The Artificial Investor - Issue 32: Is AI eating software?

My name is Aris Xenofontos and I am an investor at Seaya Ventures. This is the weekly version of the Artificial Investor that covers the top recent AI developments.

This Week’s Story: Klarna shuts down Salesforce and Workday software due to AI

We are taking advantage of a relatively uneventful week (with the exception perhaps of OpenAI’s DevDay and Tesla’s “We Robot” events) to write about the noise created in the Tech community by Klarna’s CEO comment that the company is shutting down important software, such as its CRM (Salesforce) and HR management system (Workday), and is replacing them with software developed internally using AI. A few weeks later, Sebastian Siemiatkowski backpedalled on his statement and mentioned that this is an aspiration and that, for now they are consolidating software vendors in simpler areas, such as project management.?

Nevertheless, the question remains: Is AI eating software? What is going to happen to the Software industry in the long term??

If you prefer to listen to this issue, click here for the audio version.

Powered by ElevenLabs

?? In a freefall

Software development is certainly a profession under disruption. Recent advancements in Generative AI have been a driver of better language understanding, analysis and generation, and programming languages are a great low-hanging fruit (better than the English language for instance) due to their perfect structure.?

AI-powered tools can already automate certain aspects of code generation, such as scaffolding (creating the basic structure of a software project), boilerplate code writing, and even entire modules in some cases. Also, AI-driven testing tools can automatically identify and fix bugs, decreasing testing time and costs associated with later error correction. This reduces the manual effort required from developers, thus saving time and costs.?

The cost of software development is indeed going down fast. Studies show that simple tasks can see productivity improvements of 50-60%, while tasks of moderate complexity can experience gains of 30-50%. Even more sophisticated tasks can benefit from 15 to 30% improvements (also see the most scientifically-robust study here). This is impacting the Tech job market. The WSJ published recently an article about how experienced Tech workers are struggling to find new roles and many are resorting to unusual tactics, such as putting up fliers or enrolling in college courses.

How big could the impact be on the Software Development market? The time and cost to produce a marketing article is reduced by 90% with the current Generative AI capabiltiies. There are 28 million software developers globally earning an average of 30,000 dollars annually. Hence, we are talking potentially about achieving 90% savings in an aggregate spend of 850 billion dollars. Not a small figure.

The startup and VC ecosystems have taken note. Just four companies in the space (Poolside, Codeium, Tabnine and Qodo) have collectively raised more than a billion dollars to launch and scale their coding automation solutions.

??? The rise of the citizen developer

Generative AI has the potential to enable natural language-based software development. This could empower non-IT employees (the so-called citizen developers) to build applications and tools for themselves or their team members, accelerating the low-code /no-code trend. Nearly 60% of all custom apps are already built using very little to no code. This is expected to reach 70% in the next 12-18 months as an increasing number of corporations have a citizen developer initiative.?

We have witnessed this first-hand at Seaya as we have business people (investment team members) who have developed internal automation tools (in low-code environments), which everyone is using.?

?? I did it my way

There are some important advantages in developing software in house. Large corporations typically have their own processes, data structures and ways of doing things, and, therefore, typically end up customising significantly any core software they purchase, such as the CRM. Developing solutions internally means you can make them fit like a glove to your organisation.?

In addition, sometimes software can be very close to the core of an organisation and the source of its competitive advantage, e.g. routing optimisation for ride-hailing companies. If every ride-hailing company was purchasing such software from a 3rd party, then what would be their competitive advantage? Developing this solution internally at least gives companies the opportunity to develop an edge.?

???? More than just coding

The fact that a big part of the coding task can eventually get automated, doesn’t mean that the entire job of the software engineer can get automated. Reviewing the various studies and surveys out there, a developer typically spends 30%-50% of their time in pure coding. Junior developers are on the upper end and senior developers are on the lower end.?

So, where do software engineers spend the rest of their time??

There is a number of peripheral tasks:

1. Additional coding-related tasks: This includes debugging, testing and producing documentation.
2. Understanding user needs. It involves gathering requirements through various means (interviews, surveys, user testing, etc), is subjective and relies on interpretation and judgement based on incomplete information.
3. System integration: Software typically needs to be integrated with other systems and technologies, which are often legacy. This involves understanding the architecture of the external system, which often requires complex reverse-engineering and sometimes phone conversations with 3rd party developers.?
4. Architectural decision making: Software developers need to make many decisions, from choosing the right programming language and software architecture, to deciding on the sets of features and design of user interface. These decisions are based on a variety of factors including current and future technical requirements, business needs, etc. These require human experience and strategic thinking, particularly when balancing trade-offs between competing priorities.?
5. Security strategy and management: It involves assessing the risks and making decisions on multiple layers of security, from physical security to application and network security.?
6. Planning: It involves assessing whether the project is viable with respect to time, cost and technological constraints. It also includes the identification of potential risks and establishing strategies to mitigate them.
7. Researching: This is an activity that supports all of the above. For instance, you cannot make a judgement on a security trade-off without understanding well the risks of a certain threat.?
8. Meetings and collaboration: This is also an activity that supports most of the above. Software developers don’t operate in isolation and interact with product owners, quality assurance engineers, UI/UX designers, business units, senior decision makers, etc.

AI can indeed automate a large part of the additional coding-related tasks, as well as assist with some of the other tasks, e.g. flag security vulnerabilities, automate and test system connections, or provide data-driven insights for decision-making. However, it appears that probably about 80% of the non-coding part of a software engineer’s job requires drastic changes to a Tech organisation before it can get automated.?

In conclusion, software engineering is much more complex than pure coding execution. Even if coding execution was to be automated by 90% with AI (similarly to the creation of marketing documents), this would result in 30%-35% of the overall job of the software engineer to become automated.?

?? Keep it simple

So, we have established by now that, even though AI can automate an important part of a software engineer’s tasks, the job is not going anywhere for a while.?

The fact that non-Tech companies could develop software internally cheaper than before, does not mean that they should or they would. There are additional factors to consider beyond their theoretical capabilities. These include:?

• Time to market: A 3rd party software is ready to use. Developing software internally from scratch can take months, often more than a year.?
• Distraction: As software development is not going anywhere, in-house development means that an entire Tech organisation is required within a non-Tech company (developers, testers, designers, product managers, project managers, line managers, decision makers, etc). Particularly when the application in question is not core to the business of a company, this can be a big distraction.

• Compliance: Particularly in regulated industries, such as Financial Services and Healthcare, purchasing software is not only a matter of saving time and costs; it is also a matter of passing on the pain of ensuring compliance with industry standards and regulations to 3rd parties. What non-Tech company would want to assume this risk?
• Innovation: Developing software is not a one-time task; it requires ongoing maintenance and innovation to keep the software relevant, adapting to changing business environments and technological advancements. In-house software development is often carried out in isolation and thus bears a higher risk of obsolescence.

As a result, in many cases the return on investment of internal software development would be lower than when purchasing it, even if the cost of the pure coding task was very low.

???? It’s all about the people

As software engineers are not going anywhere, the question remains: what organisations are the ones that attract the best Tech talent? First of all, startups flush with VC funding and big Tech companies can afford larger compensation packages that are difficult for non-Tech firms to match. Second, HR departments in non-Tech companies often lack the technical knowledge needed to assess candidates' skills and qualifications accurately. Third, there are some additional softer challenges, such as cultural fit (integrating Tech professionals into a culture with different work styles and expectations), use of outdated tools and processes (non-Tech companies use legacy systems and development methods making them less attractive to Tech talent who get excited by using cutting-edge tools), and fewer technical career paths and support (e.g. training).?

As a result, a non-Tech corporation would need to consider its disadvantages compared to Tech startups and large companies in terms of quality of Tech talent, which may likely impact in quality of software developed.?

?? Conclusions

Wrapping up, we do not expect the one-trillion-dollar Software market to disappear because of AI. For what it’s worth, public market investors seem to agree, as the top 10 SaaS companies are currently worth more than one trillion dollars. Nevertheless, we do believe that AI’s impact on the market will be big.?

With regards to the “buy vs. build” decision, we believe the outcome will vary depending on the specific client segment.

The most vulnerable areas are likely to be:

• Selling software to Tech-native companies, like Klarna, that have the know-how and resources to build software.
• Selling simple software, such as an online survey tool or a simple information management system.
• Selling software to large non-Tech companies that have Tech resources, particularly in areas that are core to their business and where significant customisation is required.?

The most protected areas are likely to be:

• Selling software to mid-sized and small companies that lack the resources and management bandwidth to develop solutions in house
• Selling software to traditional sectors, such as chemical manufacturers, whose DNA is not technical.?
• Selling software to regulated sectors, where the software complexity increases due to regulation and risk management.?
• Selling complex software, such as solutions that interact with multiple diverse stakeholders within a business or integrate with legacy systems. For instance, in supply chain planning, the Tech solutions are used by sales teams, marketing teams, purchasers, procurement, supply chain planners, etc.?

Beyond affecting the “buy vs. build” decision, we expect the following impact of AI on Software vendors:??

• Increase in revenues due to growing the value of their product for their clients. Software companies are already capturing some of the value AI can bring, e.g. through more sophisticated and actionable insights and end-to-end automation.
• Decrease in revenues driven by the inevitable pricing pressure coming from lower product development costs (as soon as a competitor decides to charge less, the entire market is pushed downwards). From a profitability perspective, this is offset by the lower product costs.?
• Decline in margins due to higher competition from startups taking advantage of lower entry barriers (it is getting faster and cheaper to build products)
• Higher churn for some vendors driven by clients shifting from innovation laggers to leaders. From an aggregate market perspective, this is neutral.?

So, overall, on one hand, software vendor profits are likely to be negatively impacted by i) lost sales to certain large organisations that may build more software internally in some instances, and ii) increased competition due to lower entry barriers. On the other hand, software vendor profits are likely to grow due to increased value of their solutions. It is very hard to estimate whether the net impact of the above factors will be positive or negative, but we don’t expect a significant difference.?

What we can say with certainty is that the software industry is about to enter an important transition phase.

??? Fun things to impress at the dinner table

You are your data. Someone connected the images coming from Meta’s glasses to an online human face search and managed to reveal the personal details of everyone they looked at.

Eyes on the field. The Wimbledon tennis tournament will end a tradition of 147 years of using human line judges. They will be replaced by AI in 2025’s tournament for the first time.

See you next week for more AI insights.