The Framework JetBrains Uses to Predict the AI Market Landscape
In a Nutshell
The AIDE framework maps AI development tooling across three dimensions: 35 software creation activities, five delegation levels from simple tools to full autonomous execution, and organizational contexts. The market currently sits at L3 (general purpose executors like Claude Code), with cloud agents just reaching L4 capabilities. Future development will shift from code implementation toward solution engineering and intent management, with AI transitioning from tool to managed intelligence resource within vertically integrated production platforms.
These notes were generated by AI and may contain inaccuracies.
Speed of change in AI is manageable when viewing the bigger picture rather than tracking every individual update. The approach focuses on understanding where current developments originate and where they ultimately lead, rather than monitoring everything happening in real time. This creates a filtering mechanism that prevents change fatigue.
AI has become standard in developers' professional lives, adopted not only for personal projects but also in enterprise environments. Agentic coding represents a fundamental shift where developers delegate complete development tasks to agents for end-to-end execution, moving from code-validate-fix cycles in editors to task setting, execution, and review in agentic chat interfaces.
Anthropic's Claude Code drove significant market movement, currently used by approximately 8 million coding professionals and generating around 6 billion in yearly revenue. The release initiated a competitive landscape of IDE-agnostic CLI coding agents, with offerings now available from OpenAI, Google, and specialized companies.
Agents are transitioning to cloud environments, enabling autonomous work delegation. Current cloud agents can handle simpler tasks such as fixing linting errors from failed CI runs, though capabilities remain limited to what are described as garbage tasks.
Despite widespread AI usage, adoption concentrates in two primary activities: literal code writing and brainstorming. Significant portions of the software development lifecycle remain underutilized for AI, representing substantial growth opportunities.
A dedicated cross-functional research team conducts quarterly global surveys across eight national languages, targeting the general population of coding professionals rather than only existing users. Identity concealment reduces bias. The surveys collect data on individual and organizational AI adoption, satisfaction metrics, and NPS scores, enabling market size estimation and revenue analysis for individual players.
The progression began with single-line code completion, advanced to multi-line completion with generative AI, then incorporated conversational AI chat interfaces. This evolved into integrated AI features across development environments, leading to AI code editors like Cursor and Windswept. Agentic development environments emerged next, followed by cloud-based autonomous agents.
The framework emerged from applying physics concepts of fundamental theories to AI development tooling. The Artificial Intelligence Development Environments framework (AIDE framework) was developed to explain not current market states but what is fundamentally possible in AI development tooling.
Artificial Intelligence System (AIS) refers to any computer system demonstrating intelligence traits while helping users achieve their goals, independent of whether intelligence derives from LLMs or deterministic heuristics.
Artificial Intelligence Development Environment (AIDE) applies AIS concepts specifically to creative software development tools.
Principal-agent terminology from economics and sociology describes delegation relationships, where both principals and agents can be either human or AI systems.
Software creation terminology replaces traditional software development language to acknowledge that developers represent only part of a larger group contributing to software creation.
Four key assumptions underpin the model: software creation needs will persist indefinitely, gradual delegation to artificial systems drives market change, humans retain task setting and oversight roles permanently, and higher delegation levels reduce personal immersion in specific development activities.
The framework operates across three dimensions: stages of the software creation process (35 high-level activities grouped into five activity groups spanning ideation through maintenance), levels of delegation, and organizational context.
L1 Tool involves delegation of limited, scoped actions such as code completion.
L2 Assistant delegates well-defined sequences of actions or tasks with specific boundaries, creating the sensation of having a third hand.
L3 General Purpose Executor shifts focus from process to deliverables, where agents execute tasks requiring expert knowledge input from principals acting as consultants. Current state-of-the-art agents like Claude Code and Codex operate at this level.
L4 Supervised Executor moves beyond individual scope to organizational perspectives, where agents manage specific development functions with principals limited to approving key decisions.
L5 Computer Center represents complete delegation of development as an organizational competence, where principals define objectives and metrics while agents handle all execution decisions including technical architecture and infrastructure.
Delegation levels represent attributes of principal-agent relationships rather than agent capabilities. Principals determine delegation levels based on their perception of agents, potentially delegating less than technically possible in critical scenarios.
The framework differentiates between individual development, SME development, and enterprise development, each presenting distinct constraints and organizational dynamics that affect development decision complexity.
Data indicates the market sits at approximately L3 general purpose executor level, with cloud agents only beginning to touch L4 capabilities. Significant gaps exist across all directions, indicating substantial room for growth.
The rise of abstractions will shift developer focus from code implementation toward solution engineering, architecture, data flows, and business impact.
Vertical integration of AI across the entire software development lifecycle will replace current tool fragmentation with integrated development platforms.
AI will transition from being perceived as a tool to being managed as an intelligent resource, analogous to raw materials in production processes where intelligence becomes a variable input resource.
When organizations understand that AI is not just a hammer, but it's a wood furniture factory, that's probably where you might experience a fundamental shift in how software development is done. As a result, we might end up with something like agentic software production platforms.
A vertically integrated software production system where AI is working as an input raw intelligence resource, and which is just organized around the essence of every kind of work, not just software development. That is intent management, execution, and governance. Rather than traditional DevOps cycle. And human developers sort of like managing this production system and orchestrating it.
This looks really exciting, but at the same time, mildly terrifying. What can we personally do to prepare for future like that?
Our model doesn't mean that this is exactly the future that will come. Our framework tells that it's fundamentally possible due to the existence of some key forces which drive the development of the software market. But, it doesn't tell you whether we particularly land in this particular version of reality and when.
On the other side, understanding this exact fundamental forces really well can be already a great help for you right now. So, try to answer to Does AI really grow the opportunities of development work delegation? Can it act as this intelligence resource and intelligence raw material? Is it possible to ultimately have a hybrid teams of human developers and AI workers? Or is there anything else going on?
Build your own model or use our 80s framework, accept it and start asking another questions, but within the boundaries of this conceptual model. What will my responsibilities as a developer look like? What skills will be more relevant? If we are speaking about management of delegation, does it mean that certain managerial skills will be a must for every developer in the future? And if we focus on pure engineering, does it mean that everybody will become software architects? But then we should have something like senior architects, middle architects, and then junior architects?
What skills do I already have? What skills will I need to develop? And am I really ready to change? And then just observe the world, validate your model and assumptions. The good thing that change does not happen overnight. And there is always continuous process, there is always some friction and some inertia.
If we have here any engineering leaders watching us now, your objective is much more complex because you have to rethink many things from how the development process is structured to how you can treat AI as this intelligence resource, especially how you should embed it into your systems where you already have intelligence as a human capability. But this can be a topic for another discussion because we already have here as some vision as a company.
Or in the end after all, you can just go to YouTube, watch some videos on fundamental physics, and get inspired as we did.
Keep JetBrains in your library
Save the videos and channels worth coming back to, and find them again in one place.
