I have been thinking deeply about what AI is doing to software development, especially from where I am building in Zimbabwe. And I have come to an uncomfortable conclusion.

For decades, building serious software required an organisation: developers, designers, testers, database engineers, project managers, DevOps engineers, business analysts and support staff. Then came the meetings to coordinate the meetings, Jira tickets explaining other Jira tickets, and conversations about why somebody’s branch broke somebody else’s branch.

Then artificial intelligence arrived, and the software industry made a dangerous assumption: AI would simply make software companies more productive.

Maybe. But AI may do something considerably more brutal: make a large percentage of the traditional company structure surrounding software unnecessary.

The uncomfortable mathematics

Imagine two competitors.

Company A

Eighteen employees. Every month it must generate enough revenue to cover salaries, management, office expenses, subscriptions, administration, recruitment, meetings, coordination and the inevitable cost of people waiting for other people.

Company B

One technically capable founder, one laptop, cloud infrastructure, AI coding agents, research, design assistance, automated testing and deployment, documentation, customer analysis and AI-assisted support.

Most importantly, Company B has one human being who understands the system well enough to decide what gets built—and what absolutely does not.

Company A has more humans. Company B may have more productive intelligence per dollar. That should terrify mediocre software companies.

Your 20 developers are not automatically an advantage

There is an uncomfortable obsession with headcount in technology. “We have 40 engineers.” Wonderful. What are they producing?

Headcount once functioned as a rough proxy for productive capacity because complicated software required many specialised humans. AI is weakening that relationship. One excellent engineer equipped with capable agents can already perform work that previously crossed several disciplines.

That does not mean one AI user equals 40 excellent engineers. It means something more disruptive:

The minimum number of humans required to create a competitive software organisation is collapsing.

A five-person company cannot simply celebrate because AI makes each employee twice as productive. It must ask what happens when its competitor no longer needs five people.

Then comes the AI coordination problem

Companies will respond by giving AI to everybody. Twenty developers become twenty developers using twenty coding agents. Management announces that the AI transformation is complete.

Except now the organisation may have forty sources of output. One agent makes an architectural decision; another makes a different one. Someone approves generated code because the tests are green without understanding it. Confidential information enters the wrong model. A dependency changes. Thousands of lines appear that nobody particularly wanted. Documentation explains code another agent already replaced.

Then everyone attends a meeting to coordinate the AI that was supposed to eliminate coordination.

Congratulations. You automated the chaos.

AI does not magically repair a badly organised company. It can amplify it. A confused organisation with AI can become a much faster confused organisation.

The rise of the one-human software company

Now imagine a different architecture: one founder—not necessarily one worker forever, but one coherent source of product authority—surrounded by an artificial organisation.

  1. Research agent
  2. Product agent
  3. Architecture agent
  4. Coding agent
  5. Testing agent
  6. Security agent
  7. Deployment agent
  8. Documentation agent
  9. Customer intelligence agent

They do not need salaries or parking spaces. They do not care about titles. They can work at 02:37. When a better model appears, the founder does not conduct six months of restructuring. The infrastructure changes.

The human stops being the person who personally performs every task. The human becomes the governor of a machine that performs work. That distinction is enormous.

The part AI enthusiasts do not want to hear

Buying ChatGPT does not make you a software company. Neither does installing Claude, Gemini, Codex or whichever agent is fashionable this month.

If you cannot explain the architecture your AI created, you do not own a technology company. You own a mystery. If you cannot verify the code, you are gambling. If nobody knows which information is authoritative, you have automated confusion. If agents have unrestricted credentials, you have built an automated security incident.

If you cannot recover when the AI makes a catastrophic mistake, you have not created autonomy. You have created dependency.

The winning solo software company will not be the person who prompts fastest. It will be the person who builds the strongest AI operating system around themselves:

  • clear authority and approval gates;
  • durable knowledge and version control;
  • automated testing and observability;
  • security boundaries and backups;
  • customer evidence and repeatable deployment;
  • the ability to understand what the machines are doing.
The future belongs less to the best prompter than to the best orchestrator.

Zimbabwe should pay attention

This transformation could become especially interesting in Zimbabwe. Historically, building a serious local technology company was expensive. Skilled developers and capital are scarce. Customers are price-sensitive. Infrastructure can be unreliable. Small businesses cannot afford enterprise software prices.

AI attacks that equation. A technically competent Zimbabwean with a laptop, internet connection, cloud infrastructure and sophisticated AI workflows can potentially build products that would previously have required a funded team.

That founder has another advantage multinational software companies frequently lack: proximity to reality.

I can walk into the shop, watch the cashier, talk to the owner, see the handwritten debtor book, watch the internet disappear, see employees sharing computers, discover that the business operates in several currencies and notice that WhatsApp is effectively part of the company’s information system. Then I can go home and change the software.

That feedback loop can be brutally fast.

Combine frontier intelligence with local ground truth. Silicon Valley has the models. We have the problems standing directly in front of us.

Traditional companies still have one enormous advantage

Established companies possess things solo founders frequently do not: institutional knowledge, customer relationships, specialists, capital, reputation, legal capacity, security expertise, 24-hour support, infrastructure experience and multiple humans capable of challenging bad decisions.

Those advantages are real. AI does not abolish organisations. It abolishes the assumption that organisation size equals capability.

The companies most at risk are not the excellent ones. Excellent organisations will use AI to become terrifyingly productive. The vulnerable companies are the comfortable middle: too expensive to compete with AI-native independents, too slow to compete with startups, too mediocre to justify enterprise pricing, too bureaucratic to change quickly and too proud of their headcount to notice it has become a liability.

The coming software-company diet

Over the next several years, I expect something resembling corporate weight loss. Work that once required dozens of developers, testers, designers, analysts, project managers, DevOps engineers and managers may be delivered by a dramatically smaller group.

Not everywhere. Not immediately. Not without risk. But enough to alter market economics.

The result may not be millions of permanent one-person companies. We may see something more interesting: elastic organisations.

One founder sits at the centre. AI handles enormous amounts of routine intellectual labour. Specialist humans enter when their judgment creates genuine value: an accountant for accounting expertise, a lawyer for legal interpretation, a security engineer for serious security work, a designer for exceptional visual judgment, a domain specialist for industry knowledge and additional engineers when scale genuinely requires them.

Then the organisation contracts again. Instead of hiring another human simply because work exists, companies increasingly ask: why does this particular work require another human?

Software companies have been warned

If you own a software company today, do not fire everyone tomorrow because you read an aggressive article on the internet. That would be stupid. Do something harder.

  • Examine every part of the organisation.
  • Ask what AI can perform, what it can assist and what absolutely requires human judgment.
  • Find where humans merely transport information between other humans.
  • Ask why deployment takes three days and why six people attend a meeting about a feature one engineer could implement.
  • Check whether developers understand the AI-generated code they approve.
  • Decide whether company knowledge belongs to the company—or lives inside employees’ heads and chat histories.
What happens if a brilliant competitor can reproduce 70% of your productive capability for 10% of your operating cost?

Prepare, because somewhere somebody is trying. Maybe in San Francisco, Bangalore, Nairobi or Harare. Maybe in a small room in Chiredzi with a laptop, an internet connection and absolutely no respect for how software companies are traditionally supposed to operate.

The real revolution

AI will not kill software engineering. It will make good engineering more important. It will not eliminate collaboration. It will make unnecessary coordination increasingly expensive. It will not make humans irrelevant. It will make mediocre human contribution harder to hide behind organisational structure.

AI will not guarantee the victory of the solo founder. Most solo AI builders will produce mountains of garbage. But the small minority who combine technical understanding, domain knowledge, disciplined AI orchestration, relentless verification and direct access to customers may build companies with economics we have rarely seen before.

The old equation

More people = more capability

The emerging equation

Human judgment × machine capability × system quality = productive power

That multiplication sign matters. A weak founder with powerful AI can produce rubbish faster. A bureaucratic company with powerful AI can bureaucratise faster. But an exceptional human operating a disciplined artificial workforce? That is something different.

So, software companies: keep the ping-pong table, the motivational posters and the headcount announcements. Just understand that somewhere your next competitor may be holding his entire engineering department in one laptop.

And he does not need your permission to enter the market.

Prepare now.