The FDE Coach blog · new articles every day
Practical guides for engineers who ship AI.
Step-by-step builds with free AI tools, what just shipped in AI and why it matters, and Forward Deployed Engineer playbooks on interviews, pay and client work.
What just shipped in AI, and why it matters for engineers.
Page 3 of 11
AI News
GPU Memory Reads: Latency, Coalescing & Why It Matters for Engineers
A warp stalls for hundreds of cycles waiting on global memory. Here's the electrical and logical reality of a GPU memory read, how coalescing works, and why Forward Deployed Engineers must care.
9 min read
AI News
AI-Boosted Homework, Crashing Exams: The Copilot Trap Engineers Must Avoid
A study reveals a brutal truth: AI tools boost homework scores but cause exam grades to plummet. Here's the cognitive science breakdown for engineers and why FDEs must master 'desirable difficulty' to avoid becoming AI-dependent operators.
8 min read
AI News
Ornith-1.5: How Self-Scaffolding Builds Self-Improving Small Language Models
Ornith-1.5 shows how small language models can bootstrap their own training data without GPT-4. We break down the self-scaffolding pipeline, why it matters for on-device AI, and how FDEs can run it today.
8 min read
AI News
AI Didn't Erase the Junior Engineer—It Made Them a Force Multiplier
The junior engineer is not obsolete. AI tooling has transformed entry-level roles from ticket-takers into high-leverage force multipliers. Here's why the FDE career path proves junior talent is more valuable, not less.
9 min read
AI News
Cerebras CS-4: Why Memory Bandwidth Is the New FLOPS
The Cerebras CS-4 packs 1.2T transistors on a single wafer. Here's why its 21 PB/s memory bandwidth matters more than raw FLOPS for AI inference, and what it means for engineers shipping real products.
9 min read
AI News
Malicious Rust Crate Arrayref: How Build-Time Payloads Evade Code Review
The Arrayref incident reveals how build.rs macros bypass code review. Engineers must treat build scripts as attack surface, not boilerplate. Here's the technical breakdown and defense.
9 min read
AI News
LLM-Extensible Software: Why Static APIs Are Dead
LLMs aren't just chat interfaces—they're a new runtime. Learn how to design software where the model writes the glue code, turning rigid APIs into fluid execution environments.
10 min read
AI News
Linear's AI Usage Data: The Real Adoption Curves and Engineering Bottlenecks
Linear's internal data reveals how software teams actually use AI—not the hype, but the messy reality of adoption curves, language gaps, and the 30% task cliff.
10 min read
AI News
GLM-5.3 Benchmarks: Where the 130B Open-Weight Model Lands Against Llama and Qwen
We dissect GLM-5.3's Artificial Analysis benchmarks. See how the 130B open-weight model stacks up against Llama 3 and Qwen 2.5 on quality, speed, and price for engineering tasks.
7 min read
AI News
Mojo Goes Open Source Under Qualcomm: What Changes for Python-Level Systems Programming
Mojo is now open source under Qualcomm stewardship. This changes the calculus for engineers who need Python ergonomics with C speed. Here's what the license shift means, how to try it today, and where it fits in your stack.
9 min read
AI News
Unsloth Dynamic 3.0 GGUFs: Per-Token Mixed Precision for Dense and MoE Models
Unsloth Dynamic 3.0 GGUFs bring per-token mixed precision quantization to GGUF, slashing VRAM for MoE and dense models without retraining. Here's how it works and how to run it.
10 min read
AI News
Claude Writes a macOS Driver from a Windows Binary: What It Means for Driver Dev
A viral demo shows Claude reverse-engineering a Windows binary to build a macOS printer driver. We dissect the technical reality, why it matters for forward-deployed engineers, and how to try similar workflows today.
10 min read
AI News
When an LLM Only Sees Grade-School Text: The Little Learner Experiment
What happens when you train a transformer on nothing but children's stories? The Little Learner experiment reveals surprising generalization, brittle failure modes, and what it means for your training data.
10 min read
AI News
Turbovec: Google's TurboQuant Lands in Rust with 4x Vector Search Speedups
Turbovec ports Google's TurboQuant scalar quantization to pure Rust, delivering 4x faster vector search with minimal recall loss. Here's how it works and why it matters for RAG pipelines.
10 min read
AI News
Anthropic Maps Multi-Agent Chaos: The Coordination Patterns That Break in Production
Anthropic's new research maps the failure modes of multi-agent systems, from goal drift to infinite loops. Here's why FDEs building production pipelines need to move from naive chaining to structured coordination.
8 min read
AI News
GPU Offload in Rust: Write Portable Kernels Without Sacrificing Safety or Speed
Rust GPU offloading is real and production-ready. Learn how to write portable, safe kernels using Rust-native tooling, bypassing CUDA C++ footguns without losing a cycle of performance.
8 min read
AI News
Qwen3.8 27B Scores 52: The New Local Reasoning King for Engineers
Qwen3.8 27B hits a 52 on Artificial Analysis, rivaling Llama 4 Maverick. We break down why this open-weight model is a breakthrough for local reasoning, how to run it on consumer hardware, and what it means for Forward Deployed Engineers.
10 min read
AI News
Cursor Launches Origin: Why AI-First Development Now Demands Its Own Hosting Layer
Cursor Origin isn't just another GitHub clone. It's a hosting layer purpose-built for an agentic coding workflow where the AI is a first-class collaborator, not a CLI sidekick.
9 min read
AI News
GPT-5.6 Sol Vision: Real-World Accuracy Gains Beyond Benchmarks
GPT-5.6 Sol isn't just topping vision benchmarks—it's changing how engineers build reliable visual AI pipelines. We unpack the real-world accuracy gains, trade-offs, and how to use it today.
8 min read
AI News
Anthropic’s Multi-Agent Research: Coordination Patterns and Cascading Failures
Anthropic’s multi-agent research reveals that coordination without clear authority leads to cascading failures. An engineer’s breakdown of the patterns, risks, and how to build safer agent systems.
11 min read
AI News
When Copilot Autofix Ships a Backdoor: The Snowflake Jira Compromise
A deep-dive into how an AI-generated GitHub Copilot 'Autofix' suggestion introduced a backdoor into Snowflake's Jira. We dissect the prompt injection, the CI/CD trust collapse, and what Forward Deployed Engineers must redesign now.
9 min read
AI News
Stripe Acquires OpenRouter: The API Gateway Is Now the Rails of AI
Stripe's reported $7B+ acquisition of OpenRouter signals a tectonic shift: the API gateway is becoming the critical infrastructure layer for the AI economy. Here's the engineering breakdown.
8 min read
AI News
AI in Drug Discovery: The Real Engineering Bottlenecks Nobody Talks About
AI hasn't cured cancer yet. Here's the unvarnished engineering reality of AI in pharma—data scarcity, validation debt, and why the lab always wins—and what it means for Forward Deployed Engineers.
9 min read
AI News
LLMs Trained Only Below a Fifth-Grade Reading Level: What Breaks?
When an LLM sees only sub-fifth-grade text, it still learns syntax, but its reasoning collapses. Explore the surprising gaps, why it matters for domain-specific fine-tuning, and how to replicate the experiment.
10 min read
259 articles and counting