AI Rundown: Brin's DeepMind Strike Team & Moonshot K2.6

Morning AI Rundown: Brin Mobilizes DeepMind, Moonshot's K2.6, Adobe's Agentic Platform, and More

Good morning, {{ first_name | AI enthusiasts }}. Today's briefing covers major moves across AI research, open-source model releases, design tools, enterprise platforms, and practical workflows. We also include brief notes on smaller launches and community use cases — plus a quick nod to gaming news where applicable.

Top Story: Sergey Brin rallies DeepMind to close Gemini's coding gap

Google co-founder Sergey Brin has reportedly formed a DeepMind "strike team" to accelerate Gemini's code-writing abilities and close the internal gap with Anthropic's Claude. Brin frames coding as the fastest path to systems that can train the next generation of AI — the so-called holy grail of self-improving models.

"The real prize is AI that trains the next AI, and coding is the capability that gets us there." — internal memo attributed to Sergey Brin

Key details:

• Research engineer Sebastian Borgeaud leads the new team under CTO Koray Kavukcuoglu and Brin's oversight.
• DeepMind researchers reportedly rate Claude's code-writing higher than Gemini's, prompting the dedicated push.
• Gemini engineers must now use internal agent tools for complex tasks; usage is tracked on an internal leaderboard called Jetski.

Why it matters: This is an internal push aimed at automating Google's systems, not an immediate product release. Closing the coding gap could enable more powerful agentic systems inside Google to rival Anthropic and OpenAI — a strategic move that could reshape enterprise tooling and, indirectly, areas like gaming news coverage where automated content generation and toolchains are increasingly used.

Together with You.com: Is your API latency metric lying to you?

You.com's guide warns that common latency benchmarks (like p50) can hide real user failures. The guide recommends measuring "time-to-useful-result" and evaluating APIs at your actual concurrency levels. It also highlights four hidden cost drivers visible only in production logs.

What you'll learn:

• Why p50 latency can mislead product decisions.
• How to adopt a "time-to-useful-result" framework.
• Which hidden costs to watch in logs vs vendor tables.
• How to benchmark APIs under realistic concurrency.

Moonshot AI: Kimi K2.6 narrows the open-source gap

Moonshot AI open-sourced K2.6, a new agentic coding model that approaches or exceeds performance from models like GPT-5.4, Opus 4.6, and Gemini 3.1 Pro on several benchmarks. K2.6 targets long-horizon, agentic workflows at low cost.

Highlights:

• K2.6 outperforms competitors on reasoning benchmarks such as Humanity's Last Exam with tools, and on coding tests like SWE-Bench Pro.
• Demonstrated endurance: K2.6 can run for 12+ hours and handle 4,000+ tool calls in long-horizon tasks.
• Agent scale: Kimi's swarms can spawn 300 parallel sub-agents, triple K2.5's concurrency.

Why it matters: While frontier labs claim open-source and China lag by months, public releases like K2.6 show the gap narrowing. For teams exploring agentic automation — in software dev, customer support, or even automated content pipelines for gaming news portals — K2.6 offers a cost-effective alternative.

AI Training: Build high-converting landing pages with Claude Design

Anthropic's Claude Design helps generate multiple landing page mockups quickly. The tool is useful for marketers, product teams, and developers who want fast iterations and direct handoff to code generation.

1. Visit claude.ai/design and select the wireframe option.
2. Provide who the page is for, what you're selling, and the desired visitor action.
3. Add screenshots of landing pages you like as references.
4. Ask Claude to create four mockup variations and answer follow-up clarifying prompts.
5. Click elements to leave notes like "rewrite this CTA to be outcome-specific." Claude will apply refinements.

Pro tip: Use Share > Handoff to Claude Code > Send to Claude Code Web to have Claude build and deploy the final site.

Adobe: CX Enterprise — an agentic platform for enterprises

At Adobe Summit, Adobe introduced CX Enterprise, an agentic orchestration platform for marketing, content, and customer workflows. The platform coordinates networks of AI agents to plan and execute multi-step goals.

Platform components:

• Brand visibility orchestration
• Content supply chain management
• Customer engagement coordination

Features include a Coworker that assembles the right agents and an agent skills catalog for reusable workflows. The Marketing Agent can interface with ChatGPT, Claude, Gemini, and Copilot to coordinate across tools.

Why it matters: Design and marketing workflows are rapidly moving toward agentic automation. Platforms from Adobe, Figma, and Canva compete to provide orchestration layers, while improvements in direct design tools (like Claude Design) threaten to bypass legacy systems.

Trending AI Tools

• Scrunch — AI site interpretation and free audits for customer reach.
• Kimi K2.6 — Moonshot AI's open-source coding and agent model.
• Chronicle — OpenAI's Codex preview that uses screen context for persistent memory.
• Qwen3.6-Plus — Alibaba's model with a 1M token context window and strong coding performance.

Other notable updates

• OpenAI previewed Chronicle, a background-agent feature that builds persistent memories from screen context (initially Pro on Mac).
• Yann LeCun weighed in on labor-market impact debates, urging economists be central to discussions.
• Lovable denied a reported data breach, attributing the issue to documentation mistakes that made some project chats visible.
• Tinder and Zoom partnered with World to let users add "proof of humanity" badges via iris scans to reduce bots and deepfakes.
• Anthropic expanded an AWS compute deal with up to $25B investment for 5 GW of capacity.
• Recursive Superintelligence raised $500M at a $4B valuation to pursue self-improving AI research.

Community Workflow: AI in everyday marketing

Reader spotlight: Julie N. from Grand Junction, CO uses ChatGPT as a digital marketing assistant. She feeds client budgets, landing pages, and goals to generate keyword-appropriate headlines, ad copy, strategies, and even package PDFs for clients.

How do you use AI? Share your workflow and we may feature it in a future issue.

See you soon,

Rowan, Joey, Zach, Shubham, and Jennifer — the humans behind The Rundown