KMS Technology 行政總裁 Jason Wojahn 訪談

Jason Wojahn is CEO of KMS Technology, with roughly three decades of experience building and scaling enterprise technology services businesses. He co-founded and led Thirdera before its acquisition by Cognizant, previously led Accenture’s global ServiceNow business, and held leadership roles at Cloud Sherpas and IBM Global Services. His work focuses on enterprise transformation, operating models, and turning advances in AI into measurable business outcomes.

KMS Technology is a digital engineering, data, and AI company helping organizations build software, modernize applications, strengthen data foundations, and deploy AI in production. Its services span technical consulting, product engineering, data engineering, and intelligent workflow automation. In this interview, Wojahn discusses the “Execution Economy,” the gap between AI pilots and production, decision authority for agents, and how businesses can measure progress through end-to-end outcomes.

You’ve spent three decades building and scaling enterprise technology organizations, from growing Navigis and leading Accenture’s global ServiceNow business to building Thirdera and now serving as CEO of KMS Technology. How have these experiences shaped your perspective on enterprise transformation, and what attracted you to the challenge of helping organizations become AI-native?

I started at IBM Global Services, running delivery for large enterprise clients. From there it was Navigis, Cloud Sherpas, Accenture, Thirdera and Cognizant, almost all of it in the ServiceNow ecosystem.

Here’s the pattern I kept seeing. Every platform wave promised transformation, but the technology rarely decided who won. The winners were the companies that changed how work actually moved through the organization.

The ServiceNow years taught me that value lives in workflows, not dashboards. For thirty years, enterprise software got very good at telling us what was happening. It was much less good at doing anything about it. People filled that gap. They became the unpaid integration team, carrying work from one system to the next.

AI changes that. Intelligence is becoming abundant. What’s scarce now is the ability to turn it into coordinated action. That’s what brought me to KMS: the engineering depth to build systems that do the work, not just track it, and the chance to become AI-native rather than AI-enabled.

Many enterprises have successfully demonstrated AI’s potential through pilot programs, yet struggle to translate those early wins into production-scale value. What are the most common obstacles organizations encounter when moving beyond experimentation, and why do these challenges often emerge only after the technology has proven itself?

A pilot proves the model works. Production proves the organization works.

Those are very different tests. A pilot runs on clean, hand-picked data, with a motivated sponsor and a small team that works around every gap. It’s designed to avoid the hard parts.

Production can’t avoid them. Now the agent has to reach into five systems instead of one. Someone has to own the outcome when it gets something wrong, decide what it’s allowed to approve, and pay to run it every day.

That’s why these challenges appear after the technology has proven itself. They were never technology problems. They’re problems of data, integration, ownership and authority, and the pilot was built to sidestep every one of them.

The fix is to stop treating pilots as demos. Treat the pilot as phase one of production. Pick a workflow that matters, connect it to the real systems, name the owner, and define what the agent can decide on day one. It’s slower to start and much faster to scale.

You’ve described a shift toward an “Execution Economy,” where competitive advantage increasingly depends on an organization’s ability to turn intelligence into coordinated action. What does this concept mean in practice, and how does it change the way enterprises should approach AI investments?

Every piece of work moves through the same chain: Information, Insight, Decision, and Execution.

Enterprise software spent decades on the first two, and AI is rapidly improving the first three. Execution is where most organizations are still stuck, and it’s where the advantage is moving.

In practice, the Execution Economy means intelligence is no longer the scarce resource. Getting things done is. Two companies can have the same models, the same data and the same insights. The winner is the one that turns those insights into coordinated action, with clear authority and an audit trail.

That changes how to invest. Most AI budgets today buy tools: copilots, assistants, better search. They make individuals faster. But a proposal AI drafts in ten minutes can still wait three days for approval. Faster tasks are not faster outcomes.

So fund workflows, not tools, and start with a different question. Not “How do we make this employee 20% more productive?” But “What work should no longer require human coordination?”

As AI evolves from assisting employees with individual tasks to autonomously executing entire workflows, how must organizations rethink their existing processes, decision-making structures, and accountability models?

Most enterprise processes were built around human limits. Information moved slowly, so we built meetings. Decisions needed oversight, so we built approval chains. Much of what we call process is really handoffs between people.

Automate that as it stands and you’ve automated the org chart. You get faster handoffs, not fewer of them.

The rethink starts with decisions, not tasks. For each workflow, map the decisions inside it. What can an agent recommend? What can it decide? What can it execute on its own? Where must a person step in?

Authority belongs to decisions, not entire workflows. An agent might be fully trusted to reconcile an invoice and still need a person to approve any payment above a set limit.

Accountability doesn’t disappear. It gets clearer. Humans and agents jointly own execution: agents carry the work, and people define the authority, handle the exceptions and own the outcome. Every agent needs a named owner, and every action should be traceable and, where possible, reversible.

Leadership shifts from coordinating work to designing the systems that execute it.

At KMS Technology, you’re focused on transforming traditional digital engineering into AI-native enterprise execution. What does this transformation look like internally, and how does KMS approach integrating AI, data platforms, and intelligent workflow orchestration to deliver measurable business outcomes?

We’re applying the same thinking to ourselves first. KMS grew up as a digital engineering company. Our next chapter is AI-native enterprise execution: build, modernize and run, with humans and agents jointly owning execution. The results show up in time and effort:

Build. Evergreen Telehealth wanted a telehealth platform built on agentic AI and the FHIR healthcare-data standard, and nothing like it existed on the market. Working AI-native, with Claude embedded across requirements, design, build and test, our team and Evergreen’s engineers took the core platform from concept to production in 75 working days. That covered three apps (web, iOS and Android), four live integrations (EHR, identity verification, payments and pharmacy) and two production agents. The model was human-led and AI-accelerated. Agents drafted user stories, generated code and ran tests. Engineers owned the architecture, the clinical data mapping and every decision that mattered.

Modernize. Another example, a maritime software company needed to replace a platform it had run on-premise for more than 20 years. It couldn’t justify a long traditional rebuild before investors saw progress, so we ran a six-week, timeboxed agentic build. The client’s requirements were loaded as context, so the agents built against the spec instead of just generating code. The team delivered 35 to 45 percent of the MVP scope in about 800 team hours, against an original estimate of 6,000 to 8,000. Agents wrote the code and more than 5,000 unit tests. Engineers set the guardrails and judged the quality.

Run. Evergreen has since extended its agent architecture across the business, with supervising agents monitoring the system and people in the loop to evaluate and fix.

We’re not just writing code faster. We’re changing how much work a team can execute.

Across all three, humans and agents jointly own execution. Agents handle the repeatable work. Our engineers and our clients’ teams own the decisions, the exceptions and the outcome.

The same shift lets us create greater delivery leverage. That’s the difference between AI-native delivery and AI-enabled delivery.

Enterprise AI systems increasingly depend on accessing information across fragmented legacy applications, cloud platforms, and data environments. How significant is the challenge of connecting these systems, and what architectural foundations should organizations establish before attempting to deploy AI-driven workflows at scale?

It’s the central challenge. Most enterprises don’t have an AI problem. They have a fragmentation problem that AI has made impossible to ignore.

Look at how work gets done today. Someone finds information in one application, chases an approval in another and copies the result into a third. People are the integration layer. An agent can’t do that work unless it can reach those systems, understand the context and act with the right permissions.

So the foundations matter more than the model. Four come first:

• a modern data architecture that gives agents trusted context
• an integration layer that lets agents act across systems, not just read from them
• identity and permissions designed for agents, not only for people
• monitoring, so you can see what every agent did and why

The stack is also compressing. Data platforms are becoming operational, workflow platforms are becoming intelligent, and applications are becoming headless. If the workflow can cross the stack, why should the user have to?

Don’t wait for perfect architecture. Pick one workflow that matters and modernize what it touches. Then do the next one.

When AI agents begin making decisions and executing actions across business operations, questions around governance, accountability, and human oversight become increasingly important. How should enterprises determine which decisions can be delegated to AI and which must remain under human control?

I’d start by changing the question. Not “Can AI do this?” The answer is increasingly yes. The better question is “Who gave it permission to act?”

Capability and access are not authority. When an agent can approve a refund, change a production schedule or commit spend, someone has to set the boundaries.

A practical test is cost and reversibility. A decision that is high-volume, rule-based and easy to undo is a strong candidate for delegation. A decision that is high-impact, irreversible or defines the relationship with a customer should stay with a person.

Delegation should also be earned. Start with agents recommending and a person approving. As the evidence builds, move that person from approving every action to supervising the system and handling the exceptions.

The goal isn’t a human in every loop. It’s a human in the right loops. Data provides context. Models reason. Agents execute. People define authority and own the consequences.

You’ve led major enterprise platform businesses at Accenture and Cognizant, where scaling operations required aligning technology, people, and processes across global organizations. What lessons from those experiences apply to today’s AI transformation efforts, particularly when organizational resistance or outdated operating models become bigger obstacles than the technology itself?

Integration taught me more than anything else. At Cloud Sherpas, then Accenture and later Cognizant, I spent years bringing organizations together. Different cultures, delivery models, incentives and regions all had to operate as one platform business.

The technology was rarely the constraint. The operating model was.

Three lessons carry straight into AI transformation.

First, resistance is usually rational. People protect what they’re measured on. Introduce agents while still measuring teams on hours, headcount and activity, and you’ll get compliance, not change. Change the measures and behavior follows.

Second, decision rights have to be explicit. In every integration, the hardest conversations were about who decides what. AI makes that question unavoidable, because some of the deciders are no longer people.

Third, be clear about what stays the same. People absorb change better when they understand the why and know what isn’t moving.

AI transformation is an operating model shift that happens to involve technology. Leaders who treat it as a software rollout will be disappointed.

Traditional AI success metrics often emphasize adoption rates, productivity improvements, and time saved. What metrics should executives prioritize to determine whether AI is genuinely improving business execution, and how can organizations distinguish between incremental efficiency gains and meaningful operational transformation?

Adoption and time saved are input metrics. They tell you people are using the tool, not whether the business is moving faster.

If an engineer saves two hours but the release still waits a week for review, nothing has changed for the customer. Faster tasks are not faster outcomes.

I’d prioritize three kinds of metrics:

• Time from need to outcome: how long a request takes to become a result, end to end.
• Where the work waits: how long decisions take, the handoffs, the queues between teams.
• Work without coordination: how much of a workflow runs without humans coordinating it, while people stay focused on judgment and exceptions.

Then tie those to the business: cost to serve, quality and revenue.

The test for transformation is simple. Did the task get faster, or did the workflow change shape? Incremental gains make the same process quicker. Transformation removes the steps, handoffs and approvals that no longer need to exist.

Measure the time from need to outcome. Find where the work waits. Change what’s keeping it there.

Looking ahead three to five years, how do you expect the relationship between AI, enterprise software, and human workers to evolve? Do you envision a future where AI-native systems fundamentally replace traditional enterprise workflows, and what should business leaders be doing today to prepare for that transition?

For thirty years, enterprise software told us what was happening. Over the next three to five years, it will increasingly do the work.

What AI will replace is coordination: the chasing, copying, handing off and following up that fills most people’s day. People move toward judgment, exceptions, relationships and owning outcomes. That’s a better use of human expertise, but it takes new capabilities, and leaders have to build them deliberately.

What should leaders do today? Pick one consequential workflow and map the decisions inside it. Fix the data underneath. Define what the agent is allowed to decide, and name the person who owns the outcome. Then move to the next workflow.

Companies that treat AI as a tool will get efficiency. Companies that reorganize around it will get leverage. AI isn’t the strategy. Execution is.

Thank you for the great interview, readers who wish to explore AI-native enterprise execution, intelligent workflows, and the transition from AI pilots to production should visit KMS Technology.