Software engineering with LLMs in 2025: reality check

Summary

The video discusses the growing impact of AI tools, particularly Large Language Models (LLMs), on the software development industry. Experts highlight their potential to revolutionize coding, offering substantial productivity boosts for individuals, but there’s still uncertainty around their organizational adoption. While CEOs and founders are enthusiastic, many engineers are less convinced, citing modest time-saving benefits. The video explores how AI is reshaping software development, with some comparing it to past technological shifts, suggesting that experimentation and lower costs may lead to exciting new innovations in the field.

Takeaways

• 😀 CEOs and founders in AI-related companies are highly enthusiastic about AI tools, often more so than engineers.
• 😀 Experienced engineers are seeing success with AI tools, particularly in enhancing their productivity and innovation.
• 😀 Large language models (LLMs) are being used to create low-code tools and have the potential to revolutionize software development.
• 😀 AI usage in software development is growing steadily, with 50% of developers using AI tools weekly.
• 😀 While some companies experience significant benefits from AI tools, the actual time saved may be smaller than some high estimates suggest.
• 😀 CEOs and founders are often more excited about AI due to the potential financial impact, while engineers are more focused on the practicalities of its integration.
• 😀 LLMs are not just a new layer of abstraction but could represent a lateral shift across the entire stack, offering new ways of working.
• 😀 AI tools are currently better suited for individual developers than for large teams or organizations.
• 😀 The shift from assembly to high-level languages in the past is similar to the productivity boost that LLMs could offer in the future.
• 😀 Veteran software engineers like Kent Beck are finding new excitement in programming due to the possibilities enabled by LLMs and AI.
• 😀 The landscape of software development is changing, and experimentation is essential to understand what works and what doesn’t, similar to past technological shifts.

Q & A

Why do engineers seem less enthusiastic about AI tools compared to founders and CEOs?

Engineers may be more cautious because, while they recognize the potential of AI tools, they are also more critical of their current limitations and are often dealing with the practical challenges of integrating these tools into existing systems. Founders and CEOs, on the other hand, tend to focus on the business and financial opportunities, which may explain their higher enthusiasm for AI's potential.

What makes LLMs different from previous technology shifts like high-level programming languages?

LLMs differ from past technological shifts because they are non-deterministic, meaning they can generate multiple outputs for the same input, making them more flexible but also less predictable. This introduces new challenges in terms of reliability and control compared to deterministic systems like traditional compilers or high-level programming languages.

How much time do developers typically save by using AI tools, according to surveys?

Surveys show that developers save about 3 to 5 hours a week on average by using AI tools, though some engineers, like Peter, report higher productivity boosts of 10-20 times. However, the actual time savings may vary depending on the developer's work and the specific tools being used.

Why do AI tools work better for individual developers rather than teams?

AI tools tend to be more effective for individual developers because they are designed to assist with specific tasks and are easier to integrate into a personal workflow. In contrast, using these tools at the team or organizational level often faces challenges related to coordination, communication, and integration into larger systems.

What does Simon Willis believe about the current state of AI development tools?

Simon Willis believes that coding agents and other AI tools have reached a tipping point in the last six months, with significant improvements making them genuinely useful for development. These tools are now capable of handling more complex tasks that were previously challenging, marking a key shift in AI's role in software development.

What is the significance of Kent Beck's statement about having more fun programming after 52 years?

Kent Beck's statement highlights the transformative impact of AI tools, which have made programming more exciting and less tedious. He finds the flexibility and capabilities of AI tools inspiring, allowing him to tackle ambitious projects like building a small talk server. This reflects how AI is rekindling passion for developers who have experienced the evolution of technology over decades.

What key insight did Martin Fowler offer about the future of AI in software development?

Martin Fowler compared the advent of LLMs to the shift from assembly language to high-level programming languages. He suggests that LLMs could provide a significant productivity boost, similar to past advancements, but with the added complexity of being non-deterministic, which will shape how developers approach software design.

How are large tech companies and startups engaging with AI tools?

Large tech companies are making heavy investments in AI, with growing usage across their operations. Startups are also experimenting with AI tools, but their adoption rates can be hit or miss depending on the company's focus and size. Overall, AI usage is more prevalent in certain sectors like AI-related startups and large enterprises.

What does Brigita from Thoughtworks believe about the role of LLMs in software development?

Brigita believes that LLMs are a versatile tool that can operate at multiple abstraction levels, from low-code assembly languages to high-level human languages. She sees LLMs as a lateral move across the tech stack, enabling new ways to interact with technology and pushing the boundaries of how software is developed.

What is the main takeaway from the speaker's discussion on AI tools in development?

The main takeaway is that while AI tools are showing promising potential in software development, there are still several unresolved questions about their impact on productivity, their application at the organizational level, and how to effectively use the time saved. Experimentation and continuous learning are key as the industry navigates these challenges.