I recently sat down with Daniel Sogl, an AI specialist for software developers who has presented on this subject at multiple conferences. Our conversation covered how AI will shape our work in the near, medium, and distant future.
Which AI tools are you currently using?
GitHub Copilot leads the pack by a wide margin, mostly due to its seamless integration in corporate settings. It works across nearly every IDE, though the available features differ. I also rely on Cloud Code, which operates through the command line. Then there's ChatGPT – probably the most widely accessible option. My grandfather, who is over 70, now writes Python with it. There's a whole ecosystem of tools like Cursor, Aider, and Claude, but many of them have started to converge. Features get copied quickly between tools, so the real differences are often quite small.
You've got practical experience using AI with Angular. What stands out?
Angular presents a unique challenge because it has evolved dramatically in recent years. React, by comparison, has stayed relatively consistent. LLMs like ChatGPT were trained on publicly available code, and much of that code is stale. When you request Angular components, you frequently end up with an odd blend of old and new syntax. You might receive a standalone component that still references *ngFor. This is a common frustration among Angular developers. Fortunately, there are ways around this – for instance, instruction files that bring the LLM up to speed.
Can you share an example where AI genuinely saved you time?
Absolutely, the NGRX Signal Store comes to mind. I built an instruction file that documents how we organize our stores. Once that's in place, I can ask Copilot to generate a new store, and it's correct roughly 90% of the time. AI is also handy when I switch between client projects. I use it to grasp the domain or codebase quickly – this is incredibly useful for debugging and getting onboarded.
How do these instruction files actually function?
It varies by tool. In Copilot, it's .instructions.md, in Cursor it's .cursor-rules, and in Cloud Code it's .cloud.md. In essence, they're markdown files containing coding rules that get silently appended to every prompt. This lets you guide the LLM toward the right Angular APIs and ensure it follows your project's conventions.
How frequently do you integrate AI into your Angular workflow?
I use it on a regular basis – for writing stores, services, unit tests, and basic UI components. It shines especially on repetitive tasks. For more involved debugging or sophisticated UI work, I usually handle it manually, though I still tap AI for brainstorming or to explore possible strategies.
Have you ever gotten trapped in a loop where AI keeps missing the mark?
All the time. That's what's known as "Wipe Coding" – endlessly prompting "please fix" without making headway. Sometimes the best move is to close the tool and troubleshoot on your own. Developing that tolerance for frustration is crucial, particularly for junior developers.
Any advice for using AI during debugging?
Yes – MCP servers are an excellent way to provide richer context. These are tools that let your LLM connect to external systems like GitHub, Confluence, or even logs. An LLM can examine a GitHub issue, fetch logs, and consult documentation to figure out why an acceptance criterion isn't being satisfied. That can save a significant amount of time.
GitHub claims a 55% boost in productivity. Does that match your experience?
Not exactly. We spend only about one-third of our day actually writing code. So even if AI cuts 30% off that time, it translates to just a 10% improvement across the entire workday. That's still worthwhile, especially since it lightens the cognitive burden. But we need to keep our expectations grounded.
What about junior developers? How does AI impact them?
There are two sides to this. First, we still need developers who comprehend what AI-generated code is doing. Even if we're not writing every line ourselves, we're responsible for reviewing and maintaining it. Second, AI can serve as a learning aid – kind of like a personal mentor that answers questions. But leaning on it too hard can breed complacency. Developers must continue honing their problem-solving abilities.
How does AI affect code quality?
Instruction files are helpful here as well. You can embed your linting and formatting standards directly into them. This raises the quality bar for generated code. However, LLMs are still prone to duplicate logic or overcomplicate things. That's why static code analysis remains a must-have.
When something breaks, do you refactor by hand or prompt AI to take care of it?
It depends on the situation. If fixing it manually is faster, I just do that. If it's something that keeps coming up, I'll update the instruction file. You have to weigh which approach is quicker – prompting or typing.
Have you come across AI-generated bugs that looked fine on the surface?
Absolutely – it happens constantly. The AI says "done," but then ng serve fails. LLMs don't verify their own output unless you explicitly instruct them to run a linter or a build. So testing stays essential.
Looking forward: How should we get ready for autonomous AI development?
That's a big question. Autonomous AI agents are already being introduced. GitHub Copilot, for instance, can now resolve issues on its own. For now, this works best with small jobs like updating translations. More complex changes still fall short. But the direction is unmistakable: AI will handle more and more tasks. Developers will shift into positions like technical supervisors – overseeing architecture and reviewing AI-generated code.
How can I convince a skeptical dev team to embrace AI?
Be open about what AI can and can't do. Set achievable expectations. Don't promise 50% productivity gains within a month. Look for repetitive tasks where AI can add value – like generating stores or refreshing Swagger docs. Above all: developers remain in control. It's a co-pilot, not an autopilot.
You offer a workshop too. What can attendees look forward to?
It's a two-day session focused on understanding what tools exist and how to use them effectively – GitHub Copilot in particular. We cover instruction files, prompt engineering, MCP servers, and even building your own MCP server. The aim is to equip developers with the skills and know-how to apply this in their own development process.
Workshop Details: AI for Developer Productivity
What's the bare minimum every Angular developer should know about AI?
How LLMs work and how they're trained. That's what explains why they occasionally deliver outdated or off-target results. The specific tool matters less – grasping the underlying principles is essential.
What's one action our readers should take immediately after finishing this interview?
Two things, really:
- Look into Custom Instructions – they dramatically boost output quality.
- Explore MCP Servers – they elevate AI tooling by providing LLMs with access to contextual information.
Further Reading
Be sure to check out our workshop covering this subject: