AI will result in much more software. What this will mean for companies.
AI is going to result in much more software. From working through the implications, we think that fears that software proliferation because of AI will be bad for most companies are wrong – our conclusion is that it will be very positive for the average company in terms of profits and valuation. What is needed though is for almost all existing companies to reassess the new “rules of the game” as this effect on software will change almost every industry – and adapt their strategy to continue creating value in this new, more productive world.
In this short note, we first think about software and its general impact on business (looking at the last 30 years), we then think about how AI will dramatically increase the amount of software in the world over the next five years, we then think about the effects that all this extra software will have on the world and on companies, and finally we think about what companies need to do now in response to this fundamental change.
Thinking about software and its general impact on business (looking at the last 30 years)
Software has changed the world over the last 30 years.
It has changed how people work, how humans communicate with each other, how democracies work, and the policies governments implement.
This has had big implications for companies. Many of the world’s biggest companies 30 years ago no longer exist, and many of the world’s biggest companies now did not exist 30 years ago.
Software has increased standards of living. It has driven GDP increases in developed and developing markets – and has created a lot of value that is not reflected in GDP (things like the benefit of having a smartphone or fewer deaths from healthcare improvements are not fully reflected in GDP measures).
The big constraint has been software developers - even simple software has been expensive to create - because there are a limited number of people who can write code. Software developers have been among the best-paid professionals in society because of this high value that their software creates.
This has meant that we have had far less software than we would ideally have if we were not limited by the number of software developers that we have.
How AI will dramatically increase the amount of software in the world over the next five years
AI has dramatically reduced this societal constraint – it has reduced the number of software developers we need to write each piece of code.
AI can already produce code. Companies like Google say that over 25% of their code is now written by AI.
At the moment though, AI-produced code is still highly imperfect - it is still full of errors. But this is improving fast. These improvements are partly coming because LLMs are improving, but they are also coming because of other software that is being created to work with LLMs to write code together. These include software tools which get LLMs to write lots of possible versions of code for a prompt - and then run all of those versions in a “walled-off computer for testing” to see which ones work, which ones work fastest, and which ones can’t be hacked.
As LLMs improve and as these bigger systems that work with LLMs to create code get developed, we will be able to produce much more code.
This is not a new phenomenon for us - it is similar to the entire history of programming. We went from very basic programming languages like punch card patterns, to assembly languages, to early high-level languages like FORTRAN - all the way to modern high-level languages supported with large numbers of frameworks and libraries. The outcome of these improvements in programming languages and the outcome of AI-written code generation is likely to be the same – more software created per programmer. Today, one programmer can maybe produce as much software (in terms of what it can do) as perhaps 100 programmers 30 years ago because of programming language advances and advances in the computing environment in which their programs run. AI-generated code means that today, one programmer can produce as much software as perhaps four programmers a year ago. In five years from now, one programmer might be able to produce as much software as 20 programmers today.
This is not likely to mean that in five years, we are creating a similar amount of software to today – just employing far fewer developers.
We expect that the result will be that we have much more software produced. We expect that the total number of software developers will also significantly increase - as we have seen historically as output per programmer increases. The nature of what a “software developer” is will change - as it has over the last 30 years. The average developer will be working at a higher level of abstraction than developers today. There is a chance that we are wrong, and that AI systems will actually be able to fully design and create all software alone in this time. But this seems unlikely - at least in the next decade. In this scenario, the rate of software growth would be even faster.
First-order, second-order, and higher-order effects
We can break down the effects we expect into first-, second-, and higher-order effects.
First-order effects
We expect the first-order effects to be that much more software produced. This will be both custom software produced in-house and third-party software. In-house software will replace some software that companies currently buy. Large software companies will expand the features of their software - creating much more value for buyers. Some software services will be protected by non-software assets attached to them - like the data of Bloomberg, or the network effects of Microsoft’s Office file formats. New software companies will be created that solve new, often more specific problems. Existing large software firms that adapt will often have competitive advantages that help them continue to succeed in this new environment. New firms will replace some existing large software firms that do not adapt well to this new paradigm.
Second-order effects
More software will mean more output. Standards of living will increase. The value of companies’ assets (customer bases, brands, assets, etc.) will rise in general. Some industries will get highly disrupted - but in general for companies that are proactive in adapting, we expect the future to be increasingly profitable.
Higher-order effects
More output will increase standards of living - in developed and developing countries. This will affect social, geopolitical, political, and regulatory systems. There are a lot of wildcard outcomes that could occur because of how powerful software can be - including scientific breakthroughs in energy and longevity.
What companies need to do now in response to this fundamental change
The key takeaway for all companies is to adapt. The future should be better on average for companies - but the rules of the game have changed so companies need to adapt to those new rules. Companies need to look at their products - and work out if the effects of much more software in the world will mean they need to change their products. Companies also need to look at their operations - to work out how to make best use of the new capabilities available to companies. This includes finding better ways of getting new customers and delivering value to customers. They also need to find new ways to increase efficiencies that are unlocked by this new lower-software development price structure (and new capabilities of software) - so they can stay competitive on price.