Ideas lab
How enterprises can move beyond AI innovation theatre
Article
24 Oct 2025
By Ismail Amla, Senior Vice President of Kyndryl Consult
We’re entering the era of Software 3.0, a phase defined not by code but by conversation.
Large language models (LLMs) and AI-powered coding tools are transforming how software gets made, replacing lines of code with natural-language prompts. And it’s in this new paradigm that text becomes more than just an interface — it’s the new programming language of business, expanding creation beyond the domain of developers and data scientists.
As Andrej Karpathy observed from his time at Tesla, this evolution builds on, rather than replaces, what came before: the rigid codebases of Software 1.0 and the data-heavy model pipelines of Software 2.0. What’s emerging now are intelligent agents that can reason, generate, and automate — radically simplifying how ideas turn into applications.
We’re already seeing it at AI-forward companies. Take Zapier, itself a no-code automation platform, where the HR lead builds a system to screen out job applicants who are likely fraudulent. Or look to Shopify, where revenue teams build code modules to make it easier to generate new products.
As Karpathy notes, models are now capable of generating code autonomously to resolve problems or extend functionality. But people remain an indispensable link — translating ideas into intent, setting boundaries, and steering these systems toward meaningful outcomes. The work doesn’t disappear; it shifts, becoming more about direction than execution.
As we move deeper into this new era of AI, the senior leaders are going to confront big new questions. CEOs should be asking: If prompts are programs, who is writing your company's software now? The answer increasingly includes non-engineers, but also the very software systems themselves. CIOs must rethink software development pipelines and expand their purview to include code from non-engineers. CTOs must plan for hybrid stacks where retrieval-augmented generation sits upstream of APIs and MCP servers wrap most critical services to enable coordination with other AI agents and natural language interactions.
In short, software 3.0 isn't a technical upgrade. It's a leadership moment, giving employees greater agency and ability to affect their companies in ways previously requiring far more coordination and costs.
For incumbents that traditionally struggle to move as quickly as startups, Software 3.0 can be transformative. Traditional development bottlenecks disappear as domain experts build systems directly. Engineers can focus less on business logic and more on infrastructure and delivery. The companies that thrive will leverage Software 3.0 while building defensible positions.
So the question becomes: What's your moat when text is the programming language?
Control the specialized knowledge
Domain expertise becomes more crucial when code becomes a commodity. Palantir's moat isn't code complexity — it's deep knowledge of intelligence workflows, data integration patterns, and government requirements gained painstakingly through frontline experience. That moat can't be easily replicated through prompts alone.
How to moat it: Make domain experts who understand nuanced business problems and contexts your prompt architects to multiply their impact.
Own the critical data
When models can generate functionality, proprietary data becomes the key differentiator. Bloomberg built its empire on financial data access, not just software capabilities. Tesla’s entire business model is premised on its cars collecting data. In Software 3.0, that data becomes even more valuable as the foundation for specialized AI applications.
How to moat it: Identify unique data assets within your organization. Build systems that make this data AI-accessible while keeping it proprietary.
Create compound AI workflows
Simple prompts are commoditized, but complex multi-step AI workflows that combine reasoning, retrieval, and action create sustainable advantages. KPMG built a 100-page multi-step prompt to power its TaxBot for corporate clients. Companies building sophisticated prompt orchestration — where multiple AI agents collaborate on complex tasks — can tackle more complex problems more effectively.
How to moat it: Develop multi-agent systems that handle end-to-end business processes. Focus on breaking down workflow complexity, not individual prompt sophistication.
Build the people-AI feedback loops
The most successful Software 3.0 implementations aren't fully automated — they create tight feedback loops between our own expertise and AI’s capabilities. This creates institutional knowledge that's harder to replicate than standalone AI systems.
How to moat it: Design systems where human experts continuously refine AI outputs. Make the human-AI collaboration itself your competitive advantage.
Embed AI natively in workflows
Rather than building standalone AI tools, embed intelligent capabilities directly into existing business processes where users already work. This creates natural stickiness and reduces the likelihood of replacement by external solutions.
How to moat it: Integrate AI capabilities into core business applications rather than offering separate AI tools. Make intelligence feel seamless, not additive.
Software 3.0 as an advantage and differentiator
Rather than viewing Software 3.0 as a threat to technical differentiation, forward-thinking companies will see it as an opportunity to compete on business insight, data assets, and workflow sophistication. This means smart organizational design, better human-AI collaboration, or more traditional approaches like owning unique data for competitive advantage. As Software 3.0 removes barriers to building software, every company needs to identify new sources of sustainable competitive advantage. The democratization of programming doesn't eliminate moats — it just moves them.
