Feeling Overwhelmed by AI? Here's Where It's Going – Welcome to Software 3.0
- Anastasia Karavdina
- Aug 2
- 2 min read
If you’ve been following the rapid rise of AI and feel overwhelmed by all the talk about large language models (LLMs), agents, and prompting, you’re not alone!
Many people are asking the same question: Where is all of this heading?
Recently, Andrej Karpathy (ex-OpenAI, Tesla) gave a fantastic talk titled “Software Is Changing (Again)” at Y Combinator’s AI Startup School. It puts today’s transformation into perspective and I love the connection he shows between our time and 60s of 20 century.
From Software 1.0 to 3.0 – A Quick Timeline
Karpathy describes the evolution of software development in three major phases:
Software 1.0 – Explicit CodeWe wrote everything line-by-line. Traditional programming. Logic was hard-coded.
Software 2.0 – Machine LearningWe began training neural networks. The “code” became the model weights, learned from data rather than written by hand.
Software 3.0 – Prompting LLMsNow, we don’t write all the behavior ourselves. We describe what we want in natural language, and LLMs generate the code or response.In short, English is becoming the new programming language.
What This Shift Means
This isn’t just a technical transition—it’s a massive change in who can build software and how it's built.
Here’s what’s changing:
Anyone with clear ideas and good communication skills can build with AI.
APIs and systems need to be designed not just for human developers, but also for AI agents.
LLMs aren't perfect – but when used with thoughtful human-in-the-loop workflows, they can be incredibly powerful.
Karpathy also introduces the idea of "autonomy sliders": you decide how much control the AI has, and how much oversight you want.
One practical approach is the generator–verifier loop:Let the AI generate a draft, then review it, test it, and iterate. This keeps the system agile and safe.
Key Takeaways
If you’re navigating this space—or trying to understand how to apply it in your work—here are the main ideas:
Prompting is the new coding
Build APIs for LLMs, not just for human developers
Keep humans in control – review, audit, and validate
Autonomy is not automation – we are designing for oversight, not surrender
So... What Are We Actually Building?
Despite all the hype around AGI (artificial general intelligence), this moment isn’t about creating sentient AI.It’s about building practical agents that support and accelerate human work:
Faster feedback
Smarter tools
Fewer technical barriers
We're rebuilding the foundations of how software is created and consumed.
That means designing agent-ready systems:
With clear interfaces
That are understandable to LLMs
And easy for humans to guide and review
The Decade Ahead: A New Kind of Developer Experience
The way we design software tools, services, and workflows is changing. The developer experience now includes AI systems. Your API isn’t just for your teammate anymore—it might be called by an autonomous agent.
This shift is significant. And it’s just beginning.
Are you ready for Software 3.0?
Comments