Biweekly Tech Radar: AI Agents, Foundation Models, and Infra

The last weeks of November 2025 pushed AI and cloud from "exciting" to "non‑negotiable." New foundation models, powerful agents, and multi‑billion‑dollar infrastructure bets signaled that AI is no longer a siloed add‑on—it is becoming the default interface and compute substrate for modern products.

This biweekly digest highlights the launches and announcements that actually matter if you build, run, or product‑manage technology: GPT‑5.1 and Gemini 3, agentic workflows, major cloud–model partnerships, and a new generation of AI‑native tools. The goal is simple—give you enough context to make decisions, not just more headlines.

1. Foundation Models: GPT‑5.1, Gemini 3, and Thinking Agents

November 2025 cemented a new pattern: frontier models aren't just getting bigger, they're getting more agentic. OpenAI's GPT‑5.1, highlighted in multiple round‑ups, introduced distinct "Instant" and "Thinking" modes—one optimized for speed, the other for longer, tool‑heavy reasoning chains—exposed through ChatGPT and the API.

In parallel, Google announced Gemini 3 and made it available in AI‑mode for Search, bringing state‑of‑the‑art reasoning into an interface billions of users already rely on. Gemini 3 powers richer layouts, interactive tools, and "Thinking with 3 Pro" workflows for paid tiers, moving advanced models from research demos into everyday search behavior.

• Why this matters: As models expose explicit "fast vs deep thinking" modes, teams can architect systems that dynamically trade off latency and deliberation per request rather than picking one model for everything.
• Action for builders: Design APIs and UX so that critical actions (payments, HR decisions, medical assistance) can route to deeper, slower reasoning, while routine queries stick to instant responses.

2. AI Infrastructure: Massive Bets and Consolidation

Under the surface, infrastructure deals are shaping which models you can realistically run at scale. Commentary across November highlighted multi‑billion‑dollar investments such as Microsoft's expanded commitment to Anthropic and broader capacity agreements that tie model access tightly to specific clouds.

At the same time, surveys and round‑ups point to AI becoming a line‑item in enterprise P&L, not an experiment. Reports describe AI adoption exceeding half of working‑age adults in some markets, with corresponding pressure on data centers, GPUs, and energy consumption. This infra layer is now a competitive moat, not a back‑office detail.

• Why this matters: Vendor choice today defines your cost curve, data residency options, and which models you can realistically deploy in production tomorrow.
• Action for builders: Treat "AI infra strategy" like cloud strategy—negotiate portability (open formats, containerized serving, abstraction layers) early instead of discovering lock‑in the hard way.

3. Agents Everywhere: From Browser Help to Group Collaboration

A clear theme in November's updates: agents are leaving the sandbox and entering real workflows. Multiple recaps emphasize AI systems that not only answer questions but also browse, operate tools, and orchestrate multi‑step tasks across tabs, documents, and services.

One notable shift is collaborative and group‑oriented AI usage. Updates described group chat capabilities in ChatGPT, where multiple humans can co‑work inside the same AI‑augmented thread, while the assistant maintains consistent context across the conversation. This changes AI from a solo helper into a shared teammate for product, design, and engineering squads.

• Why this matters: Agentic AI is evolving from "single‑prompt helpers" into workflow engines that can manage multi‑step, cross‑tool processes with human oversight.
• Action for builders: Think in terms of tasks and roles, not just prompts: where can an agent reliably own a workflow segment (triaging tickets, compiling weekly reports, validating logs) with humans just reviewing deltas?

4. Google's November AI Push: Gemini 3, SIMA 2, and AI Shopping

Google's November AI update was dense. Beyond Gemini 3 in Search, the company unveiled SIMA 2—a step toward more embodied, game‑capable agents that can follow natural‑language instructions in virtual environments, reflect on goals, and improve over time. This is framed internally as a milestone on the path toward more general, interactive agents.

Google also leaned hard into AI‑powered commerce, rolling out holiday shopping features that use conversational AI to find products, compare options, and even call stores or perform "agentic checkout" on your behalf when conditions are met. Shopping flows are shifting from page‑by‑page browsing to conversational, outcome‑driven interactions.

• Why this matters: When AI becomes the default shopping layer, product discovery, SEO, and UX patterns change—your "interface" may be an agent's reasoning, not your homepage.
• Action for builders: Make your APIs, product data, and documentation agent‑friendly: clear schemas, up‑to‑date metadata, and robust authentication flows for third‑party agents.

5. Emerging Models and Research: Kimi K2, Healthcare, and Governance

November round‑ups also spotlighted emerging models outside the US Big Three. Beijing's Moonshot AI introduced Kimi K2, a trillion‑parameter "thinking agent" designed for long‑horizon reasoning and 200–300 sequential tool calls, emphasizing extended planning and research‑style workflows.

On the applied side, researchers reported AI systems detecting dementia with over 97% accuracy using EEG signals and explainable AI techniques, while national initiatives like France's AI‑augmented lung cancer screening pilot showed governments moving from pilots to population‑scale deployments in healthcare.

• Why this matters: Advanced reasoning plus domain‑specific data is moving AI from "assistant" to "co‑diagnostician" and "co‑researcher," with direct implications for regulation, ethics, and liability.
• Action for builders: If you work in regulated domains, invest early in auditability: logs, data lineage, and clear separation between retrieval, reasoning, and final recommendations.

6. The Macro View: AI as Infrastructure, Not a Feature

Across industry roundups, surveys, and think‑tank reports, one message repeats: AI is now infrastructure. November summaries describe AI adoption levels exceeding half of adults in some regions, AI permeating security, productivity, and scientific research, and governments racing to update digital governance frameworks to keep up.

This shift mirrors early cloud computing: at first, a competitive differentiator; now, table stakes. The difference is speed—AI is compressing that adoption curve into years, not decades, which raises stakes for talent, governance, and technical debt. The teams that treat AI as a core capability, not a bolt‑on, will compound advantage the fastest.

• Why this matters: "AI strategy" is increasingly indistinguishable from "product strategy" and "infra strategy." Waiting for the space to "settle down" is itself a strategic choice—with opportunity cost.
• Action for builders: Start small but systematic: one or two production‑grade AI use cases, clear KPIs, and a governance framework that can scale. Use these biweekly waves of updates as a trigger to refine—not rewrite—your roadmap.