This week: AI operating legacy systems the way a person would, the future of local AI and an official tool to update Go code automatically.
Claude Sonnet 4.6 advances in computer use and reaches systems without APIs
Anthropic released Claude Sonnet 4.6, with improvements in coding, computer use, long-context reasoning and office tasks, and a 1 million token window in beta. Pricing stays the same as Sonnet 4.5, starting at $3 and $15 per million tokens. The company highlights that the model operates software by clicking and typing like a person, which makes it possible to automate older software without APIs, such as navigating a complex spreadsheet or filling out a multi-step web form.
Legacy systems without integrations stop blocking automation, since AI can operate their own interface. For critical processes, API integration is still more reliable, and computer use can serve as a bridge during modernization.
How Dexpert helps: Legacy System ModernizationSources
- Anthropic: Introducing Sonnet 4.6
The founding team of llama.cpp joins Hugging Face
ggml.ai, the team that created the ggml library and llama.cpp, announced it has joined Hugging Face. The projects stay open and community-driven, and the team keeps maintaining the libraries full-time. The announcement says llama.cpp has become a building block of countless projects that run AI on consumer hardware, and that there will be a focus on ease of use and better integration with the transformers library.
Long-term backing gives more confidence to those betting on running models locally, with data that never leaves the company and no per-call cost. It is a solid base for internal assistants and applications with privacy requirements.
How Dexpert helps: Artificial Intelligence
Go 1.26 ships a rewritten go fix to modernize code automatically
Go 1.26 includes a completely rewritten go fix command, which finds code that can use more modern language and standard library features and applies the changes itself. The Go team recommends running it on every Go version upgrade, starting from a clean Git state, and using the -diff flag to review first. You can also enable only some analyzers to split large changes into smaller reviews.
Keeping code up to date gets cheaper when the language itself does part of the work. Companies with Go services can add go fix to their upgrade process and pay down technical debt gradually.
How Dexpert helps: Backend & APIsSources
- The Go Blog: Using go fix to modernize Go code
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.