
Why the smartest software founders may stop selling software and start owning the businesses that use it. And why the next scarce skill in an AI economy may be knowing what is worth doing.
For roughly fifteen years, there was an excellent answer to the question of what a technically competent entrepreneur should build.
Software.
Preferably software for other businesses.
Find an industry with inefficient processes. Build a SaaS product. Charge a monthly subscription. Acquire customers. Scale.
The appeal was obvious.
Software has low marginal distribution costs. A product built for one customer can often be sold to thousands. Recurring revenue makes the business easier to forecast. And a sufficiently successful SaaS company can grow without acquiring factories, warehouses, or a particularly impressive collection of delivery vans.
Meanwhile, the actual businesses using the software continue doing their complicated, physical, regulated, occasionally unpleasant work.
The accounting firm handles the accounts.
The insurance broker deals with customers and claims.
The property management company handles tenants.
The law firm deals with clients.
And the software founder sells them another dashboard.
For a long time, that was a very sensible division of labour.
But AI is beginning to change the economics.
Building software is becoming cheaper.
Businesses can increasingly create tools for their own needs.
And the competitive advantage of producing another application may be shrinking in markets where existing products are easy to reproduce.
The obvious question is what happens to the value that software still creates.
Because the value has not disappeared.
Quite the opposite.
If a business can use AI and automation to operate substantially more efficiently, someone will capture that economic benefit.
And I'm beginning to suspect that more technology entrepreneurs will decide they would rather capture it directly.
Not by selling software to an accounting firm.
But by owning the accounting firm.
The Strange Problem with Selling Software
The traditional SaaS model depends on an important economic relationship.
Software is expensive to create, but relatively cheap to reproduce.
That makes selling the same product to many customers attractive.
Suppose building a specialised system costs hundreds of thousands of dollars.
A single small company may not be able to justify developing it.
But if a software vendor sells the system to hundreds of companies, the development cost can be distributed across the customer base.
The vendor earns recurring revenue.
Each customer receives capabilities that would have been too expensive to develop independently.
Everyone has a reason to participate.
Now change one assumption.
What if the cost of building the specialised software falls dramatically?
Not necessarily to zero.
Production software still needs testing, security, integration, maintenance, and support.
But enough that a small business can create useful internal systems without hiring a large development team.
Suddenly, the calculation changes.
Why pay for an expensive specialised application if a coding agent can build most of the functionality you actually need?
Why subscribe to five different workflow tools if your own system can integrate the relevant operations?
Why accept a generic product's limitations when you can automate precisely the way your business works?
This doesn't mean SaaS disappears.
Reliable, secure, maintained software still provides enormous value. Shared products also benefit from integrations, compliance expertise, network effects, and improvements distributed across many customers.
But the decision to buy rather than build is becoming less obvious.
And that is not just speculation.
McKinsey's 2026 survey of AI adoption found that 32% of respondents at organisations regularly using AI had decided against purchasing at least one software product or feature because agentic coding tools enabled them to develop the functionality internally.
That is not evidence that a third of all software spending has disappeared.
It is evidence that the economics of buying software are changing.
And this creates a new question for technical entrepreneurs.
If software has become much cheaper to build, why keep selling it instead of using it to run a business better than your competitors?
The Software Is No Longer the Entire Business
Imagine two entrepreneurs.
The first builds AI-powered software for accounting firms.
The product automates document classification, transaction categorisation, preliminary reconciliations, and preparation of financial reports.
The founder must find customers.
Explain the product.
Demonstrate reliability.
Negotiate contracts.
Integrate different accounting systems.
Handle support.
And convince established firms to change their workflows.
The second entrepreneur acquires or starts an accounting practice.
They develop similar automation.
But instead of selling the software, they use it internally.
The economic benefit appears directly in the operating business.
Less repetitive administrative work.
More clients handled by the same team.
Faster turnaround.
More consistent documentation.
Potentially better margins.
The second entrepreneur doesn't need to persuade hundreds of other businesses to adopt the system.
They need one business in which the system produces enough value to justify its cost.
This is a fundamentally different proposition.
The software becomes an operational advantage rather than the product being sold.
And the customer pays for the outcome.
Completed accounts.
Processed insurance policies.
Managed properties.
Translated documents.
Legal work delivered by qualified professionals.
The client does not need to know which model or workflow produced the result.
They care about quality, reliability, price, and whether the work gets done.
This is what I find so interesting about the shift.
When building software becomes cheaper, owning the business that benefits from the software may become more attractive than owning the software alone.
Why One Business May Be Enough
The emerging AI-rollup industry has already recognised part of this opportunity.
An AI rollup acquires several businesses, often in a fragmented services sector, then attempts to improve their economics using shared technology, processes, and management.
Accounting, legal services, insurance brokerage, property management, and other professional services are natural candidates.
The attraction is clear.
Many such firms have established customers and recurring demand.
They may still rely on fragmented software, manual administration, and procedures that have changed surprisingly little over time.
Acquiring them provides something an AI startup often lacks.
Customers.
Revenue.
An operating history.
Relationships.
Domain expertise.
And an existing mechanism for delivering a service people are already willing to buy.
By 2026, companies such as Thrive Holdings and its accounting platform Current had helped make this acquisition-led model visible.
But a technical entrepreneur doesn't necessarily need to start with a complicated rollup strategy.
You don't have to buy ten companies.
You could buy one.
Improve it.
Learn the industry.
Develop automation around its actual problems.
Increase its capacity.
And grow from there.
This is closer to entrepreneurship through acquisition than to a conventional venture-backed software startup.
And for some founders, it may be a more sensible route.
A functioning service company has customers on day one.
Of course, it also has problems on day one.
That's the part the acquisition presentations tend to mention rather less enthusiastically.
The Accountant Who Still Needs to Be an Accountant
There is an important limitation to this idea.
Knowing how to automate accounting workflows does not automatically make someone qualified to run an accounting firm.
Professional services are often regulated.
They involve legal responsibility, licensing, confidentiality, and standards that vary by country.
An accounting firm's work may require review or sign-off by qualified professionals.
Insurance brokerage involves licensing and compliance.
Legal practice has strict rules concerning professional authority and ownership in many jurisdictions.
Notarial activities can be even more restricted.
The fact that an AI agent can draft a document does not make the person operating it legally authorised to provide a regulated service.
And this distinction matters enormously.
A technical founder cannot simply acquire a regulated business, replace everyone with agents, and declare the problem solved.
They need the right legal structure.
Professional leadership.
Reliable quality controls.
Data protection.
Insurance.
Human accountability.
And an understanding of how the service is actually delivered.
There are also tasks that are difficult to automate because their value lies partly in human relationships.
A client may want someone to explain a difficult financial decision.
A family may need sensitive legal advice.
An insurance customer may require support during a distressing claim.
A property owner may need someone to handle an argument with a tenant.
Some tasks are not simply information transformations.
They are relationships involving responsibility, judgment, and trust.
But this doesn't invalidate the opportunity.
It changes the strategy.
The most promising business may not be one where AI eliminates the professionals.
It may be one where a small number of excellent professionals can serve substantially more clients because AI removes much of the work surrounding their expertise.
Research is beginning to support this distinction.
A 2026 study in the Journal of Accounting Research, based partly on more than 200,000 transaction records from an AI-enabled accounting platform, found productivity improvements associated with AI use and a shift in accountants' work toward communication and quality assurance.
It also found that following certain questionable AI recommendations could increase errors.
In other words, the technology can improve the economics of the service.
But the professional judgment remains essential.
That sounds less exciting than replacing the entire accounting profession.
It also sounds considerably more practical.
The Best AI Business May Be One Where Nobody Talks About AI
Consider what traditional service businesses actually sell.
They rarely sell technology as such.
They sell completed work and confidence that the work will be done correctly.
An insurance broker sells access to appropriate coverage and guidance through the process.
A property manager sells the ability to stop worrying about the daily operation of a property.
A translation agency sells accurate communication across languages.
An accounting firm sells reliable financial work and compliance.
The technology is part of the production system.
Not necessarily part of the customer's identity.
This creates an interesting difference from many AI startups.
A software founder may spend enormous effort explaining why their product uses the latest model.
A service business owner may achieve better results by never mentioning the model at all.
The customer doesn't need another AI dashboard.
They need the problem solved.
And if the service becomes faster, more reliable, or less expensive, that may be enough to create a competitive advantage.
There is also an attractive economic consequence.
Software companies often have to prove that their customers will adopt a new workflow.
An operator can redesign the workflow directly.
They control the process.
They see where the bottlenecks occur.
They can compare the results with actual costs.
And they can observe whether automation improves the economics of the business rather than merely producing an impressive demonstration.
This is a much tighter feedback loop.
The technology is connected directly to financial outcomes.
And financial outcomes have a wonderful habit of being less impressed by AI demos than venture capital presentations are.
The Real Bottleneck Is Moving
This brings me to a related shift.
For much of the 2024–2026 period, enterprise AI adoption was discussed primarily through task automation.
Which activities can be delegated?
How many hours can be saved?
Which documents can be processed automatically?
Which internal jobs can be simplified?
And, less pleasantly, how much labour cost can be reduced?
That phase isn't over.
Many organisations are still struggling to make even basic automation reliable.
But I suspect the next stage, particularly during 2027, will place much greater emphasis on something else.
Building complete systems for setting objectives, executing work, checking results, and deciding what happens next.
The distinction is important.
An AI assistant can perform an individual task.
An agent can execute a sequence of tasks.
A workflow can organise several agents and conventional software components.
But a business requires more than completed tasks.
It requires objectives.
Priorities.
Constraints.
Budgets.
Decisions about risk.
And a way to establish whether the work produced anything valuable.
As agents become more capable, the bottleneck moves.
Initially, the difficult part is performing each operation.
Then it becomes supervising the execution of the complete workflow.
Then it becomes deciding when exceptions require human judgment.
Eventually, a greater share of management attention can shift toward deciding which outcomes the organisation should pursue.
And that is a surprisingly difficult problem.
The Agent Completed Every Task. The Business Still Lost Money.
Imagine an AI-powered property management company.
Its agents answer messages promptly.
Prepare excellent documents.
Analyse maintenance requests.
Generate weekly reports.
Schedule contractors.
Update databases.
And produce beautifully formatted summaries of everything they have accomplished.
From an automation perspective, the system is performing magnificently.
But suppose it consistently prioritises inexpensive tasks over costly problems that matter to customers.
Or approves unnecessary maintenance.
Or focuses on reducing response times while ignoring tenant satisfaction.
Or pursues occupancy growth without considering payment defaults and legal obligations.
The agents are working.
The workflows are running.
The dashboards are green.
And the business is becoming worse.
The problem is not necessarily intelligence.
It is the objective.
A system can execute a badly chosen goal with extraordinary efficiency.
In fact, automation can make a bad objective more dangerous because it allows the organisation to pursue that objective at a much greater scale.
This is why the ability to define useful goals may become increasingly valuable.
Not goals such as:
Use AI to improve productivity.
Or:
Automate 80% of customer communication.
Those are activity targets.
They may be useful intermediate metrics, but they do not necessarily describe an economic outcome.
A stronger objective might be:
Reduce the time required to onboard a new accounting client by half while maintaining review quality and client satisfaction.
Or:
Increase the number of properties managed per employee without increasing unresolved maintenance cases.
Or:
Expand translation capacity while preserving accuracy in high-risk documents.
These objectives connect a technical capability to a business result.
They also contain constraints.
And constraints matter.
Without them, an agent may find remarkably efficient ways of achieving the metric while damaging the underlying business.
From Execution to Goal Setting
I see the evolution of AI-enabled management as a gradual shift through several levels.
<box gap={3}> <box border radius="lg" padding={3} gap={2}> <text weight="medium">1. Task execution</text> <text color="secondary" size="sm">Humans define and supervise individual operations. AI helps perform them.</text> <row align="center" justify="center"><icon name="arrow-down" color="tertiary"/></row> <text weight="medium">2. Workflow execution</text> <text color="secondary" size="sm">AI handles more of the process. Humans monitor results and intervene when something fails.</text> <row align="center" justify="center"><icon name="arrow-down" color="tertiary"/></row> <text weight="medium">3. Decision governance</text> <text color="secondary" size="sm">Humans retain authority over high-impact actions, budgets, hiring, payments, legal responsibility and exceptions.</text> <row align="center" justify="center"><icon name="arrow-down" color="tertiary"/></row> <text weight="medium">4. Goal design</text> <text color="secondary" size="sm">Human leadership increasingly concentrates on choosing outcomes, priorities, acceptable risks and measures of success.</text> </box> </box>
This is a directional model, not a claim that every company will pass through four clean stages.
And I wouldn't assume that an agent becomes completely reliable simply because it has performed a task successfully a hundred times.
Rare failures, unexpected inputs, malicious instructions, and changing external conditions remain important.
The appropriate level of autonomy should depend on evidence, risk, monitoring, and the consequences of error.
But as bounded tasks become more reliable, the nature of human work can change.
You no longer need someone to supervise every individual step.
You need someone who can design the system.
Understand when it is failing.
Decide which outcomes matter.
And take responsibility for decisions that should not be delegated.
That is a different management skill.
And I think it will become significantly more valuable.
The Scarce Resource Is Not Another Prompt
The technology industry has become extraordinarily focused on how to make agents perform tasks.
Better prompts.
Better tools.
Better memory.
More context.
More elaborate orchestration.
These things matter.
But eventually, the ability to execute a task becomes less interesting than the ability to choose a worthwhile task.
If an agent can build an application cheaply, which application should exist?
If it can create a marketing campaign, what should the campaign accomplish?
If it can analyse an industry, which opportunity is worth pursuing?
If it can run a complex workflow, what result should the workflow optimise?
This is where understanding customers, markets, and incentives becomes particularly important.
A technically excellent entrepreneur who understands nothing about the economic structure of an industry may struggle to identify useful objectives.
Meanwhile, someone who deeply understands a business can recognise opportunities that a general-purpose model might overlook.
They know which customer complaints matter.
Which processes are expensive.
Which regulations cannot be ignored.
Which services customers actually value.
And which improvements translate into better economics.
The combination becomes powerful.
Domain knowledge provides direction.
AI provides execution capacity.
Software engineering provides the infrastructure.
And business ownership captures the resulting value.
This is why I expect more technical founders to move toward traditional industries.
Not because software skills have become irrelevant.
But because those skills may become more valuable when applied directly to an operating business.
There's Also a Less Rational Reason
There is another reason I expect this shift.
Some technology entrepreneurs are simply exhausted.
I increasingly encounter people who no longer seem particularly excited by the idea of launching another AI product.
They have spent years keeping up with model releases.
Frameworks.
Agents.
Benchmarks.
New interfaces.
New development tools.
And an endless stream of announcements explaining why the thing they built last month is now obsolete.
The pace is extraordinary.
But the emotional experience is not always particularly pleasant.
There is a strange pressure to remain permanently available for the next technological breakthrough.
A new model appears.
You must test it.
A new coding agent is released.
You must rebuild your workflow.
A new competitor launches.
You must investigate.
A framework changes.
You must migrate.
Then, just as everything begins working, someone publishes a demonstration of an agent doing the entire job with three lines of code and a slightly smug thread on X.
Excellent.
Back to work.
For some founders, the problem is no longer a lack of opportunity.
It is the absence of any point at which they can reasonably stop.
And I understand the attraction of doing something more tangible.
Running a local business.
Providing a real service.
Working with customers whose needs aren't redefined every time a frontier model receives a new release number.
Perhaps occasionally going outside.
Touching grass.
Experiencing daylight without checking whether an autonomous coding agent has finished its seventh refactor.
This isn't a claim that traditional businesses are stress-free.
Anyone who has managed employees, customers, payroll, regulators, or commercial property would find that suggestion hilarious.
The difference is the nature of the work.
A traditional service business has relatively stable underlying needs.
People still require accounting.
Insurance.
Housing.
Legal assistance.
Maintenance.
Translation.
Those needs may change in how they are delivered, but the demand does not disappear simply because a new model is released.
For a founder tired of competing on the frontier of AI capabilities, that stability can be attractive.
The technology becomes a tool.
Not the entire reason for the company's existence.
Owning a Business Is Harder Than Building Its Software
I don't want to romanticise this transition.
Buying a small business is not the same as downloading a repository.
You need capital.
Due diligence.
Legal work.
An understanding of liabilities.
A realistic assessment of customer concentration.
Staff relationships.
Cash flow.
Operational risks.
And enough industry expertise to recognise when something that looks inefficient is actually preventing a serious problem.
An apparently redundant manual review may exist because an earlier automated process caused expensive errors.
An old-fashioned customer relationship may be the reason clients remain loyal.
A complicated compliance procedure may be legally mandatory.
And a business with attractive margins may depend heavily on an owner whose departure would take the customers with them.
These are not problems a coding agent automatically solves.
Nor is acquisition financing universally easier than startup financing.
Lenders and investors will examine cash flow, collateral, management experience, deal structure, and risk. A business with established revenue may support different financing options from a speculative software startup, but a poor acquisition remains a poor acquisition.
There are several reasonable ways to begin.
A founder with capital may acquire a business.
A founder without sufficient capital may partner with an existing operator.
A technical person may join an established practice and negotiate an ownership position linked to performance.
Or they may bootstrap a new service firm in an area where licensing and startup costs permit it.
The best path depends on the industry.
But the central idea remains.
You don't necessarily need to own a portfolio of businesses.
One good business may be enough.
Especially if you understand precisely where automation can create durable value.
How I Would Evaluate the Opportunity
I would start with the economics of the service, not the capabilities of the model.
First, is there recurring demand?
Second, what proportion of the work involves repeatable information processing?
Third, which activities require professional judgment, licensing, or personal relationships?
Fourth, how much of the current cost structure can realistically be improved without damaging quality?
Fifth, can the resulting improvement be captured through higher margins, greater capacity, better service, or growth?
And finally, is the competitive advantage durable?
That last question is essential.
If everyone can build similar automation, your advantage may disappear quickly.
Perhaps the real asset is the customer relationship.
Perhaps it is a proprietary dataset.
Perhaps it is a trusted brand.
Perhaps it is a specialised distribution channel.
Perhaps it is operational knowledge accumulated over years.
Or perhaps the business has no meaningful advantage beyond being first to deploy a temporary technological capability.
That's not necessarily a reason to avoid it.
But it changes the investment thesis.
The most promising opportunity is not always the industry with the greatest percentage of tasks theoretically exposed to AI.
It may be the industry where improvements can actually be implemented, accepted by customers, and converted into durable economics.
There is a substantial difference between a model being able to perform a task and a business being able to profit from deploying it.
The 2027 Shift: From Automation to Operating Systems
My expectation for 2027 is that the more sophisticated AI-enabled businesses will increasingly redesign themselves around complete operating systems.
Not just agents performing individual jobs.
But systems that connect objectives, execution, observation, decision-making, and improvement.
For example, a business might define an objective for customer onboarding.
The system identifies incoming requests.
Checks required documentation.
Prepares responses.
Routes exceptions.
Monitors completion times.
Measures customer satisfaction.
And reports whether the process is improving.
A human remains accountable for the high-impact decisions.
But much of the routine coordination becomes automated.
The goal is not to replace human judgment everywhere.
It is to reserve judgment for the places where it matters most.
The organisation becomes less dependent on people coordinating every small activity.
And more dependent on people who can recognise which activities should happen in the first place.
This also explains why successful AI adoption may look less like a collection of chatbots and more like a redesigned business process.
Deloitte's 2026 research found that only a small minority of surveyed organisations considered their processes highly prepared for agents.
That is a useful reality check.
We are not already living in a world of fully autonomous companies.
But the direction is visible.
The question is whether organisations can build systems in which AI capabilities produce measurable results without sacrificing accountability.
And I suspect that companies which solve this problem inside real businesses will have an advantage over companies that merely offer more AI features.
Meanwhile, Some People Are Going to Monasteries
There is a rather strange cultural development unfolding alongside all this.
While some entrepreneurs are trying to make their businesses run with less human intervention, others are becoming deeply concerned with what AI is doing to human purpose.
A recent New Yorker investigation by Sam Kriss examines MAPLE, the Monastic Academy for the Preservation of Life on Earth, a Buddhist-inspired community in Vermont.
The organisation has developed an elaborate view of AI as a profound threat to human life and society.
Its founder, Soryu Forall, argues that modern technological thinking reduces lived reality to abstract information.
The community combines intensive meditation with an ambition to influence how people working on AI understand their responsibilities.
The article also describes a much stranger set of claims.
Former members told the magazine about a secret doctrine involving collective minds, referred to as egregores, and AI-related entities described as cybregores.
Some former residents alleged that the community's practices were coercive and psychologically harmful.
MAPLE's representatives disputed significant parts of those accounts. The organisation has also published a response denying earlier allegations from a former trainee.
Those are contested claims, not established conclusions about every aspect of the community.
And it would be a mistake to treat MAPLE as representative of Buddhism or of meditation communities in general.
But the story is revealing for another reason.
It illustrates the extraordinary amount of meaning people are now attaching to AI.
For some, the technology promises wealth.
For others, the end of work.
For others, superintelligence.
For others, catastrophe.
And for a few, apparently, the emergence of a new kind of spiritual being.
The rhetoric becomes almost theological.
AI will save us.
AI will destroy us.
AI will replace humanity.
AI will finally tell us what humanity is supposed to do.
Meanwhile, somebody in an ordinary town still needs an accountant.
The Search for Something Real
I don't think the connection between technology burnout and stories like MAPLE is entirely accidental, although one community cannot establish a broader psychological trend.
AI is changing more than the economics of work.
It is changing how people think about their own usefulness.
If a machine can write software, what does it mean to be a programmer?
If an agent can perform complicated research, what does expertise mean?
If an entire business process can be automated, what should management do?
And if technical execution becomes cheap, what exactly remains valuable?
One response is to compete harder.
Use more agents.
Move faster.
Launch more products.
Become the person who can coordinate the largest amount of machine intelligence.
Another response is to seek some form of meaning outside the competition.
Meditation.
Community.
Physical work.
A more predictable business.
A life not permanently organised around the latest model release.
These are obviously not equivalent choices.
And there is an important distinction between seeking perspective and surrendering independent judgment to a charismatic authority.
But they reveal a common tension.
When the cost of execution falls, the problem of deciding what is worth doing becomes more visible.
That is true for a company.
And perhaps for an individual life as well.
The danger is treating AI as either a universal answer or a universal enemy.
It is neither.
It is a collection of technologies with increasingly powerful capabilities, embedded in institutions and markets shaped by human decisions.
The question is what we want those capabilities to accomplish.
And who gets to decide.
The Founder Who Owns the Outcome
This brings me back to traditional businesses.
A technical entrepreneur who opens an accounting firm is making a different kind of commitment from one who builds accounting software.
They are responsible for the result.
Not merely the tool.
If the accounts are wrong, the fact that an advanced model produced them does not solve the problem.
If the client is dissatisfied, a beautiful dashboard provides little comfort.
If the automation creates a compliance failure, the business has to deal with the consequences.
But the rewards are also different.
The founder owns the customer relationship.
Controls more of the operating process.
Can capture the financial benefit of improvements.
And can build a company around a real service rather than the temporary novelty of its technology.
This is not necessarily easier entrepreneurship.
It is entrepreneurship with a different source of value.
For years, the ideal founder was someone who could build software and find people willing to use it.
In the next phase, an increasingly attractive founder may be someone who understands a traditional business deeply enough to redesign how it operates.
Someone capable of asking:
What are customers really paying for?
Which parts of the process create value?
Which parts exist only because nobody has found a better way?
Where does human expertise remain essential?
And what would this company look like if we could rebuild its operating model today?
Those are not questions a coding agent can settle through software generation alone.
They require commercial judgment.
And that judgment may become one of the most valuable forms of expertise in the AI economy.
The Next Startup May Not Look Like a Startup
I don't think the end of SaaS is approaching.
I don't think traditional business ownership is automatically superior to building software.
And I certainly don't think every burned-out developer should rush out and buy an accounting firm.
Some will discover that managing employees, clients, regulation, and cash flow is substantially less relaxing than writing code.
Which would be an expensive way to learn that touching grass is available free of charge.
But I do think the economic opportunity is shifting.
Software creation is becoming more accessible.
Agentic automation is becoming more capable.
Businesses can increasingly build systems tailored to their own operations.
And the value of those systems can sometimes be captured more directly by the business using them than by a vendor trying to sell the same capability to hundreds of customers.
At the same time, the scarce resource is moving.
From writing every piece of software.
To organising reliable execution.
To making consequential decisions.
And ultimately to deciding which outcomes are worth pursuing.
The most capable agent in the world is of limited value if it is working toward the wrong objective.
The most sophisticated automation is not a competitive advantage if it makes the business more efficient at doing something customers don't want.
And the most impressive AI company is not necessarily the one that has built the most technology.
It may be the one that has used technology to build the most effective business.
I suspect we'll see more software engineers, AI developers, and technical founders moving into industries that used to seem profoundly unexciting.
Accounting.
Insurance.
Property management.
Legal services.
Specialised consulting.
Translation.
Local services.
And many other businesses where much of the work involves information, coordination, and repeatable decisions.
Some will start companies.
Some will acquire them.
Some will partner with experienced operators.
And some will discover that a business with predictable demand, real customers, and better operating margins is considerably more attractive than another AI SaaS startup competing against thirty nearly identical products.
Perhaps the most interesting thing about the AI revolution is that it could make technology entrepreneurs less interested in selling technology.
They may finally be able to own the value their software creates.
And perhaps, after automating enough of the routine work, they might even get to enjoy the occasional afternoon outside.
Without checking whether the latest agent framework has made their entire business model obsolete.
That would be a rather nice technological achievement.
Not artificial general intelligence.
Just a profitable business, a clear purpose, and a few hours in which nobody needs to open a terminal.
Sources
- McKinsey: The State of AI in 2026
- Choi & Xie: Human + AI in Accounting (2026)
- NBER: The Microstructure of AI Diffusion (2026)
- PYMNTS: Buy the Firm, Then Rebuild It With AI
- Caritas: Do AI Rollups Work?
- Deloitte: AI Agents and Enterprise Readiness (2026)
- BCG: AI at Work, Why Strategy Matters More Than Tools
- Sam Kriss: The Buddhist A.I. Sect That Wants to Take Over Silicon Valley
- MAPLE: An Initial Response to False Accusations
