A week about how AI is changing engineering: reasoning for hard technical problems, real-time coding and a new platform for teams that work with agents.
Gemini 3 Deep Think is updated for science and engineering problems
Google released a major upgrade to Gemini 3 Deep Think, its specialized reasoning mode, developed with scientists for problems with no single answer and incomplete data. In one example, a Rutgers mathematician used the model to review a technical paper and it found a logical flaw that had passed peer review. Deep Think is in the Gemini app for Google AI Ultra subscribers and, for the first time, in early access through the API for selected researchers, engineers and enterprises.
Engineering, R&D and technical analysis teams get a tool to review complex work and explore alternatives. Companies with these challenges can request API access and test it on real cases.
How Dexpert helps: Artificial Intelligence
GPT-5.3-Codex-Spark generates code at more than 1,000 tokens per second
OpenAI released a preview of GPT-5.3-Codex-Spark, a smaller version of GPT-5.3-Codex and the company's first model built for real-time coding. It runs on Cerebras hardware and exceeds 1,000 tokens per second, with a 128k token window and text only. The idea is to complement models that work on their own for hours with one for quick edits, where the developer interrupts and redirects the work on the spot.
Near-instant answers change how people work with AI on interface and logic tweaks, bringing the agent closer to a pair programmer. Teams can combine a fast model for daily work with a stronger one for long tasks.
How Dexpert helps: Artificial IntelligenceSources
- OpenAI: Introducing GPT-5.3-Codex-Spark
Former GitHub CEO founds Entire, a development platform for teams with agents
Entire, a new company from a former GitHub CEO, announced a $60 million seed round to build a development platform where people and AI agents work together. The company argues that issues, repositories and pull requests were designed for collaboration between humans and do not scale to the volume of code agents generate. Its first product is an open source CLI that records the agents' context in Git on every push.
Teams adopting coding agents need to review their process, not only their tools: how to record what the agent did, why it did it and how to review it. Keeping that context next to the code makes audits and maintenance easier.
How Dexpert helps: Dedicated Squads & Staff AugmentationSources
- Entire: Hello Entire World
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.