AI Native Development: Context Engineering and AGI Productization
Logan Kilpatrick from Google DeepMind discusses the shift from prompt engineering to context engineering, the rise of agentic coding, and why AGI will emerge as a product ecosystem rather than a single model. Key insights on developer productivity and future software roles.
The Shift from Prompting to Context Engineering
The landscape of AI-native development is undergoing a fundamental paradigm shift. Logan Kilpatrick, a member of technical staff at Google DeepMind, argues that traditional prompt engineering is effectively a "bug" in the current workflow. The new standard is context engineering, where the human role shifts from crafting precise instructions to curating and providing disparate data sources. Modern models, particularly those with deep research capabilities, can now autonomously browse, retrieve, and synthesize information. This allows developers to ask high-level, context-thin questions while the system handles the heavy lifting of finding relevant data, significantly reducing manual effort and improving output accuracy.
Agentic Coding and Productivity Gains
A critical operational change is the move toward batch processing in agentic coding. Kilpatrick notes that the rules of AI interaction are being rewritten rapidly. What was effective six months ago is now obsolete. To achieve frontier-level productivity, developers must stop asking for minimal, iterative changes and instead request comprehensive, multi-part solutions in a single prompt. Current models are robust enough to handle complex, multi-step logic without fumbling, making the rate limit the primary constraint rather than model comprehension. This shift requires mental plasticity from engineering teams to continuously adapt their workflows to evolving tool capabilities.
Redefining AGI and Developer Roles
The conversation around Artificial General Intelligence (AGI) is decoupling from academic definitions of model generality. Kilpatrick posits that AGI will not be a single model release but a product experience created by orchestrating multiple systems, agents, and tools. This "narrow super intelligence" is already visible in coding tools that can build anything with code. For the software industry, this means the value of developers is shifting away from typing characters and toward problem-solving, architectural intuition, and handling edge cases. As the volume of software increases, the gap between what average users can achieve with AI and what experts can architect will widen, ensuring that deep technical expertise remains a high-value asset. The future of development is not about fewer developers, but about a broader population touching code, with a specialized tier handling complex system design and reliability.
Key insights
-
Prompt engineering is being replaced by context engineering, where the human provides raw data sources and the model autonomously retrieves and synthesizes information. This shift leverages deep research capabilities to reduce manual instruction crafting.
Impact: Reduces developer time spent on iterative prompting and increases the accuracy of AI outputs in complex, data-heavy tasks.
-
Current AI models can handle comprehensive, multi-part requests in a single interaction, making iterative, minimal prompts inefficient. The primary constraint for productivity is now rate limiting rather than model comprehension.
Impact: Enables faster development cycles by allowing developers to define entire features or modules in one go, maximizing throughput per API call.
-
AGI will manifest as a product ecosystem rather than a single model. It requires the orchestration of agents, tools, and infrastructure to solve general problems, representing a 'capability overhang' where existing models are underutilized by current product designs.
Impact: Shifts competitive advantage from model training to product integration and orchestration, favoring companies with strong platform ecosystems.
-
Reusable 'skills' act as token-saving efficiency mechanisms, preventing models from wasting resources on failed API calls or redundant web searches. These skills encode domain-specific context and architectural decisions.
Impact: Lowers operational costs and improves reliability in agentic workflows by standardizing how models interact with specific APIs and codebases.
-
The value of software engineering is shifting from code generation to problem-solving and architectural decision-making. As AI handles routine coding, the premium is placed on handling edge cases and making high-level technical judgments.
Impact: Reshapes hiring and training criteria to prioritize systems thinking and intuition over syntax proficiency, ensuring long-term relevance of engineering roles.
Action items
-
Audit current AI workflows to identify areas where manual prompt crafting can be replaced by providing raw context sources. Implement deep research tools to allow models to autonomously gather and synthesize data.
Impact: Reduces manual effort in data preparation and improves the accuracy of AI-generated insights by leveraging autonomous retrieval capabilities.
-
Train development teams to batch complex requirements into single, comprehensive prompts rather than iterative, minimal requests. Update internal guidelines to reflect current model capabilities for multi-step logic.
Impact: Increases development speed and reduces API usage costs by maximizing the output value of each interaction with the AI model.
-
Develop and maintain a repository of reusable 'skills' that encode domain-specific context, API interactions, and architectural preferences. Integrate these skills into agentic workflows to standardize execution.
Impact: Improves the reliability and speed of agentic tasks by reducing token waste and preventing models from fumbling through trial-and-error execution.
-
Reframe engineering performance metrics to focus on problem-solving, architectural decisions, and edge-case resolution rather than lines of code written. Update hiring criteria to prioritize systems thinking.
Impact: Ensures the engineering team remains valuable in an AI-augmented environment by focusing on high-level strategic contributions that AI cannot easily replicate.
-
Evaluate product roadmaps for opportunities to orchestrate existing AI models into broader product ecosystems. Focus on integration and tool-use capabilities to capture the 'capability overhang' of current models.
Impact: Positions the company to deliver AGI-like experiences through product innovation rather than relying solely on next-generation model releases.
Quotes
“my worldview has always been that like prompt engineering was a bug like if you go and talk to users they don't want to prompt engineer and actually like the things that you're asking them to add in already exist somewhere else”
“AGI is not going to be a model it's going to be a product that somebody creates”
“the difference between those two things is where I think traditional software engineering developers will add a huge amount of value because that gap is going to change the way to use AI tools today”