There is something about AI productivity that has been bothering me.
We are getting very good at making individual tasks faster.
Writing is faster.
Research is faster.
Coding is faster.
Analysis is faster.
Presentations are faster.
Customer responses are faster.
And yet I keep coming back to a simple question:
What if making every task faster does not necessarily make the business better?
In fact, in some situations, AI may allow us to become extremely efficient at producing the wrong things.
That sounds counterintuitive.
But anyone who has worked seriously with process improvement will recognize the pattern.
A process can have highly efficient individual steps and still produce a terrible overall outcome.
We have known this problem in operations for decades.
AI is simply giving us a much more powerful engine for creating it.
“The fastest way to improve a bad process is not always to accelerate it. Sometimes it is to stop doing it.”
The Difference Between Task Productivity and System Productivity
Imagine an employee who previously needed two hours to prepare a report.
AI reduces that to twenty minutes.
We celebrate.
And rightly so.
That employee has gained 100 minutes.
But then something interesting happens.
Because the report is now cheap to produce, the organization asks for more reports.
One report becomes five.
Five become ten.
Managers now have more information.
Analysts have more outputs.
Meetings have more dashboards.
Executives receive more summaries.
The organization becomes extremely productive at producing information.
But has decision-making improved?
Maybe.
Maybe not.
This is the distinction I think we need to make much more carefully:
Task productivity
How efficiently did we complete the activity?
versus
System productivity
Did the entire system produce a better business outcome?
Those are not the same thing.
And AI makes the distinction more important because AI dramatically lowers the cost of producing certain kinds of work.
AI Is Removing the Cost of Doing Things We Should Have Questioned
This is where my Lean Six Sigma thinking naturally comes in.
One of the most useful questions in process improvement is not:
“How can we do this activity faster?”
It is:
“Why are we doing this activity?”
That question becomes surprisingly powerful when AI enters the picture.
Consider a simple example.
A manager asks an analyst to prepare a weekly report.
The analyst spends six hours collecting data, cleaning it, formatting it, and writing commentary.
AI reduces that effort to thirty minutes.
Wonderful.
But before celebrating, ask:
Why does the report exist?
Perhaps it exists because leadership needs to make a decision.
If that decision could instead be supported by a live dashboard and automated alerts, then the report itself may be unnecessary.
We have two very different outcomes.
Scenario A
AI makes the report cheaper.
Scenario B
AI makes the report unnecessary.
Scenario B is much more interesting.
But organizations often stop at Scenario A because the productivity gain is easier to measure.
The Productivity Trap
I think we are entering what I would call the:
AI Productivity Trap
It happens when:
AI increases the efficiency of individual activities → the organization increases the volume of those activities → the additional work creates new downstream demand → overall system complexity increases → the expected productivity benefit gets diluted.
The trap isn’t that AI failed.
The AI worked exactly as intended.
The problem is that we optimized the wrong level of the system.
This is a classic local-optimization problem.
And it is one of the reasons I believe AI adoption requires much deeper process thinking than we often acknowledge.
A Simple Example From Customer Service
Consider a customer-service organization.
Before AI:
A customer sends an email.
An employee reads it.
They research the issue.
They write a response.
Average handling time: 12 minutes.
Now AI arrives.
The system reads the email, retrieves information, drafts the response, and reduces human effort to three minutes.
That sounds like a fourfold productivity improvement.
But now suppose the organization decides:
“Since responses are much faster, let’s increase the number of customer interactions we proactively initiate.”
The team sends more communications.
Customers reply.
More tickets arrive.
The contact center handles more volume.
Managers see higher productivity per interaction but higher overall workload.
Eventually another AI system is introduced to manage the additional volume.
Then another system handles escalations.
Then another system monitors quality.
Then humans spend more time reviewing AI-generated outputs.
The organization has not necessarily become simpler.
It may have become faster and more complicated at the same time.
That is a dangerous combination.
Faster Is Not the Same as Better
This is perhaps the most important idea in this article.
We often use:
Speed = Productivity
But business productivity is multidimensional.
A process has to be evaluated through things like:
- Cost
- Quality
- Speed
- Reliability
- Customer experience
- Risk
- Employee effort
- Business value
A ten-second process that produces a wrong answer is not better than a ten-minute process that consistently produces the right one.
A report generated in thirty seconds that nobody uses is not more valuable than a report produced in two hours that enables a million-dollar decision.
A customer-service response generated instantly that frustrates the customer is not a productivity improvement.
It is merely faster output.
This is why I think AI evaluation needs to move beyond model-level metrics.
The question should eventually become:
Did the AI improve the system?
Not simply:
Did the AI perform the task?
The Hidden Cost of AI-Generated Work
There is another side to this problem.
AI can reduce the cost of producing content, analysis, decisions, code, documents, and recommendations.
When production becomes cheap, demand tends to expand.
And that creates a strange organizational phenomenon:
The amount of work grows because the cost of creating work has fallen.
Think about email.
Or presentations.
Or dashboards.
Or reports.
Or software tickets.
Or marketing content.
Or internal documentation.
AI can make all of these dramatically easier to produce.
But easier production does not automatically create more value.
Sometimes it creates more organizational noise.
The organization starts drowning in outputs.
More documents.
More dashboards.
More recommendations.
More code.
More messages.
More “insights.”
Someone still has to:
- Read them
- Validate them
- Prioritize them
- Act on them
- Store them
- Govern them
- Maintain them
- Decide which ones matter
The cost has not disappeared.
Sometimes it has simply moved.
The Bottleneck Moves
This is one of the most important concepts I learned from process improvement:
Improving one step does not necessarily improve the system if the bottleneck simply moves somewhere else.
Suppose AI makes analysis extremely fast.
The bottleneck may move to decision-making.
Then AI makes decision preparation faster.
The bottleneck may move to approval.
Then approvals become automated.
The bottleneck moves to implementation.
Then implementation becomes faster.
Now the bottleneck moves to customer demand.
The system keeps changing.
This is why optimizing isolated AI use cases can be misleading.
You are not optimizing a collection of independent tasks.
You are optimizing a system of interconnected activities.
This Is Why I Stopped Thinking About AI Agents in Isolation
This is also where my thinking about AI agents has changed.
I used to find the question:
“What can this agent do?”
interesting.
I now find a different question much more useful:
“What should the entire system accomplish, and what is the simplest architecture that can achieve it?”
An agent may be part of that architecture.
But it might not be necessary.
Sometimes the best solution is:
- A deterministic workflow
- A database query
- A simple automation
- A retrieval system
- A human decision
- A small language-model call
Or some combination of them.
I wrote about this more deeply in my DEV.to article Stop Building AI Agents. Start Building AI Systems.
The idea is simple:
Don’t optimize the sophistication of the AI component. Optimize the system that produces the outcome.
That distinction matters enormously for productivity.
The Question We Should Ask Before Automating
Whenever I see an organization identifying an AI automation opportunity, I would now ask a sequence of questions.
Not:
“Can we automate this?”
But:
What outcome does this activity support?
Then:
Does the activity actually need to exist?
Then:
What happens before it?
What happens after it?
Where does its output go?
Who uses the output?
What decision depends on it?
What happens if we eliminate it?
Only after those questions should we ask:
“Where should AI enter the process?”
This is slower at the beginning.
But it can save enormous amounts of time later.
Local Optimization vs System Optimization
This distinction deserves a simple visual way of thinking about it.
Local optimization
Task → AI → Faster task
The focus is the individual activity.
System optimization
Business outcome → Process → Workflow → People + AI + Systems → Outcome
The focus is the complete chain.
The first approach asks:
How much time did we save?
The second asks:
What changed because we saved that time?
That second question is the one I care about.
Because saving ten hours is not the objective.
Creating more value with those ten hours is the objective.
What Should Employees Do With the Time AI Saves?
This may be the most important question in an AI transformation.
Suppose AI genuinely saves an employee ten hours a week.
What happens next?
There are at least three possibilities.
Option 1: More of the same work
The employee produces more reports, more analysis, more content, or more transactions.
This creates throughput.
But not necessarily value.
Option 2: The organization reduces headcount
The organization captures some of the productivity financially.
Again, this may be rational in some situations.
But it doesn’t necessarily create new organizational capability.
Option 3: The employee moves upstream
The employee spends more time on:
- Root-cause analysis
- Customer relationships
- Innovation
- Process improvement
- Strategic thinking
- Complex decisions
- Experimentation
- Exception management
This is the most interesting possibility to me.
Because now AI is not simply making people faster.
It is changing the kind of work the organization is capable of doing.
That is transformation.
AI Can Also Create Negative Productivity
There is an even more uncomfortable possibility.
Sometimes AI doesn’t merely fail to improve productivity.
It can reduce it.
For example:
AI generates inaccurate information.
A human reviews it.
The human discovers errors.
They investigate the source.
They correct the output.
Another employee verifies the correction.
The process now contains:
AI generation → Human verification → Error investigation → Rework → Final output
The original manual process may actually have been simpler.
This is why I am skeptical when I see AI projects measured primarily by:
- Number of AI outputs
- Number of prompts
- Number of users
- Number of automated tasks
Those are activity metrics.
They tell us that AI is being used.
They don’t tell us whether the system is better.
The Real AI Productivity Equation
I don’t think we should measure AI productivity as:
Output ÷ Human Time
That is useful, but incomplete.
I would rather think about it as:
AI Productivity = Valuable Business Outcome ÷ Total System Effort
And “total system effort” includes more than the employee sitting in front of the AI.
It can include:
- AI infrastructure
- Human review
- Error correction
- Governance
- Integration
- Maintenance
- Monitoring
- Change management
- Downstream work
- Customer impact
This is why an AI tool that makes one employee 50% faster can still be a poor business investment.
The local productivity number may look fantastic.
The system economics may not.
The AI Waste Question
Lean thinking gives us another useful lens.
Ask:
What new waste is AI creating?
That is a question I don’t hear often enough.
AI can potentially reduce traditional forms of waste:
- Waiting
- Overprocessing
- Manual effort
- Rework
- Information searching
- Handoffs
But it can also create new forms of waste:
AI Overproduction
Generating information nobody needs.
AI Rework
Correcting unreliable AI outputs.
AI Review Waste
Humans spending excessive time checking low-risk outputs.
AI Tool Waste
Using multiple AI systems to accomplish something that could be done with one.
AI Coordination Waste
Managing increasingly complicated chains of agents and tools.
AI Decision Waste
Generating recommendations faster than the organization can make decisions.
That last one is particularly interesting.
What happens when the organization can generate decisions faster than it can absorb them?
We may discover that the bottleneck was never intelligence.
It was organizational capacity.
This Is Where Process Excellence Becomes More Important, Not Less
Some people assumed AI would make process improvement less important.
I think the opposite is happening.
The more capable AI becomes, the more dangerous poorly designed processes can become.
Because AI gives us the ability to scale them.
A bad manual process may waste ten hours.
A bad AI-enabled process can waste ten thousand hours.
Automation increases the consequences of process design.
This is why I wrote Why AI Projects Fail Without Process Excellence.
And it is also why I believe organizations should examine the process before implementing AI, rather than automatically searching for an AI use case.
Don't Automate Waste. Don't Accelerate Waste.
There is a simple principle here that I think deserves to become part of every AI transformation conversation:
Before asking AI to make a process faster, ask whether the process deserves to be faster.
That sounds obvious.
Yet it is surprisingly easy to forget.
Organizations have incentives to demonstrate quick AI wins.
A task that goes from ten minutes to two minutes produces an attractive metric.
A task that disappears completely can be harder to celebrate.
There is no “90% efficiency improvement” when the process no longer exists.
But sometimes elimination is the bigger win.
A Better AI Transformation Sequence
This changes how I think about the sequence of AI transformation.
The conventional sequence is often:
Find task → Find AI tool → Automate → Measure time saved
I would prefer:
Define outcome → Understand process → Remove unnecessary work → Simplify → Redesign → Identify AI opportunities → Build system → Measure business outcome → Improve
Notice where AI appears.
It is not first.
It is somewhere in the middle.
That is intentional.
AI should be a design capability, not the design objective.
What Does Good AI Productivity Actually Look Like?
I think we will know we are getting AI productivity right when we see something different happening inside organizations.
Not simply:
Employees completing more tasks.
But:
Organizations needing fewer unnecessary tasks.
Not:
More reports produced.
But:
Better decisions made.
Not:
More customer messages sent.
But:
Customer problems resolved more effectively.
Not:
More code generated.
But:
Better software systems delivered.
Not:
More AI agents deployed.
But:
Better business processes operated.
That is the shift I want us to make.
From Productivity to Capacity
Perhaps the most useful way to think about AI is not:
“How much work can AI do for us?”
but:
“What becomes possible when routine work requires dramatically less human effort?”
That second question opens a much bigger space.
An organization might use the freed capacity to:
- Improve processes
- Explore new markets
- Serve customers better
- Experiment more frequently
- Make better decisions
- Develop employees
- Build new products
- Solve problems that were previously too expensive to address
That is where AI becomes genuinely transformative.
The goal isn’t maximum automation.
The goal is maximum useful capacity.
Where This Connects to Agentic Process Excellence™
This is also why I see Agentic Process Excellence™ as something broader than deploying autonomous agents.
The core question is not:
“Where can we put an agent?”
It is:
“How should intelligent systems, people, processes, and governance work together to continuously improve business outcomes?”
That requires us to think simultaneously about:
- Process
- AI workflows
- Agents
- Systems
- Human judgment
- Governance
- Measurement
- Continuous improvement
My AI Operating Model explores the organizational structure required to scale AI.
My AI Evaluation Framework looks at how we evaluate AI beyond simplistic measures of performance.
And my AI Process Optimization work focuses on improving workflows once we understand where waste and opportunity exist.
But the underlying philosophy remains the same:
AI should improve the system, not merely accelerate a task.
The Question I Would Put on Every AI Project
If I had to reduce this entire article to one question, it would be this:
“If AI makes this activity ten times faster, what happens to the rest of the system?”
Don’t accept:
“We save 100 hours.”
Ask:
Where do those 100 hours go?
Does the employee create more low-value work?
Does another department become the bottleneck?
Does the organization generate more information than it can consume?
Does quality fall because review becomes harder?
Does complexity increase?
Or does the organization redirect that capacity toward something genuinely valuable?
That is where the real answer lies.
Final Thoughts
I am excited about AI productivity.
But I am increasingly skeptical of productivity metrics that stop at the individual task.
Because the business does not exist as a collection of isolated tasks.
It exists as a system.
And systems behave differently from individual components.
AI can make a person faster.
It can make a workflow faster.
It can make a department faster.
But if the organization doesn’t redesign the system around that new capability, we may simply create more work, more information, more complexity, and more decisions.
We should not measure the success of AI by how quickly we can produce output.
We should measure it by what valuable outcomes become possible because of that speed.
That is a much harder measurement.
But it is also a much more meaningful one.
The AI era gives us an extraordinary opportunity.
We can use intelligence to automate work.
We can use it to augment people.
We can use it to redesign workflows.
And, perhaps most importantly, we can use it to finally question processes that have existed for years simply because nobody had a good enough reason to redesign them.
So before we celebrate another 80% productivity improvement, I think we should pause and ask:
Did we make the business better or did we simply make the work faster?
Because those two things are not the same.
And the organizations that understand that difference may be the ones that get the most from AI.
A Note for Leaders
The next time someone presents an AI business case with:
“This will save 10,000 employee hours per year,”
don’t immediately ask:
“How much money will that save?”
Ask five additional questions:
- What work is actually being eliminated?
- Where will those hours go?
- What downstream process will receive the additional output?
- What new risks or complexity does the AI introduce?
- What higher-value capability will the organization gain from the freed capacity?
Those questions may turn a simple automation project into a much better transformation decision.
Or they may reveal that the project shouldn’t exist at all.
Both outcomes are valuable.