New technology shows up every week, and some of it becomes an edge for those who move first. Dexpert Radar follows what changes in AI, software, apps, data and cloud, and shows how to use each novelty to sell more, cut costs and innovate.
GLM-5.2 takes the lead among open models, the IETF publishes the HTTP QUERY method and Browser Use runs Firecracker VMs inside EC2 to cut costs.
Apple rebuilds Apple Intelligence on models made with Google, Apple's container project gains Linux environments integrated with the Mac and PgDog raises $5.5M.
OpenAI models and Codex arrive on Amazon Bedrock, the company behind Vite joins Cloudflare and Microsoft open-sources pg_durable for workflows inside Postgres.
Opus 4.8 flags its own uncertainties more often, DBOS argues for Postgres as a workflow orchestrator and Cloudflare launches a feature flag service.
Google releases Gemini 3.5 Flash for agents and coding, Anthropic buys SDK and MCP server maker Stainless and Deno 2.8 brings audit fix, ci and pack.
Anthropic brings Claude to QuickBooks, HubSpot and other small business tools, DuckDB gains a client-server mode and Codex arrives in the ChatGPT app.
Multi-token prediction drafters speed up Gemma 4, agents create accounts, buy domains and deploy on Cloudflare and OpenAI explains its WebRTC architecture for voice.
GitHub Copilot swaps premium requests for token-based credits, the Warp terminal opens its code with agents and Mistral releases an open 128B model.
OpenAI releases GPT-5.5 for long tasks, DeepSeek opens V4 with a 1 million token context by default and Google presents separate chips for training and inference.
Opus 4.7 takes on harder coding tasks, Google launches a CLI that makes agents 3 times faster on Android and Cloudflare opens its email service.