Software 3.0: What product teams need to know
Redesigning APIs for AI Agents, Generative Engine Optimization (GEO) and more. OpenAI’s co-founder says we’re now in the era of Software 3.0. Here’s everything that matters to product teams.
🔒The Knowledge Series is available for paid subscribers. Get full ongoing access to 80+ explainers and AI tutorials to grow your technical knowledge at work. New guides added every month.
Many of us are still scarred by the promises of the “Web 3.0” era, which ultimately ended up being little more than a phrase used by blockchain and crypto enthusiasts to convince people to invest in a vision of a tokenized web that never fully materialised.
The current AI-powered transformation of the web may have been a far more appropriate time to adopt the term Web 3.0 but instead of confusing things, OpenAI’s co-founder Andrej Karpathy recently opted for Software 3.0.
Here’s the slide from his recent presentation which puts this fundamental shift into context:
But what exactly is Software 3.0 - and what are the implications for product teams?
In this Knowledge Series, we’ll dig into some of the core concepts and expand upon these to understand how product teams can prepare for this new era. From updating API docs to accommodate AI Agents and non-engineers building prototypes through to building features with “partial autonomy” that allow AI and users to work together, Software 3.0 upends many of the ways of building products that we previously took for granted.
Coming up:
What exactly is “Software 3.0?” The key principles explained
Why this matters to product teams
Real world examples of Software 3.0 principles in action - insights and perspectives from chief design officers and how companies like Stripe, Perplexity, Cursor and others are adopting Software 3.0 principles in practice
GEO vs SEO - the new ways to attract organic users in a post-click web
How to prepare your API products for the new era of AI agents
What exactly is “Software 3.0?” The key principles explained
The phrase “Software 3.0” was coined by Andrej Karpathy, co-founder of OpenAI, who also came up with the term “vibe coding”. Here’s a snapshot of some of the core principles of Software 3.0 and the potential impact of these on product development teams:
1.Programming in natural language
Programming is shifting away from hand writing code to writing prompts with context. This is freeing up the time of engineers, with over 50% of pull requests now using AI and the average time saved per pull request has grown from 8.9 hours to 13.7 hours according to a new report. This means that while engineers can focus on solving deeper problems, tasks like building high fidelity prototypes to test with customers are often now in the hands of product designers and product managers. Gumroad’s CEO recently claimed that this marks the death of traditional front end engineers.
This shift has also led to the explosion of vibe coded apps - particularly for internal use. Tools like Lovable now have over 500,000 users and the use of so-called “Shadow apps” built using AI but without permission is set to grow to over 160,000 by Q2 next year.
OpenAI’s former head of research still calls1 vibe coded apps a “net liability” overall but it’s increasingly looking like the era of clogging up development backlogs with work to build internal apps that users never see may be coming to an end, as building internal apps and connecting them to the relevant infrastructure becomes easier over time.
We’ll take a closer look at how one company incorporates vibe coded prototypes into their product development and design process later.
2. Optimizing infrastructure for AI
The second broad principle of Software 3.0 is the idea that product teams should not only be optimizing their infrastructure and APIs for human users any more - but also for AI agents / LLMs. We’ll take a look at some of the ways you can optimize your APIs and documentation to accommodate AI agents in a second.
Infrastructure changes can also be made to accommodate LLMs - with some companies already implementing changes that make it easier for AI models / LLMs to crawl their website. More on that later.
3. UX and human / AI collaboration
As more companies release more AI-powered features, users are increasingly coming to expect that products will ship with AI features. In practical terms, this is creating what Andrej Karpathy calls the “autonomy slider” - a way for users to decide how much control and autonomy to give AI to control products vs humans.
Check out previous DoP Deep dives on new AI features if you’re interested in learning more about what companies are shipping:
Why and how this matters to product teams
If we take each of these 3 broad categories of Software 3.0 principles as our foundation, we can start to unpack some of the ways this impacts product strategy and the software development processes that product teams follow.
First, let’s consider APIs, documentation and AI agents. According to one engineer, we are fast approaching the point where LLMs and AI agents are reading API and SDK docs more than humans:
This shift has a significant impact on product teams, shaping the way APIs and their respective documentation is designed.