Running the Machine While Changing the Machine: Why Brownfield AI Transformation Is the Real Game
Most AI transformation advice assumes you can start again.
New systems.
New processes.
New teams.
New data.
That is not how most organizations work.
Most organizations are brownfield organizations.
They already have employees, customers, policies, vendors, platforms, workflows, reporting lines, technical debt, and commitments that cannot simply be switched off.
The real challenge is not building an AI-native company from scratch.
It is running the machine while changing the machine.
That is the real game of AI transformation.
Brownfield Is Not a Technical Footnote
A greenfield organization can design its operating model around AI from day one.
A brownfield organization has to introduce AI into a system that already works well enough to keep people employed and customers served.
That difference changes everything.
Brownfield AI transformation means introducing new AI capabilities into an existing organization without breaking the work that keeps the organization alive.
It is not only an IT problem.
It affects:
- how work gets done;
- how decisions get made;
- how managers lead;
- how teams are structured;
- how risk is owned;
- how employees build skills;
- how HR measures productivity;
- how leaders prove return on investment.
Not a technology rollout. An operating model change.
This is why the standard AI playbook often fails.
It starts with the model.
Then the vendor.
Then the pilot.
Then the dashboard.
Only later does someone ask what changed for the people doing the work.
By that point, the organization has added another tool without removing any old work.
That is not transformation.
That is capability add.
The Moment the Strategy Became Clear
I recently had a conversation with a product leader who asked what I was seeing in organizations that were approaching AI transformation well.
I did not give him a list of tools.
I described two broad patterns.
The first pattern was familiar.
Everyone says, “Do not just use AI. Transform the process.”
That advice was useful a few years ago.
Now it is incomplete.
The business still has to happen as usual.
Payroll still has to run.
Employees still need answers.
Leaders still need reports.
Customers still expect service.
The second pattern was more practical.
Find the work causing the most pain.
Then use AI to solve that problem, sometimes through a language model, sometimes through deterministic code that AI helps you build.
That conversation reinforced a simple point.
The best AI transformation work does not begin with, “Where can we use AI?”
It begins with, “What work is creating the most friction, cost, delay, or risk?”
That is a better question for a CHRO.
It is also a better question for a consultant who wants to charge for meaningful work instead of another strategy deck.
The Brownfield Constraint
Existing organizations carry history.
Their processes often contain decisions nobody remembers making.
Their systems contain workarounds.
Their data contains conflicting definitions.
Their teams contain informal knowledge that never made it into a policy document.
Their leaders may agree on the goal but disagree on what the process actually is.
This creates a difficult reality.
Before you automate a process, you need to understand the process.
Before you give an agent permission to act, you need to understand what the agent is allowed to do.
Before you promise productivity, you need to identify which work will disappear.
Otherwise, the organization gets a new AI workflow and keeps the old workflow too.
People check the AI output.
Then they check the original system.
Then they update the spreadsheet.
Then they prepare the report.
The new tool produces activity.
The old work survives.
AI added. Work not removed.
The Four Choices Leaders Must Make
Brownfield AI transformation does not require one universal migration strategy.
It requires a clear choice about what happens to each part of the organization.
I see four practical choices.
1. Remediate in place
Keep the existing system.
Improve the data.
Document the process.
Add controls.
Build tests.
Use AI to make the current workflow more reliable and useful.
This works when the underlying structure is sound and the main problem is poor documentation, weak governance, or manual effort.
For HR, this might mean improving an existing workforce planning process instead of replacing the entire planning platform.
2. Build beside the old system
Create a new AI-enabled capability beside the existing one.
Connect the two carefully.
Move work across in stages.
Retire old components as the new capability proves itself.
This is often the most realistic path for large organizations.
The core system remains stable.
The new layer can evolve faster.
The risk is creating two systems that nobody fully owns.
That is why integration ownership matters.
3. Rebuild selectively
Sometimes the old structure cannot support the work the organization now needs.
The process may have too many exceptions.
The data model may be too rigid.
The system may be held together by undocumented workarounds.
In those cases, rebuilding one capability may be better than endlessly repairing the old one.
But rebuilding is not a blank cheque.
The new system must have a narrow scope, clear success measures, and a cutover plan.
Otherwise, the organization creates a second system that inherits every problem of the first.
4. Isolate and bypass
Some legacy systems should not be modernized.
They should be maintained at the lowest reasonable cost while new value is built around them.
This can feel uncomfortable.
Leaders often believe every old system must be fixed.
That is not true.
Some systems still perform a narrow system-of-record function but should not become the foundation for future innovation.
Do not modernize everything. Modernize what matters.
Start With Work, Not Tools
Vendors want to start with their product.
Consultants often want to start with their methodology.
Technology teams may want to start with the model.
Start with the work.
Map the workflow from beginning to end.
Identify where people wait.
Identify where information is re-entered.
Identify where decisions are repeated.
Identify where managers spend time collecting status instead of developing people.
Identify where employees search across five systems for one answer.
Then ask what can be:
- removed;
- compressed;
- automated;
- delegated to an agent;
- kept human because judgment still matters.
This is where HR transformation becomes AI transformation.
The People function sees work patterns across the organization.
It sees roles, skills, decision rights, incentives, learning gaps, and the human cost of change.
That gives HR a valuable position.
Not as the department that approves training after the technology has been selected.
As a co-architect of how the organization will work.
The Human Layer Is Part of the Architecture
A common mistake is treating people change as communication.
Send the email.
Schedule the town hall.
Publish the FAQ.
Move on.
That is not enough.
When AI changes work, it can change identity.
A recruiter may worry that their judgment no longer matters.
A manager may discover that status checking is no longer a large part of their job.
An analyst may be expected to build tools instead of only using them.
A consultant may need to stop selling access to information and start selling judgment under uncertainty.
These are not training problems alone.
They are role design problems.
They are management problems.
They are economic problems.
People need to know what is changing, what is not changing, and how value will be measured.
They also need a path to build new capability.
The emerging progression may look like this:
Use software. Configure software. Build lightweight software. Direct agents.
This does not mean every employee becomes an engineer.
It means more employees will need to understand how work can be structured, encoded, tested, and improved.
That is a major shift in the meaning of knowledge work.
What This Means for CHROs
The CHRO does not need to own every technical decision.
But the CHRO needs to own important organizational questions.
What work are we removing?
What work are we adding?
Which roles will change first?
What capabilities will become scarce?
Where does human judgment remain essential?
Who is accountable when an AI-enabled process fails?
How will productivity gains be used?
Will the organization reduce cost, increase capacity, improve quality, or pursue growth?
If the leadership team cannot answer these questions, it does not have an AI transformation strategy.
It has a collection of experiments.
The most valuable People leaders will help the organization connect AI investment to changes in work, workforce design, and business performance.
That is how HR moves from implementation support to enterprise leadership.
What This Means for Consultants
The consulting opportunity is not selling another AI readiness assessment that produces a colorful maturity chart.
The value is in helping a client make a consequential choice.
Should this process be remediated, rebuilt, isolated, or replaced?
Which work should disappear?
What controls are needed before automation?
What should the client build internally?
Where should it use a vendor?
What must change in roles, incentives, skills, and management?
Those are decisions clients will pay for.
Clients do not need more content about the future of work.
They need help changing work without damaging the business.
That is a more valuable offer.
It is also harder to deliver.
You need to understand the current operating reality.
You need to ask uncomfortable questions.
You need to show where the business is paying twice.
You need to connect technical decisions to workforce consequences.
Do not charge for information the client can find online. Charge for judgment that changes what the client does next.
That is the difference between a presentation and a transformation engagement.
The Practical Sequence
A brownfield AI transformation can begin with a simple sequence.
First, map the current machine
Document the systems, processes, data flows, decisions, roles, and failure points.
Do not trust the official process map until the people doing the work confirm it.
Second, identify work removed
Separate genuine work reduction from new AI-enabled activity.
If the new capability creates more review, maintenance, and coordination than it removes, question the business case.
Third, choose the transformation mode
Decide whether each capability should be remediated, built beside the old system, rebuilt, or isolated.
Do not use the same answer everywhere.
Fourth, create the safety layer
Define permissions, escalation paths, audit trails, fallback processes, and human decision points.
An AI agent should not receive broad authority because a demo looked impressive.
Fifth, redesign the work
Update roles, skills, manager expectations, incentives, and measures.
Do not leave the people system untouched while the workflow changes underneath it.
Sixth, ship a narrow prototype
Pick a painful process.
Build the smallest useful version.
Test it with real users.
Measure the work removed.
Then decide whether to scale, change direction, or stop.
This is not glamorous.
It is how transformation becomes real.
The Bigger Implication
AI-native startups get to design the machine around new technology.
Existing organizations have to change the machine while it is running.
That constraint is difficult.
It is also where much of the economic value will be created.
The organizations that win will not be the ones with the most pilots.
They will be the ones that connect AI to fewer handoffs, faster decisions, better service, stronger controls, and more capable people.
Not AI for its own sake.
Not transformation theatre.
Not vendor promises hidden inside a roadmap.
Work removed. Capability improved. Accountability preserved.
That is the standard.
If you are a People leader, the question is not whether your organization will use AI.
It is whether you will help design what the organization becomes when AI starts doing the work.
If you are a consultant, the question is not whether you can explain AI.
It is whether you can help a client make the machine better without stopping it.
Takeaway
Brownfield AI transformation is the real game because most organizations cannot start over.
They must protect today’s performance while building tomorrow’s operating model.
That requires more than choosing tools.
It requires understanding work, selecting the right transformation mode, removing real effort, redesigning roles, and governing new capabilities.
Run the machine. Change the machine. Measure what changed.
That is the work.
For more practitioner-focused thinking on AI transformation in organizations, explore the work of Vic Akosile and the AI Agents Course.
Sources and further reading
- Brownfield AI: Why the Specification Is the Migration Asset
- Brownfield and the AI Gap
- Brownfield Engineering Strategy
- The Strangler Fig Application Pattern
- Research, Review, Rebuild: AI-Assisted Legacy Modernization
