The next billion-dollar software company could look nothing like today's tech giants

AI has already transformed how software is written. Developers now use AI to generate code, explain complex systems and accelerate everyday tasks. But this is only the beginning. The next stage of AI isn’t about helping engineers write software faster; it’s about enabling software to build itself.

 

We’re entering the era of the Dark Factory: an AI-powered software factory capable of designing, building, testing, deploying and continuously improving applications with minimal human intervention. The name borrows from lights-out manufacturing, where fully automated production facilities can literally operate with the lights off because no people are needed on the factory floor. The same principle is beginning to emerge in software development. 

 

Of course, the responsibility for outcomes, decisions and the impact of software development and delivery principally sits with people, which puts more value than ever on strong developer talent and expertise. But the reality of AI in software is set to have implications far beyond day-to-day operations. It will fundamentally reshape how software companies grow, compete and create value. 

 

From instructions to intent

For decades, scaling a software business followed a predictable formula. As demand increased, companies hired more engineers. More products required more developers, larger QA teams and growing operations. Headcount became closely linked to output. 

 

Today, teams are becoming much more hybrid in structure, with AI taking on more work as company output increases.

 

AI tools can already generate code, create documentation and automate testing. More advanced agent-based systems are beginning to coordinate these activities, with multiple AI agents working together across the software development lifecycle. But the real shift comes when AI moves beyond following instructions and starts optimising for outcomes, when we start building ‘outcome machines’ at scale.

 

Instead of providing a detailed technical specification, organisations will increasingly define the result they want to achieve. AI systems can then determine the architecture, select the appropriate technologies, generate the implementation and continuously refine the solution based on production feedback.

 

In other words, software development becomes less about prescribing every step of the journey and more about defining the destination.

 

Redefining how software companies scale

This changes what it means to build a successful technology company.

 

The next billion-dollar software business may not be a huge operation with hundreds and thousands of engineers. Instead, relatively small teams could orchestrate highly autonomous AI systems capable of delivering work that previously only much larger enterprises could resource. Competitive advantage will shift away from building the largest engineering organisation towards being able to embed AI to the right level to help them to scale instead. The organisations that can do this, regardless of size, will be able to make better strategic decisions, define the right objectives and bring products to market faster than competitors.

 

However, this doesn’t mean engineers become obsolete. Far from it. As software construction becomes increasingly automated, the role of the engineer becomes more valuable in different ways. Instead of spending hours writing repetitive implementation code, developers will focus on designing resilient systems, defining architecture, orchestrating AI agents and validating AI-generated outputs. The future of engineering is something like game theory: the ability to facilitate great outcomes from carefully decided initial conditions, the ability to negotiate well with hard-to-understand things.

 

Software engineering is evolving from manually writing every line of code to directing intelligent systems that can produce it. The technical expertise doesn’t disappear; it becomes the foundation that enables AI to operate effectively.

 

Dark Factories still need human governance

While engineers will spend less time on implementation, organisations themselves cannot step away from responsibility. A common misconception is that autonomous software development removes people from the equation altogether. It doesn’t.

 

AI may generate code and execute tasks at unprecedented speed, but humans remain legally and ethically accountable for defining business objectives, setting guardrails, managing risk and deciding whether AI-produced outcomes should be trusted. Oversight isn’t about checking every line of code, it is about ensuring autonomous systems are operating within the right technical, commercial and ethical boundaries.

The companies that succeed won’t be those that automate everything. They’ll be the ones that understand where autonomy accelerates delivery and where human judgement remains essential.

 

Safety cannot become an afterthought

Dark factories come with a different class of risk than conventional forms of automation. A truly dark factory, something whose core is completely or almost completely composed of autonomous agents, is an unbelievably difficult thing to keep track of. Already, the industry is seeing not only a huge glut of code coursing through their systems, but that code is becoming more difficult to understand. We’ve seen autonomous agent swarms craft code that is truly superhuman, leveraging all kinds of creative techniques to squeeze more and more efficiency out of a system. So the question becomes really clear – how do you actually manage something you don’t completely understand? 

 

The answer to this question is to really laser-focus on the interface between your infrastructure and the world. This means being incredibly intentional about what your agents have access to and the kinds of things they can do. Theorist Jay Springett calls this “building a hard world,” where permissions and possibilities are supremely tightened up. 

 

This new class of safety risk also means an extremely strong focus on production, not just in terms of observability (which can only get you so far when working with things that are superhuman) but actually in terms of control. Guardrails and killswitches and re-routing mechanisms should be built into every runtime environment that takes place within production, with the intention of gathering total, live control over your entire agentic surface area. This is a non-negotiable, in most cases.

 

The evolution of software development

In many ways, the Dark Factory represents the next evolution of software development.

 

Today’s AI tools still resemble highly capable assistants. They generate code, answer questions and automate individual tasks. Tomorrow’s systems will increasingly coordinate entire workflows, making design decisions, allocating resources and continuously improving software against measurable business outcomes.

 

The companies that thrive won’t necessarily be those with the largest engineering teams or the biggest AI budgets. They’ll be the organisations that become best at orchestrating autonomous systems while maintaining the governance, resilience and human judgement needed to keep them aligned with business goals.

 

---

 

Marek Poliks is Head of AI at LaunchDarkly

 

Main image courtesy of iStockPhoto.com and pressureUA