From Autocomplete to Agents: AI Coding State of Play

From Autocomplete to Agents: AI Coding State of Play

The presentation, delivered by Birgitta Böckeler from Thoughtworks, explores the evolution and current state of AI coding assistants, from basic auto-completion to advanced agent-based systems. The session provides insights into the features, capabilities, and workflows enabled by these AI tools, aiming to both enthuse skeptics and ground enthusiasts.

Key Points:

• History and Features: Initially, AI tools began enhancing coding with auto-suggestion, evolving into chat-based systems where questions could be directly asked within an IDE. Current tools offer enhanced IDE integration, allowing direct interaction with code bases through inline chat and contextual queries.
• Agentic Assistants: These systems, referred to as 'agents,' orchestrate prompts by leveraging the entire codebase context, interacting through tools within the IDE. This allows agents to automate and execute commands, enhancing task automation and efficiency.
• Working with Agents: Users are encouraged to collaborate with agents by planning tasks together and using concrete instructions. Smaller working sessions and maintaining memory of tasks are advised to maximize effectiveness and minimize cognitive load.
• Challenges and Risks: Issues such as coding errors, security vulnerabilities, and dependency on external MCP servers highlight the need for vigilant supervision and trust in source credibility. Additionally, balancing automation with manual oversight remains crucial to maintaining code quality.
• Future of AI in Coding: While AI increasingly assists in complex tasks, it is not yet fully autonomous. The presentation stresses the role of AI as a supplement to human expertise, requiring continuous evaluation and integration of improvements in workflows.

The talk provides a comprehensive overview of how AI coding assistants have transformed software development workflows and the unforeseen challenges and opportunities they bring. Continuous evolution in AI capabilities mandates active engagement and adaptation from developers.

This is the end of the AI-generated content.

After almost 2 years of intense attention givent to coding assistants, at the beginning of 2025, many organizations were disappointed by the measurable results. The recent arrival of coding assistance agents though has stirred up and reset expectations and the hype.
 

This presentation gives an overview of the tooling and features in AI coding assistants today, and what new workflows they enable. It will particularly focus on the new agentic assistants that have brought us the vibe coding meme. How do they work? Are they replacing developers? What does this do to code quality? This whirlwind tour of the AI tooling landscape is based on Birgitta's experience as Thoughtworks' full time global SME for AI-assisted software delivery.