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.
LatestAI News
Your AI Agent Has Root: Threat Modeling LLM-Driven System Administration
LLMs are being given sudo. We break down the Infernal Code incident, map the attack surface of an AI agent with root access, and provide a practical threat model for engineers deploying autonomous sysadmin tools.
11 min read
Read the articleMore ai news
AI News
Claude Code Session URLs in Commits: Audit Trails Every FDE Needs
Claude Code now appends session URLs to commits by default. Here’s what that means for audit trails, compliance, and debugging AI-assisted code in production.
10 min read
AI News
Warp's Self-Improving Claude Agents: The Feedback Loop Architecture
Warp's engineering team built self-improving coding agents on Claude using a tight feedback loop—agents generate code, tests score outputs, and failures feed back as few-shot examples. Here's how the loop works, why it matters for production AI, and how to prototype it today.
10 min read
AI News
Domain-Driven Agents: Bounded Contexts for Reliable AI Workflows
AI agents fail when they mix data. Domain-Driven Agents apply bounded contexts from software architecture to LLM workflows, scoping prompts and tools for deterministic reliability.
11 min read
AI News
vLLM v0.28.0: How Automatic Prefix Caching Slashes LLM Serving Costs
vLLM v0.28.0 ships automatic prefix caching (APC) that reuses KV-cache across requests. We break down the hashing internals, the cost math, and how to enable it in production.
9 min read
AI News
StemDeck: Run Studio-Quality Audio Stem Separation Locally, No Cloud Needed
StemDeck brings AI-powered audio source separation to your local machine. Here's how it works, why it matters for engineers building audio pipelines, and how to run it today.
9 min read
AI News
AI Sludge in OSS: Why Maintainers Are Blocking LLM-Generated Contributions
Maintainers are drowning in AI-generated pull requests that waste time and erode trust. Here's why the 'CV-driven contribution' model is broken, how it hurts forward deployed engineers, and what actually works.
9 min read
AI News
When LLM Context Becomes a Tracer: Accidental Program Analysis via Memory Injection
How a misconfigured LLM agent turned its memory into a runtime tracer. We unpack the mechanism, why it matters for forward-deployed engineers debugging production AI, and how to harness it deliberately.
8 min read
AI News
OpenRouter That Learns: Optimizing Model Routing with Usage Data
Experiential Labs open-sourced a model router that logs usage to optimize future selections. Here's why engineers should care about data-driven routing and how to deploy it.
11 min read
AI News
GLM-5.3 Goes Open-Weight: What This MoE Model Means for Local Inference
Z.ai's GLM-5.3 is now open-weight under MIT license. We break down what a Mixture-of-Experts architecture means for local inference, how to run it on consumer hardware, and why this matters for engineers building private, specialized toolchains.
9 min read
AI News
Gemini 3.5 Transcribe: Timestamp-Precise Speech-to-Text for Engineers
Google’s Gemini 3.5 Transcribe brings word-level timestamps and 100+ language support to speech-to-text. Here’s why it matters for FDEs building voice pipelines, how to use it today, and the cost/accuracy tradeoffs.
9 min read
AI News
Why musl Can Tank Performance: malloc and Locale Overhead Explained
musl is lightweight but its malloc and locale handling can silently kill throughput. An engineer's deep dive into the bottlenecks, benchmarks, and practical fixes.
9 min read
AI News
Terminal-Bench-Science: The First Real Benchmark for AI Research Agents
Terminal-Bench-Science evaluates AI agents on gritty, real-world research workflows—not contrived puzzles. Here's why the FDE skill set is the perfect lens to judge these models.
8 min read
AI News
DNS Cache Shrink: How Cloudflare Saved 100TB of RAM on 1.1.1.1
Cloudflare cut 100 terabytes of memory from 1.1.1.1 by rewriting the DNS cache with a bespoke hash table and arena allocation. Here's the engineering breakdown and why it matters for FDEs.
11 min read
AI News
HTTPX2 Is a Ground-Up Rewrite: What Changes for Python HTTP Clients
HTTPX2 is not a minor bump. It's a ground-up async-native rewrite from the pydantic team. Here's what changes for engineers who rely on Python HTTP clients for high-throughput data pipelines and API integrations.
6 min read
AI News
GLM-5.3-Flash and Ox Alpha: Open-Weight Rivals Reshape Model Economics
Z.ai's Ox Alpha and GLM-5.3-Flash are open-weight models that rival DeepSeek on performance while cutting inference costs. Here's what happened, why it matters for FDEs, and how to deploy them today.
9 min read
AI News
Agentic Context Management: Why Memory Is an Architecture Problem, Not a Prompt Fix
Forget stuffing system prompts. The paper on Agentic Context Management reframes memory as an architecture problem with massive cost, latency, and accuracy implications. Here's what engineers need to know.
12 min read
AI News
WebMCP: Ship a Native API for AI Agents on Your Website
WebMCP turns any website into a tool-calling endpoint for AI agents. Learn how the Model Context Protocol meets the browser, why this shifts FDE integration patterns, and how to try it today.
11 min read
AI News
Six Months Coding with Agents: Speed Gains and Sharp Limits
A raw, six-month field report on coding exclusively with AI agents. We unpack the 10x speed boosts, the catastrophic failures, and the concrete workflows that separate hype from production-grade engineering.
9 min read
AI News
Gemini Omni 1.1 Flash: Native Multimodal Output for Builders
Google's Gemini Omni 1.1 Flash adds native image and audio output. We break down why multimodal I/O changes agent architecture, how to use it, and the real limits.
9 min read
AI News
OpenAI Restores 5-Hour Codex Limits: Engineering Workflow Impact
OpenAI restored 5-hour Codex limits for ChatGPT Plus. We break down the quota mechanics, what it means for daily coding workflows, and how Forward Deployed Engineers can integrate this into a high-signal development loop without hitting the wall.
9 min read
AI News
OCR It: Build a Document-to-LLM Pipeline When Copy-Paste Is Blocked
Stuck behind a locked PDF or unscrapable document? Learn how to build a local OCR-to-LLM pipeline that extracts text from screenshots, bypassing UI copy-blockers for your RAG and agent workflows.
7 min read
AI News
Why VMs Won't Contain Cyber-Capable AI: A Technical Breakdown
Virtual machines aren't the sandbox you think they are. We break down why modern AI agents escape VM containment through side-channels, hardware passthrough, and prompt engineering—and what engineers should build instead.
10 min read
AI News
Qwen3.8-Flash-Next: Under-the-Hood Upgrades for Agentic Workflows
Alibaba's Qwen3.8-Flash-Next isn't a bigger model—it's a surgical upgrade for tool-calling, structured output, and multi-turn stability. Here's what changed and why it matters for agentic pipelines.
11 min read
259 articles and counting