The Great Clearing · Article 2 · AI Doesn’t Fix Broken Processes. It Industrializes Them.
There is an understandably intoxicating idea spreading through business right now.
Take every repetitive task. Give it to an agent.
Give the agent another agent.
Let them work nights.
Let them work weekends.
Give them access to the CRM, the calendar, the inbox, the inventory feed, the phone system, the website, the reporting stack, and perhaps—once everyone gets sufficiently comfortable—the nuclear codes.
Then stand back and admire the efficiency.
There is only one inconvenient question:
Was the thing we automated worth doing in the first place?
That question matters enormously in automotive retail because our industry enters the agentic era carrying decades of accumulated process. Some of it exists because it genuinely protects customers, employees, dealers, lenders, and manufacturers. Some exists because automotive transactions are complicated and high-stakes. Some represents hard-earned operating knowledge that would be foolish to discard simply because a model can now complete a form.
And some of it exists because somebody created a workaround in 2007, trained the next person to do it, added three fields to the CRM, and nobody has been brave enough to ask why since.
AI does not know the difference automatically.
Leadership has to.
AI is not inherently transformative. It is an amplifier. Give it a great process and it can create extraordinary leverage. Give it a bad process and it can make the bad process faster, cheaper, more persistent, and much harder for a human being to escape.
This is where the next phase of automotive AI becomes less a technology conversation and more an executive leadership test.
For most of automotive history, labor imposed a natural constraint on bad process.
You could ask an employee to perform an inefficient task, but eventually you had to pay somebody to perform it. You could require a BDC agent to make another round of calls, but there were only so many calls one person could make. You could ask the marketing team to manually rebuild content across six systems, but eventually somebody would run out of Tuesday.
Human limitation created friction, but it also created a kind of accidental governance.
At some point somebody would say: This is ridiculous.
Agentic systems can remove that constraint.
An AI agent can monitor thousands of records, respond to thousands of triggers, create thousands of variations, initiate thousands of workflows, and theoretically continue until the infrastructure underneath it asks for mercy.
That is powerful.
It is also why the executive question changes.
When labor becomes dramatically cheaper, the cost of the task can no longer be the primary reason to question the task.
The organization has to understand its purpose.
Imagine an automotive lead process built around the assumption that persistence is inherently valuable. A shopper raises a hand. The CRM creates tasks. Email begins. SMS begins. Calls begin. Another system starts nurturing. The salesperson receives an alert. The BDC receives an alert. A vendor may receive the same lead. Someone creates a cadence because the last cadence didn’t produce enough contact attempts.
Now add autonomous agents.
Congratulations.
The customer who ignored twelve follow-ups can now ignore twelve hundred for a fraction of the labor cost.
The efficiency gain is spectacular.
The experience is still terrible.
That sounds obvious when presented as an absurdity. The dangerous versions will not be that obvious. They will arrive as clean dashboards showing response-time reductions, task completion, increased activity, lower labor costs, and impressive demonstrations in which the AI never becomes tired, distracted, annoyed, or late for lunch.
Those are legitimate capabilities.
They simply do not answer the most important question:
Did we improve the experience, or did we merely remove the cost that previously forced us to notice how bad the experience was?
Technology leaders talk about technical debt: shortcuts, outdated architectures, brittle integrations, and old decisions that become progressively harder to maintain as systems evolve.
Automotive has plenty of that.
We also have process debt.
Process debt accumulates when an organization solves yesterday’s operating problem and then continues performing the solution long after the conditions that created it have changed.
A customer could not complete a financing step digitally, so we built a handoff. A platform could not talk to another platform, so an employee became middleware. A salesperson could not know when a customer returned to the website, so we created another alerting product. Lead response was inconsistent, so we built increasingly rigid cadences. Pricing lacked transparency, so we developed scripts for explaining the lack of transparency. Reporting systems disagreed, so teams learned which spreadsheet to trust.
Each decision may have been rational in its original context.
Stack enough of them together and the organization eventually starts treating the workaround as the operating model.
This matters because AI encounters an existing process with no institutional memory of why the process exists unless we give it that context. If leadership says, “automate our lead follow-up,” the system’s job is to become extremely good at lead follow-up. It does not spontaneously convene an executive off-site and ask whether the dealership should still be treating every expression of curiosity as permission for a fourteen-day pursuit campaign.
The same problem exists inside the employee experience.
A service advisor may spend meaningful portions of a day moving information between systems because systems do not move the information themselves. Marketing teams routinely download data from one platform, restructure it, upload it somewhere else, confirm it rendered correctly, and then manually construct the report proving they completed the manual work. Managers reconcile dashboards instead of managing.
Automating those actions can create real value, but there is an important distinction between automating a necessary task and automating the human workaround for an unnecessary architectural constraint. If software A and software B should simply exchange information, the transformative end state is probably not a highly intelligent digital employee eternally copying fields between them.
We do not need armies of agents condemned to meaningless digital existence recreating the swivel-chair integration problem at machine speed.
FSometimes the transformative answer is an API.
Executives tend to experience technology through reports.
Customers experience it through moments.
They experience whether the information they entered online survived the drive to the dealership. They experience whether the salesperson knows what was already discussed. They experience whether an advertised vehicle actually exists. They experience whether the price changes when the conversation changes surfaces. They experience whether the service appointment knows why they are coming. They experience whether “personalization” means somebody remembered something useful or merely inserted their first name into another automated message.
This distinction is important because the emerging evidence around AI and digital retail does not suggest customers are demanding maximum automation.
It suggests they value better experiences.
Cox Automotive’s 2026 Car Buyer Journey study found that 25% of new-vehicle buyers had used AI during the shopping process, and 59% of respondents who used AI-powered assistance reported high satisfaction with it. Among mostly digital buyers using AI assistants, 84% reported high satisfaction with the overall buying process. Cox’s findings point toward something more nuanced than “customers want bots.”
They point toward customers valuing tools that make the process clearer, easier, and more controllable.
Cox has subsequently emphasized the importance of flow: customers become frustrated when progress made digitally disappears during a handoff, information must be re-entered, numbers need to be reconstructed, or the next employee lacks the context established earlier in the journey.
That’s a crucial distinction.
The customer does not care how many processes we automated. The customer cares whether the business feels easier to deal with.
This is why the AI transformation cannot be measured only through labor reduction, response time, or task completion.
If an agent answers a customer’s question instantly but answers incorrectly, we did not improve the experience.
If AI removes an employee from the conversation but also removes empathy precisely when the customer needs judgment, we did not improve the experience.
If an automated workflow moves faster but forces the customer through the same unnecessary steps, we improved throughput.
Throughput and experience are not synonyms.
That difference becomes particularly important in automotive because buying and servicing a vehicle contains moments with dramatically different emotional and economic stakes. Checking whether a vehicle is available is different from discussing negative equity. Rescheduling an oil change is different from explaining why a transmission failed. Answering a question about cargo capacity is different from helping a family understand whether the payment still fits after their financial circumstances changed.
Good operating design recognizes those differences.
Bad automation treats them as workflow objects.
Before executives ask where to deploy agents, I think they should force every meaningful process through four different decisions.
Some work should not be automated because it should not exist.
Duplicate entry is an obvious example. So are manual transfers between systems that could exchange structured information directly, reports nobody meaningfully uses, approval chains created around technical limitations that no longer exist, and customer questions that exist primarily because the business failed to make something understandable upstream.
The cheapest agent in the world cannot beat deleting unnecessary work.
This category deserves much more executive attention because AI makes it tempting to preserve process debt indefinitely. Why redesign the workflow when software will perform the bad workflow for almost nothing?
Because customers still experience the workflow.
Some work genuinely needs to occur and does not become more valuable because a human performs it.
Data reconciliation. Routine status updates. Document classification. Information retrieval. Monitoring inventory changes. Preparing summaries. Scheduling predictable tasks. Checking defined conditions. Moving approved information through repetitive workflows.
This is where agents can create enormous operating leverage.
The goal should be reliable execution inside clearly understood boundaries, with escalation paths proportional to the consequences of being wrong.
This may be the most strategically interesting category.
Some work becomes dramatically better when AI increases the capability of the human rather than removing the human from the work.
Give a service advisor immediate access to the relevant vehicle history, technical knowledge, customer context, and likely questions before the conversation begins.
Give a salesperson an intelligent briefing on what the customer researched, what inventory actually fits, which incentives apply, and which unresolved questions remain—without forcing the customer to reconstruct the journey.
Give a technician faster access to diagnostic history and trusted technical information.
Give a marketing leader the ability to see what customers are asking across search, sales, service, social, and AI environments and identify which employee inside the dealership possesses the best answer.
None of those examples removes the person.
They remove the information disadvantage surrounding the person.
Finally, some moments should remain intentionally human because human judgment, accountability, empathy, discretion, or trust is part of the value being created.
That boundary will differ by dealership and use case, and it should evolve as the technology proves itself. The point is not that humans must ceremonially touch every decision forever.
The point is that leadership should identify meaningful human moments deliberately rather than discovering after deployment that the organization accidentally automated them away.
The strategic question is not “Where can we use AI?” It is “Which work should disappear, which should become autonomous, which humans should become more capable, and which moments are important enough to protect?”
That is a much harder exercise than buying an agent.
It is also much closer to transformation.
The word ethics can make executive technology conversations sound philosophical when the consequences are often intensely operational.
Every autonomous system eventually encounters decisions about authority. Who is allowed to make a promise to a customer? Which information can be used in which context? When does an automated recommendation become consequential enough to require review? What should be logged? What can be reversed? Who owns the outcome when software acts incorrectly?
Those are ethics questions, but they are also architecture questions because the answers have to be represented somewhere in the system.
NIST’s AI Risk Management Framework approaches responsible AI as a lifecycle discipline rather than a disclaimer attached after deployment. Its emphasis on governing, mapping, measuring, and managing risk is useful because trustworthy automation depends on understanding context, consequences, responsibilities, and failure modes before the system is given meaningful authority.
That becomes increasingly important as the industry moves from generative systems that produce language toward agentic systems capable of initiating actions. A generated paragraph can be reviewed before publication. An autonomous system may change a record, send a communication, schedule an appointment, alter a workflow, publish an offer, or eventually participate in increasingly consequential commercial actions.
The distance between an incorrect answer and an incorrect action is strategically enormous.
Leadership therefore has to look beyond whether the model is accurate and examine the objective encoded into the system. An agent told to maximize appointments may become extraordinarily persistent unless the business has also defined what respectful customer engagement means. A system rewarded for reducing handle time may become hostile to the complicated customer who genuinely needs more time. A tool optimized around gross may discover forms of persuasion leadership would never tolerate from a human employee.
Every optimization target contains an implicit value judgment.
AI ethics begins long before the disclaimer. It begins with what leadership tells the system to optimize.
Humans have always made these choices. AI simply gives organizations the ability to encode them into systems capable of executing at extraordinary scale. That makes accountability more important, not less.
The organization chooses the system, determines its access, defines the data it can use, establishes the objective, selects the metrics, and benefits from the productivity. Responsibility cannot conveniently disappear at the moment automation produces an outcome nobody intended.
There is another reason I think automotive should resist framing the AI transition primarily as a labor-replacement exercise.
Our industry’s competitive advantage is unusually human.
Dealerships have physical presence. They have technicians with decades of accumulated expertise. They have service advisors who can hear something in a customer’s description that never appears on a diagnostic code. They have salespeople who can recognize when the actual objection is different from the stated objection. They have managers who can solve exceptions. They have employees who live in the community and understand how vehicles are actually used there.
Those human assets are expensive when we bury them in administrative work.
For decades, automotive technology has often promised to make employees more productive and then rewarded that productivity by giving them more systems to operate.
A salesperson becomes the integration layer between the CRM, inventory tool, desking platform, messaging app, digital retail tool, and customer.
A service advisor becomes a highly trained data-entry professional who occasionally gets to advise someone about service.
A marketer spends time resizing, reformatting, republishing, exporting, importing, reconciling, and reporting instead of understanding the market.
AI gives us an opportunity to reverse that relationship.
Technology can become responsible for more of the machinery surrounding the employee so the employee can become responsible for more of the value only a capable human can create.
The salesperson can listen.
The advisor can advise.
The technician can diagnose.
The marketer can understand.
The manager can manage.
That isn’t anti-automation.
It is a more ambitious use of automation.
Use AI to eliminate meaningless work before you use it to eliminate meaningful people.
There will absolutely be jobs that change. There will be tasks that disappear. Roles will be redesigned, and some organizations will operate with different staffing models than they do today. Pretending otherwise is not useful leadership.
But “How many people can we replace?” is a remarkably unimaginative first question for a technology capable of making every person in the organization dramatically more capable.
The better question is what an exceptional salesperson, advisor, technician, creator, marketer, or manager becomes when the administrative friction surrounding their expertise begins disappearing.
That is where the customer experience gets interesting.
The irony of autonomous systems is that they may require leadership to become more intentional.
When a person performs a task, culture, experience, judgment, social pressure, supervision, and ordinary human hesitation all participate in the decision. None is perfect, but together they create context.
When a system performs the task, leaders have to decide which pieces of that context must become explicit.
What are the permissions?
What is the acceptable confidence threshold?
Which sources can be trusted?
When should the system stop?
When should it escalate?
What does the employee need to know before taking over?
What can the customer contest?
What gets logged?
What can be undone?
Who reviews performance?
How do we know the agent is improving the customer experience rather than merely achieving its assigned metric?
This is governance, but it should not become governance theater.
The purpose is not to create an AI committee capable of producing a 94-page PDF explaining why nobody is allowed to use AI.
The purpose is to create enough organizational clarity that powerful systems can be deployed confidently where they create value and constrained where consequences demand more care.
That is what responsible acceleration looks like.
And the organizations that do this well will probably move faster than organizations without governance, not slower, because teams will understand where autonomy is encouraged, where experimentation is appropriate, and where the boundaries actually are.
Clear roads are easier to drive quickly than roads with invisible cliffs.
Article 1 of this series argued that the executive opportunity is larger than AEO, GEO, or the current optimization vocabulary. The business itself needs to become more intelligible: a coherent environment of knowledge, context, provenance, permissions, human expertise, and activation.
This article introduces the responsibility that arrives with that architecture.
Once the organization becomes intelligible to machines, machines can do more with the organization.
That is exciting.
It should also make leadership considerably more interested in what gets amplified.
A great service process supported by AI can become faster without losing care.
A knowledgeable employee supported by AI can become dramatically more useful to the customer.
A clean knowledge environment can prevent customers and employees from repeatedly reconstructing context.
An agent can eliminate hours of low-value work nobody will miss.
Those are real operating advantages.
But the inverse is equally real.
Bad data can travel farther.
Bad incentives can operate faster.
Manipulative processes can become more persistent.
Fragmented systems can create automated confusion.
Meaningless content can become essentially infinite.
And bad customer experiences can become exceptionally efficient.
The defining AI leadership skill may not be knowing how quickly the organization can automate. It may be knowing where to stop long enough to ask whether the process deserves to survive.
This is The Great Clearing.
Clear the unnecessary work.
Clear the inherited assumptions.
Clear the idea that activity equals value.
Then use intelligence aggressively where it makes customers’ lives easier, employees more capable, information more trustworthy, and the organization better at keeping its promises.
AI should amplify the best version of the business we intend to become.
Not immortalize every process we inherited on the way here.
And that creates the next executive problem.
Because the cost of building software is collapsing at precisely the moment every vendor, consultant, startup, agency, legacy platform, and newly minted “AI expert” has discovered the same technology.
Some of what arrives will be transformative.
Some will be genuinely useful.
Some will be a wrapper around a model with an impressive gradient.
Some will be built Thursday night and enterprise-ready by Friday morning.
Leadership is going to need a filter.
Next: Welcome to the Slop Economy: Executive Discretion Is Now a Core Competency. →
Previous: AEO Is Dead. Long Live LMNOPEO. →
Return to the series hub: The Great Clearing: AI Is About to Amplify Everything. Choose Carefully.
NIST AI Risk Management Framework →
A practical framework for governing, mapping, measuring, and managing AI risk as systems become more consequential.
2026 Cox Automotive Car Buyer Journey Study →
Current evidence on how AI, digital efficiency, continuity, and customer confidence are changing the vehicle-buying experience.
The Human Signal Surge →
Why identifiable expertise and firsthand experience become more—not less—valuable as generated information becomes abundant.
Where Car Buyers Actually Search in 2026 →
Understand the increasingly distributed customer journey across search, AI, social, video, dealer websites, and physical retail.
Many of the processes most tempting to automate exist because dealership systems still cannot securely exchange information with the tools, employees, agencies, and technology partners the dealer chooses.
The better future is not an intelligent agent spending eternity copying dealership-owned information between closed platforms. It is authorized interoperability that lets intelligence move where it is useful while preserving permissions, governance, provenance, and dealer control.
APIs. Webhooks. Authorized third-party access. Real documentation. Modern authentication. MCP and emerging agent standards where appropriate. Portability. Dealer choice.
Read and sign the Open Interoperability Demand →
Automate the value. Eliminate the workaround. Let dealers plug in.
Hrizn v6 helps dealership marketing teams move from fragmented activity toward a connected operating advantage—bringing intelligence, creation, human participation, distribution, proof and improvement into a more coherent system.
The goal is not to remove the people who make dealership expertise valuable. It is to reduce the fragmented work around them, connect what the organization knows, and give capable people better intelligence and more powerful ways to turn that knowledge into customer value.
We Rise Together.
“`
