My name is Serhii, and I’m an AI Automation Architect focusing on building agents for real-world business operations (logistics, construction, customer services).
I work with OpenAI Actions, GPT Vision, n8n workflows and Realtime assistants to create practical tools that automate communication, estimation, routing and daily tasks for small businesses.
I joined the forum to learn from the community, share my projects and better understand how to structure agent-based systems the right way. My goal is to grow into a strong contributor and participate in advanced testing of new OpenAI capabilities.
If anyone here is working with Actions + automation tools, I’d be happy to connect and learn from your experience.
Connecting with others to learn and share knowledge is challenging when working with technologies that are only months old. While I’ve been focusing on workflows and Actions + automation tools to some extent, an OpenAI diamond in the rough that has caught my attention is this: Plan / Spec Mode · openai/codex · Discussion #7355 · GitHub
One recommendation is to keep tabs on the OpenAI GitHub repositories, updates, issues, discussions, etc.
Plan/Spec Mode is exactly the direction I want to go deeper into, so I appreciate the pointer. I’ll also keep an eye on the GitHub discussions to stay on top of the experimental updates.
If you’re experimenting with agent architectures or long-running workflows, I’d be glad to exchange insights.
I’ve been thinking about what you wrote earlier, and I really appreciate how clearly you explained how fast the field is shifting because of vibe-coding tools and new AI development approaches.
I wanted to ask you something more specific, since you’ve already been working in this area and probably see the trends much better than beginners like me.
If you were starting today — in 2025 — aiming to become genuinely strong in AI systems (not just surface-level automation), what would you focus on learning first?
More precisely:
Which skillset will still matter in 5–10 years, when AI becomes even more capable?
What is actually worth studying deeply — architecture? product thinking? workflow design?
Are there any programs, schools, or mentors you would personally trust?
Do you think joining the OpenAI / Microsoft ecosystem early makes sense, or is it better to stay tool-agnostic for now?
I’m trying to build a long-term trajectory, not just “learn one tool”, so I’d really value your honest view.
Thanks again — your previous message really helped me rethink how to approach my learning path.
I will note that that is an SWI-Prolog forum and if you start asking questions not related to SWI-Prolog you may get flagged and the post hidden/removed.
1. Which skillset will still matter in 5–10 years, when AI becomes even more capable?
I’ve been programming for over 40 years, including working with AI in Prolog, and I’ve witnessed four AI winters. Here’s my perspective:
Focus on understanding the fundamentals of LLMs—how they generate the next token given a sequence, or how diffusion works for image generation. Also develop a broader understanding of neural networks and AI concepts. For example, many people think “agents” are a new concept, but the term has been used in AI for decades.
Equally important is learning to judiciously choose what to study. I’ve found the best approach is having a long-term project: if a technology directly helps with that project, I dive deep; if not, I acknowledge it exists and move on (though I may revisit it later).
Early-career developers often try to keep tabs on everything—that’s simply not sustainable. While I recognize that diffusion models are powerful for image generation, I don’t spend time on them because they don’t align with my current long-term goals.
2. What is actually worth studying deeply — architecture? product thinking? workflow design?
As mentioned above, identify a long-term project and let that guide your focus. If a technology addresses a real problem you’re working on, study it deeply; otherwise, pass. A key benefit of this approach is that working on real-world problems accelerates learning. For example, while Advent of Code gets significant attention (and I understand why), it doesn’t align with my long-term objectives, so I don’t prioritize it.
3. Are there any programs, schools, or mentors you would personally trust?
Yes—there are many knowledgeable people I trust and interact with regularly on these forums. However, I can’t provide a one-size-fits-all recommendation for programs or schools, as different approaches work for different people.
4. Do you think joining the OpenAI/Microsoft ecosystem early makes sense, or is it better to stay tool-agnostic for now?
Same principle: identify a real-world problem. If a specific ecosystem (OpenAI, Microsoft, or otherwise) helps you solve it, commit to learning it deeply. If not, maintain awareness but don’t force it. What’s valuable today may become more relevant to your work later.
Note: I’ve intentionally omitted links to key concepts mentioned here—I encourage readers to research these terms themselves as part of the learning process.