Luke Girgis: AI Workflow Automation - What Should You Automate First?

Revenue is up. The team is bigger. Everyone is busier.

Yet margins have barely moved, managers spend their days coordinating work instead of improving it, and every important new customer seems to require another hire.

For a growing company, that can look like success. It may actually be a warning.

The instinct today is often to reach for AI: add a copilot, deploy an agent, automate some emails, buy another platform. But AI workflow automation cannot fix a workflow nobody has stopped to question. In some cases, it simply helps an inefficient organization do unnecessary work faster.

That is the central challenge behind operator and founder Luke Girgis's argument in Death to the Org Chart: instead of designing companies around positions and then giving those people work, leaders should understand the work first—and organize people and technology around the workflows that actually create value. The book's official description calls this making the workflow the structure, with the org chart becoming an output rather than the starting point.

It is also increasingly consistent with the research. McKinsey's 2025 global AI survey found that, among 25 practices examined, workflow redesign had the strongest relationship with organizations reporting EBIT impact from generative AI. Yet only 21% of respondents at organizations using gen AI said at least some workflows had been fundamentally redesigned.

The technology may be new. The management problem is not.

Why Growing Revenue Can Hide a Broken Operating Model

One of the most dangerous numbers in a growing company is revenue without context.

Girgis learned that while running media businesses whose audiences and sales were growing impressively. The problem was what happened underneath the headline numbers.

Margins never exceeded roughly 4%, he recalls, because every major contract created more work and required more labor.

His retrospective description is unusually useful:

“We weren't actually scaling, we were just buying revenue with labor.”

That distinction matters.

True scale means output can grow faster than the resources required to produce it. If every additional $100,000 in revenue generates another layer of coordination, administration, reporting, approvals, project management, and hiring, the company may be getting larger without becoming stronger.

Ask a harder question than “How fast are we growing?”

What happens to our cost and complexity every time we grow?

A founder can test this by examining the last three meaningful increases in revenue. What new work appeared? Which hires became necessary? How many additional handoffs were introduced? Which activities actually served the customer—and which existed only because the organization had become more complicated?

Those answers reveal more about scalability than revenue alone.

Why Adding AI to a Bad Workflow Usually Disappoints

The easiest AI implementation is often the least transformative one.

Give employees access to a chatbot. Add an AI feature to existing software. Automate a few isolated tasks.

Productivity may improve around the edges, but the underlying system remains untouched.

Girgis uses a sharp metaphor for this:

Bolting AI onto the business as it already exists is “like driving a Ferrari in traffic.”

A powerful engine does not eliminate the traffic jam.

MIT Sloan highlighted the same distinction in 2026 when reporting on research into AI and work design: organizations can miss much of AI's potential when they optimize individual tasks instead of reconsidering how tasks are sequenced, grouped, and handed between humans and machines.

For leaders, that changes the starting question.

Don't begin with: What can this AI tool do?

Begin with:

• What outcome is this workflow supposed to produce?

• Which steps actually contribute to that outcome?

• Where is information copied, re-entered, chased, approved, or reconciled?

• Where does work wait for somebody else?

• Which decisions require judgment, relationships, creativity, or accountability?

• Which repetitive steps exist simply because “that's how we've always done it”?

Only after answering those questions should technology enter the conversation.

The Four-Step Approach to AI Workflow Automation

At an e-commerce company Girgis later led, the situation was much more urgent: he describes entering a business losing roughly $400,000 per month and breaking jobs down into increasingly small tasks to understand how orders, production, delivery, and internal work actually flowed.

That experience eventually became a four-part method: Audit, Architect, Activate, Accelerate.

1. Audit the real work

Observe what people actually do—not merely what their job descriptions say they do.

Look for repeated data entry, status chasing, unnecessary approvals, duplicate tools, recurring reports, manual transfers between systems, and work that exists mainly to coordinate other work.

The goal is not yet automation. It is visibility.

2. Architect the workflow before choosing the tool

Take the most expensive or frustrating workflow and redesign it from the desired outcome backward.

Some steps may disappear entirely. Some can be automated. Some can be accelerated with AI while retaining human review. Others should remain human because judgment is the source of value.

This sequencing is crucial: architecture before software.

3. Activate one production workflow

Don't launch an abstract “AI transformation.”

Choose one recurring workflow, give it an owner, deploy the redesigned version, and measure something concrete: cycle time, cost, error rate, customer response time, revenue capacity, or hours returned to the team.

A small production win teaches more than dozens of disconnected experiments.

4. Accelerate what works

Once one workflow performs reliably, move to the next bottleneck.

Over time, the organization changes structurally—not because somebody announced a reorganization, but because different work now requires different combinations of people and technology.

AI Should Return Human Attention, Not Just Cut Costs

Perhaps the most important implication of workflow redesign is what happens to the time that automation creates.

Girgis runs an artist-management business where managers had been spending roughly half their time working with artists and half on administration. He says the company automated around 90% of that administrative burden, allowing managers to redirect their attention toward developing artists, opportunities, strategy, and revenue. He reports that every artist on the roster subsequently earned more than in previous years.

The important metric was not “tasks automated.”

It was valuable human capacity returned.

That distinction is becoming increasingly important as companies decide whether AI is primarily a headcount-reduction mechanism or a way to increase what people can accomplish. Deloitte argued in 2026 that organizations redesigning work around human-AI collaboration were outperforming those focused narrowly on role reduction, reporting stronger ROI and as much as twice the success rate in its analysis.

Girgis summarizes the philosophy more simply:

“We don't automate work because we dislike people. We automate work because we hate wasting their time.”

For a leader, that creates a useful test.

After an automation project succeeds, what will your people do with the hours you give back to them?

If there is no good answer, you may have an automation plan but not yet a business strategy.

Which Business Workflow Should You Automate First?

The best first workflow is rarely the flashiest.

Look for work that is frequent, repetitive, costly enough to matter, measurable, and understood well enough that the organization can tell whether the new system is succeeding.

A weekly process involving five spreadsheets, three people, repeated follow-ups, and predictable decisions may be a much better first candidate than a strategically important but ambiguous executive task.

Avoid beginning with the most complex workflow simply because AI appears capable of touching it.

The purpose of the first implementation is not to prove how sophisticated the technology is. It is to prove that the organization can redesign work and produce a measurable improvement.

Will AI Workflow Automation Replace Employees?

There is no responsible universal answer.

The World Economic Forum's Future of Jobs Report 2025 projects substantial disruption in both directions through 2030: 92 million roles displaced alongside 170 million new roles created, while many existing skills are expected to change.

For individual companies, workflow redesign can produce uncomfortable workforce decisions. Girgis's own turnaround story included significant staff reductions.

But treating workforce reduction as the objective misses the more interesting opportunity.

AI can remove work without eliminating the human contribution around it. It can reduce coordination while increasing customer contact. Remove administration while expanding creative capacity. Compress reporting while giving managers more time to coach, sell, design, negotiate, or solve difficult problems.

The strategic question is therefore not simply “Which jobs can AI replace?”

It is “Which work should humans no longer have to spend their lives doing?”

What Every Founder Should Do This Week

Take one recurring process in your company and follow it from beginning to end.

Not from a dashboard. Not from an SOP.

Watch the work.

Write down every task and handoff. Mark each one as remove, automate, augment, or human. Then choose one metric that would prove the redesigned workflow is materially better.

Do not buy anything yet.

That exercise may reveal that your biggest AI opportunity is not an AI problem at all. It may be an approval that should disappear, information trapped in the wrong system, a process built around an employee who left three years ago, or five people coordinating something that should happen automatically.

The companies that gain the most from AI may not be those that acquire the most AI tools. They may be the ones willing to examine, with uncomfortable precision, how their work actually gets done.

Your org chart tells you who works for you.

Your workflows tell you how your company works.

And in the AI era, the second may matter far more than the first.

This week's challenge: choose one workflow, map it completely, and eliminate one hour of work that no human should need to repeat. Then discuss what that reclaimed hour should be used for with your leadership team.

For more of Luke Girgis's thinking on workflow-first organizations, listen to his conversation on 21st Century Entrepreneurship and explore Death to the Org Chart.