Essential developments in AI, agents and the infrastructure behind them. What changed. Why it matters.
Google has moved Gemini Computer Use to GA across two APIs, but action execution, confirmation gates and prompt-injection defenses remain responsibilities of the application.
Read the analysisGoogle has moved Gemini Computer Use to GA across two APIs, but action execution, confirmation gates and prompt-injection defenses remain responsibilities of the application.
AI Gateway now serves updated Mistral Large 4 weights and lower prices through the existing model ID, showing why production AI needs regression checks beyond application deploys.
Google now stores Generate Content requests and responses by default in new paid Gemini API projects, making retention and explicit opt-out part of deployment configuration.
Vercel Connect can now create and rotate restricted Stripe keys per agent or task, moving least-privilege payment access out of copied secrets and into the deployment layer.
The production release turns repeated classification and routing into a more practical control layer, with shared-input caching and up to 400 questions per request.
Vercel’s public beta adds containerised pull-based workers alongside an application, but restarting compute does not make the work itself durable.
Google’s new GA model can continue a response far beyond 64K tokens, but its pricing and API semantics make explicit budgets and asynchronous job handling essential.
New decision endpoints separate classification, scoring and routing from open-ended generation.
The Gemini agent announcement brings long-running work, business context and model choice into one experience.
A small lifecycle change matters for the reliability of AI coding and execution workflows.
Anthropic’s new small model puts the economics of routine agent work back in focus.
Try another topic or reset your filters.