Tech • IA • Crypto
Code with Claude London 2026: Opening Keynote
8/10TL;DR
AI coding systems are rapidly evolving from assistants into autonomous agents that can execute complex tasks, dramatically accelerating software development and business operations.
KEY POINTS
From hobbyist coding to autonomous systems
Early programming experiences often involved simple tools like calculators and HTML tinkering, where immediate feedback created a sense of “magic.” That gap between idea and execution later widened due to complex tooling. Recent advances in AI are reversing this trend, enabling users to describe problems in plain language and receive working systems in return.
AI agents delivering real-world impact
Companies are already using AI agents to handle large-scale engineering tasks. Spotify has deployed agents that migrate thousands of repositories, generating over 1,000 pull requests per month and reducing migration time by more than 90%. In social services, Binti used AI to cut 20 days from foster care licensing processes, directly accelerating placements for children.
Rapid acceleration in model capabilities
AI model progress has intensified, with major leaps occurring in shorter intervals. Systems now handle multi-step reasoning, long-running tasks, and even identify deeply hidden issues, such as a 27-year-old vulnerability discovered in OpenBSD code. Over the past year alone, eight frontier models have been released, each expanding capabilities like tool use, planning, and long-context reasoning.
Gap between AI potential and adoption
Despite exponential improvements in model performance, most organizations are adopting AI at a linear pace. This creates a widening gap between what AI can achieve and what businesses actually deploy. Developers are positioned as the key bridge, translating raw model capability into usable products and workflows.
Rising usage and developer reliance
AI platform adoption is surging, with API usage increasing by nearly 17×. Developers are integrating AI deeply into workflows, spending over 20 hours per week using coding agents. These tools are no longer occasional aids but core components of daily engineering work.
Shift toward autonomous, long-running agents
A key metric, “task horizon,” has expanded from minutes to hours of continuous autonomous work. Future systems are expected to run persistently, proactively managing goals such as maintaining project timelines or updating financial forecasts without explicit prompts.
New architecture strategies for efficiency
Techniques like the “advisor strategy” split tasks between smaller, cheaper models and more powerful ones. This approach can deliver high-end performance at significantly reduced cost, with some companies reporting fivefold cost savings while maintaining quality.
Enterprise-grade agent infrastructure
New platforms provide tools to build and scale AI agents securely. Features such as self-hosted sandboxes and MCP tunnels allow organizations to run AI workloads within private infrastructure while maintaining secure access to internal systems. This enables faster deployment without sacrificing control or compliance.
AI-native workflows and proactive automation
AI agents are increasingly proactive, monitoring systems, analyzing data, and taking action without human initiation. Examples include agents that evaluate experiments, update product features, generate code changes, and open pull requests automatically.
Transformation of software development practices
Development is shifting from synchronous coding to asynchronous orchestration of multiple agents. Tools now verify their own outputs, fix bugs, handle code reviews, and resolve CI failures automatically. Features like autofix, routines, and automated security scanning reduce manual intervention.
Enterprise-wide adoption and productivity gains
Large organizations are deploying AI across entire engineering teams. Shopify integrates AI into workflows for engineers and non-engineers alike, while Mercado Libre, with 23,000 engineers, has processed over 500,000 pull requests and modernized 9,000 applications using AI-driven systems. Some teams are targeting 90% autonomous coding.
Emergence of multi-agent collaboration
Future systems will coordinate multiple agents working together on complex goals. This includes collaborative planning, execution, and continuous improvement, enabling tasks too large for a single model to handle effectively.
CONCLUSION
AI coding tools are rapidly evolving into autonomous systems that reshape how software is built, shifting development toward continuous, agent-driven workflows and redefining productivity across industries.