AI, consistency, and evidence-led ideas

We had our third Kurd Software session today with Foad and Saman. The community had been inactive for quite a long time, so it was great to see the sessions gradually starting again over the past few weeks.

The discussion focused mainly on two connected topics: the impact of artificial intelligence on software development and the importance of focus, consistency, and steady progress in life.

Foad began by explaining how AI has made it much easier for him to learn new subjects, experiment with ideas, and build small projects. He shared examples of things he has recently explored, including a chess-related project that helps him learn and improve his game. His main point was that AI can significantly reduce the difficulty of getting started. Instead of spending a long time searching for resources or trying to understand everything alone, people can use AI as a learning partner and move much faster.

During the conversation, Saman asked Foad how he manages to stay focused and consistent while working on so many different interests. This led us to discuss one of Foad’s long-standing principles: slow but continuous progress.

I mentioned that Foad has always been a role model for me when it comes to consistency. We used to discuss this frequently when I was living in Iran, and he has always encouraged the idea that meaningful progress does not need to be fast. What matters is continuing to move, even through small steps, and trying to enjoy the learning process rather than constantly rushing towards an outcome.

The discussion then moved towards programming and the future of software engineering. Foad mentioned that people often ask him how they should start learning programming today, and he asked Saman and me for our opinions.

Saman had a relatively optimistic view, while I took a more cautious and pessimistic position. I argued that AI is becoming capable of handling more and more of the work that previously required experienced software engineers. Skills such as writing clean code, following design patterns, building good abstractions, and maintaining software quality still matter, but their value may gradually change as AI systems become better at generating, reviewing, testing, and improving code.

I also raised the possibility that many of the new AI-related engineering skills we are currently learning—such as agent orchestration, harness engineering, context management, and evaluation loops—may eventually be absorbed into the platforms created by companies such as OpenAI and Anthropic. Developers may currently build the systems around AI agents, but model providers could later automate those layers as well.

This creates an uncomfortable question: even when engineers continuously follow the latest technologies, will the abstraction layer keep moving faster than they can adapt?

We also discussed the future of SaaS. I argued that traditional SaaS products, particularly small and simple subscription applications, may face serious pressure. As AI makes software creation easier, individuals and small teams may decide to build personalised tools for themselves instead of paying recurring subscriptions for generic products.

At the same time, large enterprises will probably continue adopting AI more slowly because they need to deal with security, compliance, integration, governance, and organisational complexity. Software consultancies may also increasingly operate like AI-powered software factories, rapidly building customised solutions for clients.

Towards the end of the session, Saman introduced an interesting project he has been developing with Hermes for evaluating and testing business ideas. The system is designed to move ideas away from vague enthusiasm and towards structured evidence and real-world validation.

Instead of simply saying, “This sounds like a good idea,” the system asks:

• Who experiences the problem?
• How frequently does it happen?
• Who is the user, champion, and buyer?
• What is the smallest valuable starting point?
• What evidence would support the idea?
• What evidence would prove it wrong?
• What is the cheapest real-world experiment we can run?

Each idea is documented through structured files covering the problem, assumptions, risks, evidence, experiments, decisions, target customers, and outreach plans. The final outcome for an idea can be to advance it, hold it, pivot it, or kill it.

Overall, the session brought together three important themes: AI as a powerful learning accelerator, consistency as a long-term advantage, and the need to test ideas through evidence rather than excitement.

The central tension was that AI makes it easier than ever to learn and create, while also making it less clear which technical skills will remain valuable in the future. We did not reach a definitive answer, but it was a thoughtful and enjoyable discussion.