This week: what works to give context to coding agents, a large modernization case with Rust and a new open model focused on agents.
A docs index in AGENTS.md scored 100% in Vercel's tests
Vercel tested two ways to teach coding agents the new Next.js 16 APIs, which are not yet in the models' training data. A compact 8 KB documentation index placed in the AGENTS.md file reached a 100% pass rate, while skills topped out at 79%. In 56% of cases the agent never invoked the skill.
Teams that use coding agents every day gain more by putting essential context, such as versions, patterns and project rules, in a file the agent always reads. It is a cheap change that cuts wrong code and rework.
How Dexpert helps: Artificial IntelligenceSources
WhatsApp ships its media library rewritten in Rust to billions of devices
Meta says WhatsApp now uses Rust for media file checks, a library that automatically inspects every image and video received to block malformed files. The original C++ library was rewritten in Rust, a memory-safe language, and shipped to apps and browsers. According to the company, it is the largest rollout of a Rust library in the world.
The case shows that rewriting the critical parts of a system, not the whole thing, is a viable modernization path. Components that process external data, such as uploads and integrations, are good places to start.
How Dexpert helps: Legacy System ModernizationSources
- Engineering at Meta: Rust at Scale: An Added Layer of Security for WhatsApp
Kimi K2.5 coordinates up to 100 sub-agents in parallel
Moonshot AI released Kimi K2.5, an open multimodal model trained on about 15 trillion text and image tokens. For complex tasks, it creates on its own a swarm of up to 100 sub-agents working in parallel with up to 1,500 tool calls, which, according to the company, cuts execution time by up to 4.5 times. Moonshot highlights its coding performance, especially for front-end interfaces, and code generation from images and video.
Open models at this level widen the options for companies that want to control costs or host AI on their own infrastructure. Test them on real business tasks before choosing a provider.
How Dexpert helps: Artificial IntelligenceSources
Want to put one of these to work in your company?
Dexpert builds custom systems, apps, integrations and AI solutions. Tell us what you need and get an estimate, or take the free digital maturity assessment.