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 4 of 11
AI News
The AI Credit Resale Economy: How Token Brokers Arbitrage Inference Costs
Token brokers are exploiting pricing gaps between AI providers to resell inference credits. Here's how the arbitrage works, what it means for your API bill, and how engineers can exploit the same dynamics.
9 min read
AI News
Multi-Agent Systems: Architectural Patterns and Failure Modes Engineers Must Know
Anthropic's research reveals multi-agent systems don't always beat single agents. We dissect the architectures, emergent failure modes, and when to use them in production.
9 min read
AI News
Claude System Prompts: Operationalizing Model Behavior at the API Layer
Anthropic just shipped native system prompt support in the Claude API. For engineers, this isn't a UI tweak—it's a contract for deterministic model behavior, eliminating fragile pre-prompt hacks and finally making Claude a reliable component in production pipelines.
10 min read
AI News
Adversarial Prompts in Legal Docs: An LLM Supply Chain Attack Case Study
A man injected hidden text into court filings to hijack a suspected AI summary. We unpack the prompt injection mechanics, why this is a critical supply chain risk for engineers deploying RAG pipelines, and how to test your own systems against it.
9 min read
AI News
When a Model Feels Worse: Quantifying Regressions in Instruction Following
Why does a new model release feel clunky and pedantic? We break down a reproducible method to quantify tone and instruction-following regressions, moving from vibes to hard data.
9 min read
AI News
Beyond Linear Chat: Building Editable Context DAGs for Multi-Turn LLM Reasoning
Linear chat chains break complex reasoning. Learn how editable context DAGs let you branch, merge, and prune LLM conversations, turning chaotic threads into structured engineering workflows.
11 min read
AI News
Prompting as Delegation: Why AI-Assisted Coding Mirrors Engineering Management
AI-assisted coding isn't a faster way to type—it's a leadership exercise. Learn why shifting from syntax to specification mirrors engineering management and how to delegate effectively to LLMs.
10 min read
AI News
Claude Code Sessions: Cost Engineering and Context Reuse Patterns That Ship Faster
Stop burning tokens on repetitive context. Master Claude Code session hygiene with compaction, CLAUDE.md, and Git worktrees to slash latency, cut costs, and keep your flow state intact.
9 min read
AI News
Llama.cpp Deep Dive: Run Production-Grade LLMs on a Single Workstation
llama.cpp compresses frontier AI into a single C++ binary. Here's how it works under the hood, why it obliterates GPU vendor lock-in, and how engineers can deploy it today.
12 min read
AI News
Why AI Text Watermarks Will Always Fail: A Technical Breakdown for Engineers
Text AI watermarks are marketed as a defense against deepfakes, but they are trivial to remove. This engineer's deep-dive explains why statistical signals in token selection are fundamentally broken.
8 min read
AI News
49KB per Log Line: How systemd-journald Wrecks Your Disk and How to Tame It
A single journald log line can cost 49KB+ on ext4 and 110KB+ on btrfs. Unpack the filesystem-wrecking physics, learn to measure the real cost, and apply surgical fixes before your production SSDs burn out.
10 min read
AI News
Google's Homomorphic Encryption: Private AI Engineers Can Deploy Today
Google's open-source HE compilers and transpilers let you run ML inference on encrypted data without exposing plaintext. Here's what changed, why it matters for FDEs shipping on-prem, and how to try it now.
11 min read
AI News
GLM-5.3’s Emergent Cyber Skills: What Security Engineers Need to Know
GLM-5.3 demonstrates emergent cyber capabilities like blind SQL injection and rootkit detection without explicit training. Here’s what that means for security engineers and FDEs working at the frontier.
8 min read
AI News
Mistral OCR 4.1: Multimodal Grounding for Structured Data Extraction
Mistral OCR 4.1 moves beyond text recognition to grounded structured extraction from multi-modal documents. Here's the engineering breakdown, API usage, and why it matters for data pipelines.
9 min read
AI News
AI Bot User-Agent Spoofing: How Attackers Evade WAFs During Mass Scans
Attackers are spoofing AI crawler user-agents like ClaudeBot to bypass detection during mass vulnerability scans. Here's how it works, why it matters for your defense posture, and what you can do about it today.
9 min read
AI News
Cerebras WSE-3 Hits 4,200 Tokens/s on GPT-5.6: The Inference Wall Is Gone
Cerebras demonstrated 4,200 tokens/sec on OpenAI’s GPT-5.6 Sol UltraFast. Here’s the engineering breakdown of wafer-scale inference, why latency is the new bottleneck, and how FDEs should retool for instant AI.
9 min read
AI News
DeepSeek Harness: Multi-Step Agent Orchestration, Developer Preview
DeepSeek Harness is a developer preview for orchestrating multi-step AI agents. We break down the architecture, the Python SDK, and what it means for Forward Deployed Engineers building reliable agentic workflows.
9 min read
AI News
Gemini 3.7 Flash: Engineering Trade-offs in Sub-100ms Inference at Scale
Google's Gemini 3.7 Flash targets sub-100ms inference. We dissect the architectural trade-offs in latency, throughput, and cost, and show how Forward-Deployed Engineers can exploit this speed for real-time AI features.
9 min read
AI News
H3-metal: Native MiniMax-H3 Inference on Apple Silicon GPUs
antirez's H3-metal ports MiniMax-H3 natively to Apple Silicon via Metal. We break down the architecture, performance, and how engineers can run frontier AI locally.
10 min read
AI News
GPU Passthrough on macOS VMs: Fast llama.cpp Inference on Apple Silicon
Cua's macOS VM now passes the Apple Silicon GPU directly to llama.cpp, unlocking near-native LLM inference speeds inside a sandbox. Here's the architecture, benchmarks, and how to try it.
10 min read
AI News
WorldClaw: How Agentic Pipelines Generate Infinite 3D Open Worlds at Scale
Tencent's WorldClaw uses an agentic feedback loop to generate coherent, high-quality 3D worlds from text prompts. Here's the engineering breakdown and why it matters for FDEs building generative pipelines.
10 min read
AI News
Grok 4.6: What the 61-Point Intelligence Index Score Means for Engineers
Grok 4.6 hits 61 on the Artificial Analysis Intelligence Index. We break down the benchmark, what it means for real-world engineering workflows, and how to start using it today.
7 min read
AI News
DeepSeek V4 Pro 0813: The Engineering Reality Behind the Benchmarks
A no-fluff teardown of DeepSeek V4 Pro 0813. We cut through the hype to analyze the silent architecture upgrades, practical tokenomics, and why this matters for production pipelines.
8 min read
AI News
How Claude Embeds and Detects Watermarks in AI-Generated Text and Code
A no-fluff breakdown of Anthropic's visible watermarking for Claude: how it works under the hood, why it matters for forward-deployed engineers shipping AI features, and how to test it yourself today.
11 min read
259 articles and counting