Deep Dive: AI Assisted Coding

Software engineering is evolving with AI and VIBE coding reshaping work, emphasizing collaboration and embracing AI. The future roadmap includes transitioning from augmented to AI-first and eventually AI-native developer experiences. AI integration in coding practices shapes a collaborative future, with tools evolving for startups and enterprises. AI tools aid in design, coding, and testing, offering varied assistance. Context relevance, spec-driven development, human review, and AI implementation challenges are key focus areas. AI boosts productivity but faces verification challenges, necessitating human oversight. The impact of AI on code reviews, talent development, and problem-solving evolution in coding practices is significant.

Tejas discusses AI agents and productivity at the React Summit, emphasizing the role of AI in enhancing user experiences. The use of JSON data and system prompts to improve parsing and streaming processes is highlighted. The importance of AI agents in web navigation for increased productivity and efficient website interaction is explored. Development of tools like the Google Calendar Manager Agent for better calendar management and troubleshooting calendar event creation are discussed. The talk also delves into future possibilities in UX and technology, along with insights on AI component creation and comparison between Langflow and N8n.

Discussing building better UIs faster with AI, emphasizing the use of LLMs as predictive text models and the need to maximize their smart aspects in coding and UI development. Exploring file context management in Cursor IDE and demonstrating quick UI updates using AI for coding efficiency and real-time results. Discussing advanced workflows for efficient development using feedback-driven approaches and incorporating test-driven development for iterative and automated testing in Cursor IDE. Exploring advanced prompt tricks in Cursor IDE for efficient development by writing tests first, updating code iteratively, and enabling autorun mode for automated testing and bash commands. Cursor's Annoying Prompt Behavior, Automated Testing Benefits, and Test-Driven Development for Reliable Code. AI Feedback Importance and AI Testing Complexity with Clear Goals. AI Challenges, Visual Front-End Focus, and Fusion Preview. Integration of Figma Designs for Visual Tweaks, Exploring New Features and Design Systems, Updating Customers Tab and Making Visual Edits, Making Carousel and Visual Updates, Granular Design Updates and AI Integration, Dynamic Updates and Component Refactoring, Tagging Builder Bot for Automated Code Updates.

Chaitanya, Principal Engineer at Atlassian, discusses the design system at Razorpay, the impact of AI on UI development, and the integration of AI with design systems for enhanced productivity. Detailed prompts for AI to build UI components can be cumbersome. Imagine a seamless process where AI interprets Figma designs to create UI. Leveraging design expertise and focusing on business logic, not writing detailed AI prompts. Blade's MCP server facilitates the magic of transforming Figma designs into UI code by collaborating with Figma and OpenAI.

Cloud Code introduced sub-agent feature for better performance. Understanding the sub-agent concept and its role. Challenges of sub-agents in direct implementation. Parent agents' limited visibility. Efficient context management with MD files for sub-agents' tasks and reports boosting Cloud Code performance. Guidelines for creating useful sub-agents with specific MCP tools. Utilizing TweakScene MCP for scene design reference and defining rules for sub-agent tasks. Setting goals, output format, and rules for sub-agents. Building a chat GPT replica with chat.cn UI and Versatile AIS. Integration with VerCell AI SDK for seamless implementation.

Carly discusses memory, context in LLMs, and managing information for effective results in agentic systems. LLMs hallucinate due to knowledge gaps, biases, overfitting, and incentives. Importance of context in minimizing hallucinations and focus on memory storage for decision-making. Short term memory acts like RAM, while long term memory stores knowledge and experiences. Efficient memory storage using disk, file systems, and semantic memory. Emphasis on managing message context, retrieval, and concatenation for LLM input. Attention to context length, information evaluation, and memory optimization for accurate responses.

The Talk delves into agent decoding in Cloud Code, emphasizing the importance of planning, acting, and verifying changes systematically. It explores maximizing efficiency through Cloud Code acting as an agent, reasoning, reading, writing code, and using tools for successful outcomes. The discussion covers enhancing task planning, utilizing subagents for specialized assistance, and optimizing tool use with MCP integration. It introduces Cloud Skills, plugins for team collaboration, and efficient tool utilization with parallel workflows. The concept of non-interactive agents, guided autonomy, and steps for experiencing agent decoding are also highlighted.

Discussing top AI coding best practices in 2026, including semantic understanding and context window management. Exploring the role of Language Server Protocol (LSP) in code refactoring. Efficient code renaming using LSP and code intelligence in Kiro. Impact of not using LSP on code renaming efficiency. Manual approaches without LSP significantly impact efficiency and resource consumption. Context window usage doubles without LSP, affecting code handling. Utilizing subagents for specialized tasks enhances codebase security. Spec-driven development and detailed design documents for efficient agent alignment.

Codium is a generative AI assistant for software development that offers code explanation, test generation, and collaboration features. It can generate tests for a GraphQL API in VS Code, improve code coverage, and even document tests. Codium allows analyzing specific code lines, generating tests based on existing ones, and answering code-related questions. It can also provide suggestions for code improvement, help with code refactoring, and assist with writing commit messages.

AI design challenges include bias, safety, and security. Trust and transparency are important in AI. Design principles for AI include user control, fighting bias, and promoting good decision-making. AI can enable the design process and investors expect to see it included in products. AI empowers individuals to create and share ideas, but managing expectations is crucial.

AI in game development has evolved rapidly, with generative AI being a focus. However, game developers like Romero Games have concerns about ethics and prefer using AI to automate processes and make creative work easier. AI has been used in games for decades, from path-finding AI to decision trees. Procedural world building and advanced AI technology are pushing the boundaries of FPS games. Different teams within a company have different approaches to the use of AI, depending on their specific needs and requirements.

Hey, my name is Mikhail Levchenko. I've changed my job and became a full-time developer productivity engineer. Let me share a little story. Imagine being a wealthy trader in 15th century Europe. Delivering spices for profit was not an easy task. Middle East nations heavily taxed any traders. But the Portuguese sailed around Africa using the Mariner's Astrolabe. Fast forward, Gerardo Mercator invented a map perfectly suited for deep ocean navigation. The Dutch East India Company became the new king of the spice trade. Developer productivity is about building a cozy environment for developers, solving hard problems, and identifying process bottlenecks. It focuses on the whole development process as a product. By leveraging this perspective, you can drive change among developers and other roles. Observing developer processes as a product involves meeting stakeholders, formulating a vision, creating a model and metrics, finding productivity problems, and proposing solutions. The stakeholders include developers, CEOs, CTOs, engineering managers, and other roles involved in software development. The goal is to make everyone more productive, guided by the motto of happier developers building better software faster. There are three highlighted frameworks in this talk: DORA, DEVX, and Mertrix. DORA focuses on speed and reliability with metrics like acceleration frequency, lead time, failure rate, and recovery time. DEVX addresses the human factor with metrics related to feedback loops, cognitive load, and flow state. DX4 is a recent metric that combines DORA and DEVX, introducing proprietary metrics for top management. It includes a failure rate to encompass reliability and a percentage of time spent on new features. Engineer-developed metrics like the reliability metrics from Dora and the percentage of time spent on new features are valuable additions. Talking to top management about changes is made easier with a cool metric. Research concerning metrics, consult stakeholders, and read literature to improve understanding. Try prototypes before developing custom tools as a last resort. Confirm success with empirical metrics. Release managers often feel constantly interrupted and overwhelmed by the task of supervising a release. Sharing the role with the entire team proved to be ineffective, leading to forgotten procedures. Automation and streamlining of the release process improved satisfaction and reduced delayed releases. Make developer productivity part of company goals and culture code. Advocate for developers to top management. Drive change by starting small, experimenting, and using tools as a last resort. Find ways to make teammates more productive.

I'll walk through some of the ways I set up Cursor to help speed up my full stack workflow. The main thing to check is indexing the code base. Cursor also looks at the get graph file relationships. Another important setting is large context, which allows absorbing larger files and multiple files of context in the prompt window. Iterating on lints helps find and fix errors faster. Web search tools and uploading your own docs are also available to enhance the prompt window. These are all preferences, including auto select, YOLO mode, and models. Rules are also crucial as they guide the model to understand your workflow. Implement best practice error handling and adjust rules as the models and system prompts change. MCPs like sequential thinking, Postgres SQL, and Puppeteer help refine thoughts and automate workflows. Creating a product requirements document in a .md file allows AI to ground itself in project information. The game was successfully implemented and all items were checked off. Debugging strategies involve adding debug logs to help identify issues. Use commit messages to provide clear documentation for changes. Demonstrating the use of AI in debugging by setting up and utilizing debug logs. Showcasing the benefits of providing debug information to resolve issues. Overview of various topics including indexing rules, linting, MCPs, PRDs, tech stacks, and checklists.

Today's Talk discussed integrating context into AI coding tools using the MoloContext protocol (MCP). The demos showcased the use of Conduit, an MCP tool, for bug tracking and feature addition in software development. The Talk also highlighted the use of Fetch in Cursor to summarize GitHub issues and create Jira bug reports. It demonstrated how Conduit can be used to manage issues, make code changes, and perform Git commits. Additionally, it explored the integration of Conduit with Windsurf to add a user interface for the conduit application. The Talk also touched upon implementing MCP in the AIDER AI coding system and using MCP and Windsurf for Conduit integration. Overall, it emphasized the importance of context and choosing the right tools for efficient software development workflows.

Hello, everyone. I'm Mansa Hari, and I'm very excited to open the session on how AI can automate 70% of code review and reduce the burden on developers. AI can cut review time, automate PR suggestions, and identify code issues. Implementing AI code reviews can reduce manual review time, ensure high code quality, and accelerate development cycles. The future of AI in code reviews includes understanding business logic, intuitive code suggestions, and hybrid AI-human reviews. Tools like Co-Pilot can be used to automate code review tasks and improve code quality.