The Tech newsletter for Engineers who want to stay ahead

Tech moves fast, but you're still playing catch-up?

That's exactly why 100K+ engineers working at Google, Meta, and Apple read The Code twice a week.

Here's what you get:

• Curated tech news that shapes your career - Filtered from thousands of sources so you know what's coming 6 months early.

• Practical resources you can use immediately - Real tutorials and tools that solve actual engineering problems.

• Research papers and insights decoded - We break down complex tech so you understand what matters.

All delivered twice a week in just 2 short emails.

Join 100K+ engineers

Today:

• Self-Coding AI Matches Human Designers

• Claude Just Got Smarter for Finance Teams

• Qualcomm’s New AI Chips Take Aim at Nvidia & AMD

• MiniMax-M2 Outshines Rivals in Agentic AI

• Elon Musk’s Grokipedia Takes Aim at Wikipedia

Self Improving AI is getting wild

Researchers are moving toward self-improving AI, programs that rewrite their own code to get better at tasks. Earlier work used “family trees” of agent versions, keeping only short-term winners. A new approach, the Huxley Gödel Machine, estimates a branch’s long-term potential using “clade metaproductivity”. This lets the system explore, evaluate less often, save time, and pick promising paths. 

On coding tests, it matched top human-designed agents and transferred well to other models. The result hints at faster overall AI improvement.

WATCH THE VIDEO ON YOUTUBE

Claude Just Got Smarter for Finance Teams

Anthropic expands Claude for Financial Services with Excel integration, real-time market data connectors, and prebuilt Agent Skills. These upgrades help users automate complex financial tasks like valuation models and due diligence. Claude now fits better into industry tools, speeding up analysis and improving accuracy.

KEY POINTS

• Claude for Excel: New sidebar tool lets Claude read, analyze, and edit spreadsheets with full transparency; now in beta for select users.

• Real-time Connectors: Claude gains direct access to S&P, Morningstar, Moody’s, LSEG, and more for live data and investor insights.

• New Agent Skills: Claude can now handle tasks like DCF models, comps analysis, earnings reviews, and due diligence summaries.

Why it matters

These updates make Claude a much more useful assistant for finance professionals. Instead of jumping between tools, users can get help right inside Excel or pull live data instantly. It saves time on hard tasks like modeling and analysis, which usually take hours.

Qualcomm’s New AI Chips Take Aim at Nvidia & AMD

Qualcomm unveiled new AI accelerator chips for data centers, targeting AI inference workloads. This marks its entry into a market long dominated by Nvidia and AMD. The announcement pushed Qualcomm stock up 11% as it promises lower power use, high memory, and competitive pricing.

KEY POINTS

• New AI Chips: Qualcomm introduces AI200 and AI250, rack-scale chips focused on inference, not training, launching in 2026 and 2027.

• Nvidia & AMD Rivalry: Qualcomm joins the race in the booming $6.7 trillion AI data center market, competing on cost, energy, and memory efficiency.

• Big Clients & Ambitions: Already partnered with Saudi Arabia’s Humain, with support for 768GB memory per card and 160kW rack systems.
  

Why it matters

Qualcomm’s move into AI server chips gives companies more options beyond Nvidia and AMD. That competition could lower prices and energy use for data centers. It’s a big shift for Qualcomm and a sign that AI is reshaping the entire tech industry.

MiniMax-M2 Outshines Rivals in Agentic AI

MiniMax-M2, a Chinese open-source AI model, now leads agentic tool use and coding benchmarks. With strong performance, low deployment cost, and an MIT license, it challenges proprietary giants like GPT-5 and Claude 4.5—making it ideal for enterprise automation, developer agents, and reasoning-heavy tasks.

KEY POINTS

• Benchmark Leader: MiniMax-M2 tops open-weight LLMs in agentic tasks like tool use, coding, and reasoning—ranking just below GPT-5 and Claude Sonnet 4.5.

• Enterprise-Friendly: Offers low-cost deployment ($0.30 input / $1.20 output per million tokens), sparse MoE design, and open MIT license.

• Built for Agents: Excels at planning, executing, and verifying workflows with structured tool-calling, interleaved reasoning, and long context retention.

Why it matters

MiniMax-M2 gives businesses a powerful, affordable alternative to expensive proprietary models. It helps teams build smart AI agents that use tools, write code, and make decisions—all while being transparent, customizable, and open. This opens doors for wider, cheaper AI use in real-world systems.

🧠RESEARCH

DeepAgent: A General Reasoning Agent with Scalable Toolsets

DeepAgent is a new AI system that thinks, finds tools, and takes actions on its own. It avoids memory overload by compressing past steps and learns tool use through a reward system. Tested on eight tasks, it beats older methods, showing promise for smarter, more independent AI in real-world use.

Concerto: Joint 2D-3D Self-Supervised Learning Emerges Spatial Representations

Concerto is a simple AI model that learns about 3D spaces by combining 2D and 3D views, like how people use multiple senses. It outperforms top models in scene understanding and works well even without extra training. It also links visuals to language, helping AI understand open-world environments more accurately.

Video-As-Prompt: Unified Semantic Control for Video Generation

Video-As-Prompt (VAP) is a new method that uses a reference video to guide AI in generating new videos with clear meaning and control. It avoids common flaws of earlier models, works without extra training, and sets a new performance standard—making it a breakthrough for flexible, high-quality video creation.

🛠️TOP TOOLS

Each listing includes a hands-on tutorial so you can get started right away, whether you’re a beginner or a pro.

LTX Studio – AI Video Generator - AI studio for video production

Vondy AI – AI App Marketplace - AI-powered generators and prebuilt assistants behind a single interface

Writeless AI – AI Essay Writer / Education - AI‑powered writing tool built specifically for academic essays and research papers

📲SOCIAL MEDIA

🗞️MORE NEWS

• Elon Musk launched Grokipedia, a computer-made intelligence encyclopedia challenging Wikipedia. With 800,000 machine-written entries and politics-tinged pages, it sparked worries about information control. Wikipedia emphasizes neutrality while facing fewer human readers and more automatic copying.

• OpenAI urges building new power for artificial intelligence, warning U.S. electricity trails China. Proposals: faster permits, more energy, worker training. $1 trillion in projects could raise economic output 5%. Urges building 100 billion watts yearly.

• Amazon will cut up to 30,000 corporate jobs—about 10% of staff—undoing pandemic hiring. Cuts hit human resources, devices, and operations. CEO Andy Jassy says AI tools will reduce roles. Shares rose. Amazon’s largest layoff ever.

• A judge let authors’ lawsuit against OpenAI proceed, finding ChatGPT summaries and outlines may copy parts of books, including George R.R. Martin’s. The decision rejects OpenAI’s bid to dismiss; whether law allows use remains undecided.

• U.S. Energy Department and AMD to build two very powerful computers for $1 billion. Lux in six months; Discovery in 2029. Using AI-focused chips, they’ll speed fusion, cancer, and security research. Partners will share capacity.

• Mercor, a 2023 startup hiring people to teach chatbots, is now valued at $10 billion. It’s raising $350 million led by Felicis. With 30,000 workers and clients like OpenAI, its worth jumped fivefold since February.

• Vishal Shah appointed to lead Meta's AI team, focusing on flagship AI products like the Meta AI app. Meta's recent shake-up includes layoffs and reshuffles to speed up AI product deployment. Challenges like rushed AI video service releases and competition from OpenAI drive Meta's urgency in AI development.

• Foxconn approved up to NT$42 billion ($1.37 billion) to buy equipment for a compute cluster  and a supercomputing center, Dec 2025–Dec 2026, expanding cloud services and platforms, likely in Taiwan.

• Johns Hopkins tested a computer-guided app for adults with above-normal blood sugar. In a randomized test, it matched human coaching on weight, activity, and blood sugar targets, with higher completion rates too—suggesting easier, always-available care.