Evaluating context compression in AI agents

Explore all

GitHub gives Copilot better memory

GitHub is bringing persistent memory to Copilot, allowing it to retain repository-level context across sessions. This enhances code suggestions and reviews by building on accumulated developer interactions over time.

Cursor acquires Graphite to bridge code creation and review

Cursor’s acquisition of Graphite aims to unify code creation and review, and in the process brings the company closer to territory long dominated by GitHub.

A year in review: From vibe coding to viable code

2025 was the year AI-assisted coding grew up. What started as a wave of "vibe coding" — rapid prototyping, prompt-driven experiments, and disposable applications — matured into something more disciplined: spec-driven development, production-grade agents, and an ecosystem that's finally grappling with real-world constraints. At AI Native Dev, we've had a front-row seat. Through 140+ articles, 88 podcast episodes, and two DevCon events, we've tracked every twist in this fast-moving space. Here's our look back at the year that changed how software gets built.

AI-Native Development

With Gemini 3 Flash, Google gives developers a low-cost model ‘built for speed’

Google's Gemini 3 Flash promises a more cost-effective, high-throughput language model optimized for low-latency use-cases.

8 min read

23 Dec 2025

AI Tools & Assistants

AI-Native Development

As AI coding agents mature, metrics follow a familiar enterprise path

As AI coding agents land deeper in development workflows, organizations seek metrics to assess usage, prompting vendors like GitHub and Continue to enhance visibility into tool performance and adoption.

Forever stateful: Letta Code bets on memory as the missing layer in coding agents

Letta Code is a memory-driven coding agent that retains context over time, challenging stateless models and offering developers a flexible, open source solution to enhance productivity and reliability in coding tasks.

AI Tools & Assistants

Amp launches Thread Map to help navigate AI coding agent work

Amp’s Thread Map visualizes what AI coding agents are doing and how their work connects, helping keep context intact and making parallel tasks easier to manage by grouping scattered conversations into clear units of work.

Paul Sawers

5 min read

17 Dec 2025

AI Tools & Assistants

AI Coding Tools

Claude Code comes to Slack as team chat and coding converge

Anthropic is rolling out an integration between Claude Code and Slack, allowing developers to trigger coding tasks directly from team conversations.

Paul Sawers

8 min read

16 Dec 2025

AI Tools & Assistants

Developer Experience

Testing and Securing AI-Generated Code with Cursor

Explore Cursor's innovative context engineering and rule system, designed to enhance the reliability and security of AI-generated code, maximizing development speed without compromising quality in production environments.

Mistral debuts Vibe CLI agent and open-weight Devstral 2 models for enterprise-grade coding

Mistral, the French frontier AI model lab most recently valued at €11.7 billion, has launched a duo of open-weight coding models alongside a new terminal-based open source agent.

Paul Sawers

7 min read

11 Dec 2025

AI Tools & Assistants

Vibe Coding

Inside Kilo Code: An open source AI coding agent with plans to reshape software development

Launched in March, 2025, Kilo Code is an open source coding agent that can be configured for a range of tasks — from generating new code and refactoring multi-file changes, to debugging, writing documentation, running terminal commands, or orchestrating longer workflows across multiple agents.

AI Tools & Assistants

These Aren't the Tools You're Looking For: The Hidden Dangers of MCP

Discover the hidden security risks of Model Context Protocol (MCP) servers, as explosive adoption uncovers vulnerabilities that could turn AI tools into insider threats for developers.