AI Coding Summit 2026

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.

Ishai, CTO at LearnrB, discusses the evolution and impact of AI code reviews in software development. The transition from experimentation to maturity with autonomous AI agents in code reviews is highlighted. Challenges in AI-driven code creation and review processes, as well as the benefits of AI code reviews for productivity and efficiency, are discussed. Common mistakes and challenges in AI code review adoption, along with the importance of adapting processes for AI advancements, are emphasized. The importance of enhancing AI code review processes, measuring their impact, and leveraging AI for productivity metrics is also explored.

Ken Wheeler discusses the evolving software landscape, emphasizing the shift towards higher efficiency and less typing in job roles. The evolution of coding practices from manual formatting to automated tools like Prettier is compared to advancements in construction tools for efficiency. The use of AI for work efficiency and financial benefits, alongside considerations about the future of software engineers in light of AI advancements, is highlighted. Adapting to industry shifts, embracing new technologies like React, and leveraging early AI adoption for career growth are key points. Transitioning to higher-level roles by combining domain knowledge with AI for decision-making and seizing new opportunities is emphasized.

Welcome to spec-driven development talk by Daniel Sogel. AI agents struggle with benchmarks and real-world tasks. Developers spend extra time fixing code generated by AI agents. Security issues in AI-generated code, limitations with context management, multi-file coordination, silent failures, architecture decisions, and enterprise context challenges addressed by Spectre from development. AI agents removing safety checks, generating code with fake data, silent failures worse than crashes, spectrum development benefits like complete context upfront, structured reasoning, validation checkpoints, and autonomous implementation without manual code review. Act with AI agents like pair programmers, combine spectrum, behavior-driven, and test-driven development. Spec-driven development adoption in the industry, using Keyro IDE by AWS and popular tools like SpecKit and OpenSpec for different project types. Define tech stack, API structure in planning, task phase implementation, handover to AI agents, and choosing the right spec-driven development tools.

Louis from Vibe Kanban discusses optimizing workflows for software engineers with coding agents, focusing on planning and review. Leveraging AI tools like GitHub Copilot and ChatGPT can enhance productivity by reallocating time effectively. Reflecting on time saved by AI, the focus is on optimizing planning and reviewing to boost productivity. Emphasizing detailed planning processes and effective use of AI coding agents to ensure successful outcomes. Simplifying reviewing processes to save time and enhance productivity through efficient code review. Leveraging Codex and Claude coding agents for efficient code review to optimize time and workflow. Minimizing time spent on code review by providing feedback to coding agents within the editor, working on multiple tasks in parallel, and automating tasks requested by AI for enhanced productivity.

Today, exploring Ralph Wiggum AI method's simplicity and persistence. Ralph loops ensure eventual consistency in problem-solving. Challenges in AI chats revolve around context management, emphasizing the importance of effective loop iterations. Basic example of loop iteration and backpressure in Ralph. Using skip permissions can be risky. Running Cloud in a sandbox mode for safety measures. Ralph Wiggum AI Method: Autonomous coding with thorough spec creation. Careful planning needed. Upsides include autonomous operation and clear task handling.

Alex and Rodi discuss antigravity, a powerful coding agent that unifies Google tools. The agent manager orchestrates multiple agents, providing a higher-level view than just code. Features include an AI editor, agent-controlled Chrome browser, and best practices for agent autonomy and usage. AI steering takes time before feeling intuitive. Anti-gravity offers fast and planning modes for building efficiently. The planning mode allows fine-tuning and execution of complex tasks. Playground validates ideas before full projects. The playground in AntiGravity validates project ideas. Image generation, Nano Banana Pro, and Gemini 3 enhance design output. Skills support deep context tasks. Asynchronous nature enables efficient project development. Browser agent serves as a versatile tool for bug fixing. Manual control and inspiration amidst automation. Transition to agent manager for enhanced visibility and decision-making. Encouragement to explore diverse agent-led experiences and updates for workflow improvement.

Nikita, AI Architect at Siemens, discusses challenges in JNTX system development, emphasizing domain-driven design for scalability and stability. Clear responsibilities, bounded contexts, and structured schemas are crucial for reducing complexity. The anti-corruption layer and context maps play vital roles in integration architecture. Key takeaways include the importance of contracts, firewalls, and CI gates for system development.

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.

Speaker specializing in technical architecture discusses the impact of interruptions on deep focus, productivity, and software development challenges in global organizations. Content adaptation to multiple languages required manual glyph corrections and a structured collaboration plane. Transitioning to automated systems raises the question of why orchestration remains manual, with AI as a glue agent. Strategic stock analysis involves N8n workflows, specialized agents, and structured contracts for consistency. Workflow architecture includes JSON contracts, specialized agents, and structured HTML emails for stock analysis. Workflow reliability focuses on independence, idempotence, and robust frameworks for adaptability. Automation and AI integration emphasize process streamlining and human intervention in decision-making. Intelligent orchestration systems involve encoding intent by humans and AI turning it into action, focusing on clear workflows and architecting for control.

Introduction to AI security pitfalls, emphasizing the importance of understanding AI promises and risks, responsibility in verifying AI-generated code, the necessity of code verification and thorough review for secure deployment, ensuring code validation and sanitization for secure applications, risks of unsecure authentication logic and broken authorization due to AI usage, the importance of avoiding exposing secrets in code by trusting AI blindly, learning resources on cloud security and platforms like Flare Learning and TryHackMe, and the significance of creating AI rules, testing, and following best practices for code security.

Welcome to AgentCraft with Iddo Salomon, AI lead. Scale up agents for efficient work but consider cognitive overload. Agent orchestrators are crucial for task management and real-time reactions. AgentCraft is a tool inspired by real-time strategy games to manage agents effectively, with features like heroes and buildings. It visualizes repository files through terrain projection, simplifies code inspection, and enhances task management with cloud code support. Task delivery is streamlined with code merging, quests, and hero summoning for project improvement.

Gil Friedman from Backslash Security discussed MCP security, focusing on the Model Context Protocol and AI assistant security risks like GitHub token exposure in auto run mode. Security measures for MCP usage include setting autorun to ask every time, postmarking MCP incidents, handling malicious servers, safeguarding sensitive information, reviewing and disabling unused tools, requiring permission before triggering actions, removing MCP from environments, and revoking API keys when necessary.

A demo of an animated goose following the cursor in a unique environment with real-time updates. Advanced models for long-running agents performing complex tasks. Exploring sandbox environments for agent coding, focusing on isolation, containers, networking, and persistence. Challenges in sandbox orchestration, including handling heavy processing, managing sessions, and security concerns. Importance of agent control, Micro VMs for isolation, and networking in sandbox environments. Strategies for efficient sandbox operations with pre-built images, persistent volumes, and warm pools. Utilizing sandbox primitives for simple usage with focus on isolation, networking, and persistence.

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.

John Robert, Lead Data and AI Platform Engineer, discusses evaluating AI agents beyond success rate. Talks about failures of AI projects, the uniqueness of AI agents, and the evaluation process. Creating a Framework for AI Agent Evaluation: Four Evaluation Categories - performance, business, safety, and cost. Metrics creation to enhance AI projects. Includes task completion rates, reasoning quality, accuracy evaluation, tool execution, response time, and recovery time checks. Examining KPIs, Safety, Security, and Cost in AI Agent Projects: Consider ROI, user satisfaction, time-saving, code review, adoption, engagement, risk, regulations, unauthorized actions, and cost efficiency. Checking Infrastructure and Security Metrics: Evaluate project reliability, online presence, errors, resources, prompt injection, data leakages, authorized actions, tool usage.

Stepan Suvorov, CTO at Roadsoft, discusses the rise of Voice Agents, highlighting market growth, improved models for speech to text and voice generation, reduced latency, and cost-effectiveness. Key components include speech-to-text, LLM analysis, text-to-speech conversion, orchestration, and addressing latency. Delving into essential aspects like voice activity detection, interruption handling, and model selection criteria for Voice Agents. Exploring parameters such as latency, quality, pricing, language support, and voice cloning options. Comparing DIY and managed solutions for Voice Agents, emphasizing flexibility and cost-effectiveness. Discussing cost implications between managed and self-hosted solutions, considering usage volumes and scale. Addressing infrastructure challenges, testing approaches, WAPI limits, model selection, and future plans for voice agents.

Lugari Karasen from Red Hat AI discusses the technical aspects of large language models, including data collection, tokenization, and neural network internals. Model training involves converting human language to mathematical representations and adjusting parameters iteratively in a complex environment. Inference engines like VLLM aid in deploying models for rapid data processing. Optimizing model size for efficiency without sacrificing accuracy is crucial, with quantization reducing model size while maintaining precision. Local deployment offers privacy and control, and smaller purpose-driven models can enhance workflows and experimentation.

Todd Schiller discusses using AI to code an AI-enabled extension for global use by enterprises, emphasizing empowering users through browser extensions for versatile automation and customization. Browser extensibility covers various customization methods, including user scripts and low-code options like Pixie Bricks. Challenges in applying AI tools to browser extensions include their distributed nature, slower iteration loops, and adaptation to host sites. Understanding browser extensions as distributed systems reveals complexities such as async operations and intricate component lifecycles. Key best practices will be discussed, starting with selecting the appropriate foundation. Choose a productive foundation like libraries or frameworks for browser extension development. Select between user scripts or low-code platforms like Pixiebricks. Focus on encapsulating extension-specific code, using explicit coding, and enforcing boundaries with linters. Create fast feedback loops with tools like TypeScript, Lint rules, and realistic integration testing with mocks and fakes. Encapsulate extension-specific code using standard tactics like dependency injection and explicit coding. Create fast feedback loops with TypeScript, Lint rules, and realistic integration testing. Shift left to catch issues earlier and avoid relying on slow feedback loops. Avoid testing on host pages directly to prevent slow and potentially flaky behavior detection. Utilize dependency induction, mocks, and fakes for realistic integration tests. Test boring components with Storybook and Playwright MCP. Shift left to catch issues earlier, avoid testing directly on host pages, create test pages for encountered patterns, leverage AI for generating test pages, and ensure a stable testing environment.

Shifra guides the migration of a Repl.it app to Render with emphasis on simplicity and mock data. Ensure Render deployment requirements are met, including specifying root directory, build and start commands, and correct port. Configure GitHub repository setup, connect to Git, and create a repository. Confirm code in the repository and deploy on Render, showcasing a sample landing page.