Ollama, AI Coding, and the Future of Engineering Expertise

Participants: Arash, Saman, Pooria, Foad, and Sirwan

We discussed Ollama, customised AI solutions, vibe coding, the future of software engineering, and progress on our Kurdish automatic speech recognition (ASR) project.

Ollama and Running Models Locally

We started with a question about Ollama and how it works. We used Docker as a rough analogy: Ollama makes it easier to download, run, and manage open-source and open-weight AI models locally without having to configure everything manually.

We also discussed practical examples and the idea of building customised AI solutions. Saman shared a resource called Is AI Profitable Yet? during this discussion.

A Motorcycle Racing Game Built with AI

Sirwan demonstrated a browser-based motorcycle racing game he built using an Astra model.

Sirwan plans to publish a YouTube video showing the result. The focus will be on demonstrating what AI can build, rather than explaining the entire development process.

The Future of AI in Software Companies

This became the main discussion of the session. We compared companies with a Silicon Valley-style mentality, which move quickly and experiment aggressively, with more traditional companies that adopt AI more slowly.

The central question was: What happens to software engineering if AI models become increasingly capable but also much more expensive?

Sirwan: A Return to Manual Coding Seems Unlikely

Sirwan argued that companies are unlikely to return to traditional manual coding, even if AI becomes expensive. Instead, they might invest more money in models while reducing engineering team sizes.

One possible scenario is that companies retain a small number of distinguished engineers who design systems, make architectural decisions, and direct AI agents, rather than maintaining large teams of mid-level and senior developers.

Foad: Broad Technical Knowledge Might Be Enough

Foad suggested that future models might become capable enough for people with broad technical knowledge to manage AI-driven development without understanding every implementation detail.

Sirwan challenged this idea, arguing that architectural knowledge will remain essential. Someone still needs to understand system boundaries, trade-offs, scalability, and the long-term consequences of technical decisions.

Pooria: Where Will Future Distinguished Engineers Come From?

Pooria raised a problem with this scenario: if companies stop hiring and developing junior and mid-level engineers, how will they produce the next generation of distinguished engineers?

Experienced engineers will eventually retire. Without opportunities to gain practical experience, developing replacements could become increasingly difficult.

We did not reach a definitive conclusion, but agreed that this is an important unresolved question.

Kurdish ASR Project

Towards the end, we discussed progress on our Kurdish ASR project and agreed on two practical next steps:

• Recording statistics: Track the total duration of collected audio in hours, minutes, and seconds so we can measure progress.
• Community contributions: Encourage more people to contribute recordings. Family gatherings could be a useful source of natural Kurdish conversations, provided participants agree to being recorded.

Key Takeaway

The most interesting question was how the next generation of experienced software architects will develop if AI removes much of the hands-on work traditionally used to build engineering expertise.

AI might make software development much faster and reduce the number of engineers needed for implementation. However, companies could face a long-term problem if they optimise for immediate productivity while removing the career paths that create their future technical leaders.

This creates a possible paradox: the more companies depend on a small number of highly experienced engineers to guide AI, the more important it becomes to develop those engineers, even as AI reduces the opportunities to gain that experience.