.
The sprint then moves through four phases.
Phase one: align
The team begins by defining the business outcome. Not the tool. Not the model. Not the automation. The outcome. What needs to improve? What is the current cost of the problem? What would success look like? Who is affected? What constraints must be respected?
The executive sponsor is essential because someone must connect the technical project to the economic and strategic priorities of the company. Without that connection, AI work quickly becomes a collection of interesting demonstrations.
Phase two: analyze and challenge
The subject-matter expert helps document the current workflow. The team identifies each step, handoff, decision, delay, data source, exception, and quality-control mechanism. Then the group challenges the process. Why is this step necessary? Why does this information move through three systems? Why must a manager approve this? Why is the customer asked for the same information twice? Why are employees manually copying data between applications? Why has the organization accepted this delay?
The goal is not to preserve the existing process. The goal is to achieve the outcome in the best possible way. Automating chaos only produces faster chaos.
Phase three: refactor and build
Once the workflow is understood, the team redesigns it. Some steps may remain human. Some may be automated. Some may be eliminated entirely. Some may require AI to prepare a recommendation while a person makes the decision.
The AI engineer then builds the solution, working with technology and security partners to ensure the design respects data access, privacy, system architecture, reliability, and scalability. This is where technical expertise matters. But the engineer is not working alone. The executive provides the outcome. The subject-matter expert provides operational context. Technology provides the guardrails. The engineer translates the redesigned process into a functioning system.
Phase four: run and improve
This final phase is where I believe most companies will get the model wrong.
Launching an AI workflow is not the same as operationalizing one. A prototype can work beautifully in a controlled demonstration and still fail in the messiness of daily business. Customers provide incomplete information. Policies change. New exceptions appear. Source data becomes outdated. Model outputs drift. The organization changes its priorities.
Someone must own the workflow after it launches. In our model, that responsibility is handed to a trained global professional who becomes the operational owner of the playbook. They run the process with a human in the loop. They monitor output, handle exceptions, protect quality, identify when the system is drifting, capture new information, surface improvement opportunities, and communicate useful feedback to the AI engineer. That feedback drives incremental improvement over time.
The missing layer in most AI strategies
Most discussions about AI focus on two groups: the executives who decide what the company should do, and the engineers who build the technology. But there is a critical layer between strategy and engineering. Someone must turn the redesigned workflow into daily, repeatable execution.
Someone must maintain the operating playbook, review routine output, handle edge cases, update source information, monitor service levels, document failures, identify recurring exceptions, escalate issues requiring judgment, coordinate improvements with the engineer, and measure whether the workflow is producing the intended result.
Executives should not spend their days doing that work. AI engineers should not spend their days managing routine exceptions. Without an operating layer, promising AI initiatives stall between executive intent and technical possibility.
This is where trained global talent changes the equation. Global talent is not a substitute for leadership or engineering. It is the implementation capacity that allows both to scale.
Global talent must become more than an extra pair of hands
The traditional outsourcing model was largely built around labor arbitrage. A company identified tasks being performed in a higher-cost market and moved those same tasks to a lower-cost market. The work might have become less expensive, but it often remained fundamentally unchanged.
That model will not be enough for the next decade.
The future of global talent is not simply replicating roles in another country. It is designing roles around a new operating environment. AI will execute portions of the workflow. People will manage judgment, context, relationships, quality, and exceptions. Engineers will improve the underlying systems. Leaders will remain accountable for the business outcome.
In that environment, global professionals cannot simply follow a static script. They must understand the workflow well enough to recognize when it is no longer working. They need enough AI fluency to identify when the system is producing lower-quality output, when the instructions no longer reflect the business, when a new exception has appeared repeatedly, when the available information is incomplete, when a decision should be escalated, when a manual step could be improved, and when the engineer needs better operational feedback.
They do not need to build the system. They need to understand it deeply enough to help make it better every week. That is a much more valuable role than completing repetitive tasks. It is also a more resilient one.
The human is not merely a backup plan
Human-in-the-loop is sometimes described as a temporary safety mechanism: the AI does the work, and a person checks it because the technology is not quite good enough yet. I think that framing is incomplete.
The human is not present only because the machine may make a mistake. The human is present because many business processes contain ambiguity, empathy, commercial judgment, prioritization, accountability, and context. Those capabilities are not incidental to the workflow. They are often where the value lives.
A trained person becomes the judgment layer around the machine. The AI can classify information, prepare a recommendation, generate a draft, identify patterns, or execute a defined action. The person interprets the situation, manages the exception, protects the relationship, and remains accountable for the result.
The goal is not to assign all work to AI. The goal is to deliberately assign each part of the workflow to the resource best suited to perform it. Sometimes that is a person. Sometimes it is a machine. Often it is both.
Three kinds of ownership
For the engineered workforce to work, companies must distinguish between three forms of ownership.
The business leader owns the outcome. The leader defines why the workflow matters, what success means, which constraints apply, and what risks are acceptable. This responsibility cannot be delegated away.
The AI engineer owns the technical system. The engineer designs, builds, integrates, tests, and modifies the automation. They are accountable for the technical performance of what has been built.
The global professional owns daily operation and feedback. The operational owner runs the workflow, manages exceptions, protects quality, measures performance, and identifies where the system needs to improve.
When these responsibilities are clear, AI becomes a managed organizational capability. When they are unclear, the workflow becomes another pilot nobody trusts and eventually abandons.
AI strategy is really workforce strategy
Many leaders still think about AI strategy as a technology plan. Which tools should we approve? Which model should we use? Which vendor should we select? Those decisions matter, but they are not the strategy.
AI strategy is ultimately about the design of work. It asks: What outcomes must the company produce? Which workflows create those outcomes? What responsibilities belong to people, and what responsibilities belong to machines? What knowledge must be made accessible? Where must human judgment remain? Which new capabilities must employees develop? Who owns each workflow? How will the system learn and improve?
These are workforce questions. They are operating-model questions. They are leadership questions. The companies that win will not simply deploy more AI than their competitors. They will design better systems of work.
The engineered workforce
Put all of this together and you get what I believe the next decade of work will look like.
Leaders will remain responsible for vision, judgment, priorities, culture, customers, and business outcomes. They will also develop enough fluency in workflow design, information architecture, and instruction writing to direct AI responsibly. AI engineers will turn redesigned workflows into secure, functional systems. AI will perform portions of the repetitive, analytical, or rules-based execution. Trained global professionals will operate the playbooks, manage the human-in-the-loop responsibilities, monitor performance, handle exceptions, and drive continuous improvement. Technology and security partners will ensure that data, architecture, governance, and scalability are built into the system.
That is the engineered workforce. It is not humans versus AI. It is not replacing an entire department with a chatbot. It is not adding AI tools to an organizational chart designed for 2015. It is a workforce deliberately designed around the outcome it serves. Every human role has a purpose. Every machine responsibility is defined. Every handoff is intentional. Every workflow has an owner. Every system has a feedback loop.
Leadership leverage is being redefined
For most of business history, increasing output meant adding people. More customers required more service representatives. More transactions required more administrators. More growth required more layers of management.
AI changes that relationship. But the goal should not be indiscriminate headcount reduction. The goal should be increasing the amount of valuable output each leader can responsibly direct.
AI provides execution capacity. Engineers provide automation capacity. Global talent provides operational capacity. Leaders provide direction and judgment. When those capabilities are combined well, company output can grow without every increase requiring an equivalent increase in domestic headcount.
That is not simply a technology opportunity. It is an opportunity to redesign how organizations scale.
The opportunity in front of leaders
Seventy percent of the executives we met this summer did not yet have a clear AI strategy. I do not say that as criticism. I say it as evidence of the opportunity.
The market is still early. Most companies are experimenting. Most leaders are still forming their point of view. Most workflows have not yet been redesigned. Most organizational knowledge remains poorly structured. Most AI initiatives do not have clear operational ownership. The leaders who begin building these capabilities now can create a meaningful advantage.
But they must stop waiting for the technology to do the leadership work for them. AI will not define the business outcome. It will not resolve competing priorities. It will not determine the company’s risk tolerance. It will not decide where human judgment matters most. It will not create accountability.
Leadership still has to lead.
Start with one workflow
The idea of engineering an entire workforce can feel overwhelming. So do not begin with the entire company. Begin with one workflow.
Choose something connected to a meaningful business outcome. Map how it works today. Identify the inputs, decisions, handoffs, delays, and exceptions. Challenge whether the existing process is still the right one. Determine which parts should be performed by a person, which could be performed by AI, and which should combine both. Make the information accessible. Define the instructions. Establish the guardrails. Assign ownership. Measure the result. Improve it.
Then repeat. That repeated discipline becomes the AI strategy.
Most people are still waiting for AI to prove itself. The leaders we met in five cities this July are done waiting. They are learning how to lead it. And the organizations that learn to delegate AI without abdicating responsibility for the outcome will be the ones that turn this technology from an experiment into an enduring capability.
David Nilssen is the CEO of DOXA Talent which helps businesses build high-performing teams leveraging offshore talent. With over 1,000 employees and no office, he is an advocate for remote work. He serves on the Board of Directors of the AI Officer Institute, DOXA Talent, and Guidant Financial.
