Field Deployment Engineers (FDEs) are among the hottest new roles in the AI era, serving as both engineers and consultants who write production code directly in client environments to solve the last-mile challenges of AI deployment. First created by Palantir in the 2010s, companies like OpenAI and Anthropic now offer mid-to-senior FDEs annual salaries of $350,000 to $1.2 million, while domestic firms like ByteDance and Zhipu AI provide monthly pay ranging from 35,000 to 80,000 RMB. Becoming an FDE requires deep technical skills, client communication abilities, and problem-solving in ambiguous situations, with open-ended case studies being a key part of the interview process.
Latest insights
Insights come from up to date trends.
Sakana AI's Fugu system tops benchmarks, but its core is a multi-model orchestration layer, not a foundational model. Without owning the base model, the company lacks a durable moat—pricing power depends on underlying model costs, and orchestration alone is a temporary advantage. For indie devs and investors, the real value lies in Japan-specific applications, not in competing with US/Chinese frontier labs. The video warns against mistaking benchmark scores for long-term competitiveness.
Midjourney's founder announced plans for a medical imaging device called Fullbody Ultrasonic Computational Tomography (FUCT), claiming it could be faster and cheaper than an MRI. The video explores how the technology might work using ultrasound and computational imaging, but expresses skepticism about whether it will actually succeed given the challenges in medical hardware. It's a thought-provoking look at how AI companies are branching into unexpected domains.
Big tech companies like Google, Microsoft, OpenAI, and Anthropic are investing heavily in Brazil—opening offices, building infrastructure, and focusing on AI. This raises a critical question for Brazilian developers and founders: is this a historic opportunity to launch a new generation of AI startups, or does it deepen dependency on foreign tech giants? The video analyzes what these moves mean for the local tech market, weighing the potential for innovation against the risk of increased reliance on external players.
Notion Workers represent a major shift for Notion users, enabling external data integration, outbound data pushes, and complex business logic without requiring coding skills. With AI handling the code generation, even non-technical users can build custom automations that previously needed separate no-code tools. This tutorial walks through three types of Workers—webhook, sync, and agent tool—making it practical for anyone looking to extend Notion's capabilities.
AI coding has been rapidly commercialized because it replaces high-salary, high-ROI roles. The next area likely to attract capital is AI research, where talent is even more expensive and automation could yield greater returns. From new drugs and materials to novel training methods, the potential of AI research is astronomical—but so are the risks. Unlike large companies hedging across multiple fronts, RSI differentiates itself by focusing narrowly: not on having AI optimize a piece of code, but on teaching AI to optimize itself.
Google DeepMind's Gemma 4 model enables running 10 independent subagents in parallel on a single local machine, achieving inference speeds above 170 tokens per second. An orchestrator agent delegates tasks from a high-level prompt, and the subagents autonomously generate a complete SVG art gallery within seconds. This demonstration highlights the potential of parallel local agents for enterprise workflows and private on-prem deployments, offering a glimpse into scalable, private multi-agent systems.
Getting your first 10 customers requires personal, unscalable effort rather than relying on automation tools. Founders should start by working their warm network, showing up in person where buyers spend time, and engaging in online communities where customers complain. Outreach should be framed as advice-seeking, and messages must sound human to stand out.
An AI lab with no product yet raised $650 million at a $4.65 billion valuation, driven by investor belief in the team's judgment and execution ability. The lab, RSI, focuses on automated AI research rather than building a short-term monetizable application, betting on a self-evolving paradigm. This reflects a broader trend where top capital prioritizes people over product in fast-changing AI fields.
Claude Code is a powerful AI coding assistant that integrates directly into the terminal and VS Code, enabling developers to refactor code, manage project context, and automate tasks through natural language prompts. The crash course covers installation, modes (plan, shell), settings, and advanced features like MCPs and sub-agents, making it a practical guide for developers looking to boost productivity with AI. With support for global sessions and a Claude.md file for project instructions, it offers a structured way to leverage AI in real-world projects.
The leaked 120,000-character system prompt for Anthropic's Fable 5 reveals far more than just a model's behavior—it exposes a sophisticated AI workbench integrating agent harness, file system, tool calling, billing, and risk control. Directly copying the prompt yields only surface-level imitation, as the true value lies in the server-side infrastructure. Chinese AI models lag not in benchmark scores but in building this complete workbench ecosystem, which Elon Musk highlights as the real gap.
As companies begin to systematically extract and codify employee knowledge—essentially "distilling" workers—individual value in the AI era faces a fundamental challenge. Former Meta FAIR research director Yuandong Tian notes that while this trend makes business sense, it will intensify the conflict between employees and employers, leading major tech firms to significantly reduce their workforce within the next 3 to 5 years. The real solution isn't jumping to another "dried-up puddle," but redefining one's unique position within the AI ecosystem.
GLM-5.2, an open-source model with MIT license, has achieved coding benchmark scores that rival or surpass top closed-source models like GPT-5.5 and Claude Opus 4.8. Vercel's CEO publicly expressed being 'almost shocked' by its coding ability, signaling a major shift in the AI landscape. The model's combination of strong performance, open licensing, and local deployability makes it a compelling alternative for developers and enterprises looking to reduce costs without sacrificing quality.
Google DeepMind has seen two top researchers leave recently, raising concerns about the company's AI talent retention. The departures stem from a mix of financial incentives that failed to lock in key people, internal organizational friction, and competing AI model strategies like Gemini versus Antigravity. While Google faces real pressure in the AI race, its cash reserves, user base, and long-term research infrastructure suggest it is far from doomed.
The podcast challenges the popular narrative of rich versus poor, reframing the divide as between 'makers'—those who create value through innovation and productivity—and 'takers'—those who extract value without contributing. The hosts argue that policy should reward makers to drive economic growth, rather than pitting income groups against each other. This perspective is crucial for indie developers and AI builders who identify as makers, as it underscores the importance of protecting and incentivizing value creation.
OpenAI Codex 0.141 introduces several notable features: a Record & Replay system on macOS that lets users capture a sequence of operations into a reusable skill, and end-to-end encrypted remote execution via Noise relay, preserving native working directories and shell environments across platforms. The update also brings a per-thread MCP server activation system, a curated plugin marketplace, Computer Use support for Windows, and a preview of Sites for building internal tools hosted by OpenAI. Performance improvements include caching for tool search in large sessions, reducing latency and memory usage.
DeepSeek completed a 51 billion RMB funding round, with Liang Wenfeng securing firm control through a limited partnership structure, a five-year lock-up period, LP look-through review, and anti-poaching clauses. The National AI Industry Fund received special treatment including veto power, while other investors accepted strict restrictions. This all-RMB financing paves the way for DeepSeek's computing center construction and commercialization path—including the Harness Agent team and an A-share listing—while also reflecting new trends in China's AI competitive landscape and the monetization model for open-source models.
This week's GitHub highlights feature five notable open-source projects. agent-skills provides production-grade engineering agent capabilities, while Apple's container tool offers a new approach to containerization. NVIDIA's Cosmos platform enables physical AI world modeling, and Chatwoot delivers a full-featured open-source customer support platform. Additionally, the IPTV-org project aggregates global public IPTV playlists for easy access.
Zhipu AI's open-source GLM-5.2 model has a total of 753B parameters (with approximately 40B activated per inference) and supports a 1-million-token context, outperforming competitors like Claude Opus 4.8 and DeepSeek V4 Pro on programming and general intelligence benchmarks. Its core innovation, IndexShare, uses a cross-layer shared sparse attention mechanism to significantly reduce long-context inference costs, making 1M-token reasoning economically viable. Despite outputting around 43k tokens per task, its pricing of just $1.4/$4.4 per million tokens on OpenRouter is highly competitive, positioning it as a new benchmark in the open-source community.
The All-In Podcast discusses the US government's ban on Anthropic's Fable model over Chinese access concerns, highlighting growing tensions between AI safety and national security. They also cover SpaceX's record-breaking IPO and a $60B Cursor acquisition, fueling trillionaire predictions. The episode examines the rise of tech oligarchs and their political influence, alongside a potential Iran peace deal's market impact.