In the dynamic software product development world, one innovation stands out, transforming industries at an unprecedented pace: Artificial Intelligence. With the ability to analyze vast amounts of data and learn from patterns, AI has become one of the most notable driving forces behind the development of cutting-edge software solutions.
In our recent article, we dove into the impact of AI on the future of work in recruitment, process automation, human-AI collaboration, and the creation of new jobs. The Recursive now further explores the implications of AI in the domain of software product development, the transformative potential, and various ways AI can revolutionize the process of creating and refining software products.
Insights were shared by:
- Martin Dostál: Chief Science Officer & Partner at AI Startup Incubator, Member of Investment Committee at Look AI Ventures
- Bartłomiej Poniecki-Klotz: AI/ML Field Software Engineer at Canonical, Solution Architect
What are some key benefits startups can gain from using AI in their product development?
By embracing the power of AI in software product development, developers can revolutionize their approach, allowing them to stay competitive in a rapidly evolving market. Martin Dostál has commented on several of the benefits for The Recursive below.
Automation and efficiency
Martin Dostál: Startups typically have small to medium size development teams. For them, the automation and optimization of software development help develop software more rapidly, which is critical for highly innovative software companies. Artificial intelligence speeds up development and reduces the needed human power, especially in repetitive tasks. The latest contribution is represented by LLMs (Large Language Models) used in, for example, ChatGPT or similar products. These models can be applied not only to natural language content but also to artificial languages such as programming languages.
LLMs help to automate the software development process where AI can write a certain, but not insignificant, amount of code. However, these language models can be greatly complemented by AI techniques and models that can model and understand a program code’s semantics (meaning).
Enhanced quality and bug detection
Martin Dostál: AI can also help improve software quality in terms of code, user interface, or performance optimization. We have a company in our Look AI Ventures portfolio, OpenRefactory, which focuses on detecting and fixing security-related bugs in software code. By combining large language models and their unique technology related to bug detection, the solution can detect bugs with much higher accuracy than the existing tools.
What potential issues can startups face when integrating AI into their software product development process?
While AI brings tremendous potential to software product development, it also presents challenges that must be understood and addressed. Bartłomiej Poniecki-Klotz shared with The Recursive some of the main challenges arising from the complexity and evolving nature of AI technologies.
Reliance on external providers
Bartłomiej Poniecki-Klotz: One of the issues is reliance on external providers for the core part of the business. If a big part of your business is not controlled by you, it can lead to many unpleasant surprises. This is a risk tradeoff for speed of innovation.
Data privacy and compliance
Bartłomiej Poniecki-Klotz: Data privacy presents a potential problem as you cannot use external APIs (Application Programming Interfaces) with PII (Personal Identifiable Information) data. For example, in countries like Switzerland, customer data cannot leave the country’s borders. When using APIs, you usually do not have certainty.
Bartłomiej Poniecki-Klotz: Another issue is related to the cost. One of the most frequently used APIs by startups is OpenAI. While you only pay for what you use, not keeping the costs in check can kill any newly founded startup, and API usage costs can grow massively overnight.
How can startups integrate AI into their software product development process?
Using AI in software product development offers startups multiple opportunities to enhance the process, including streamlining the product design process and improving the efficiency and accuracy of testing and validation.
Bartłomiej Poniecki-Klotz: Outside of integrating AI into the products, there are multiple ways to use it in the product lifecycle. AI can be used on a strategic level to support the business creation process, decision-making, or creating step-by-step instructions. The purpose is to shorten the time needed for research and quickly get access to aggregated information.
On the opposite side of the spectrum, AI helps with work on the task level, where it creates new logos, writes short and long texts, or transforms existing content into a more engaging one. However, startups need to be careful when using AI, as it may duplicate or copy content from someone else.
Another interesting usage of AI is code generation for POC (Proof of Concept) or first MVP (Minimum Viable Product). Projects that took months before are now prototyped in minutes with the hands of an experienced engineer and AI. As a result, using AI in the product design stage allows startups to innovate faster than ever.
“AI is a powerful tool but only a tool in the end,” says Bartłomiej Poniecki-Klotz.
Testing and validation
Martin Dostál: AI disruption in software development is already here, and we can see it in many areas, including coding, testing, bug detection, UI development, and usability monitoring.
These days, we have many products already powered by AI. These AI models, or more specifically, Machine Learning models, are typically developed and tested in laboratory environments. There is a huge opportunity and necessity to validate and test AI products that are themselves powered by AI techniques, such as deep learning models for monitoring or anomaly detection algorithms, in real operations. The validation and testing process helps collect feedback on how your AI is performing and in which use-cases and environments there is happening something that is not expected from the existing model or AI. Typically a poor performance, to give an example,” says Martin Dostál.
Realize how great such tools can be. You can detect an issue in your model, while this is something you would not be able to reveal in laboratory testing and validation. You can detect the issue sooner than you start losing customers because your product is not working as expected.
It is important to understand that Machine Learning models cannot be typically exhaustively tested in advance in laboratory environments, so real field monitoring is a critical piece of the software toolchain.
“We are just at the beginning of the transformation of the software development processes,” says Martin Dostál.