Blog

Engineering notes from Poket Dev. Python, AI/ML, cloud and DevOps, hand-written in Detroit.

Recent

Coordinated actions by an autonomous AI agent against production infrastructure, illustrating the Hugging Face agentic attack and its implications for security teams.

News • Jul 20, 2026 •12 min read

Hugging Face Agentic Attack Redefines AI Security Response

The Hugging Face agentic attack proves autonomous AI agents can bypass traditional defenses. Learn how to revise your AI threat models and incident response.

Financial review of AI agent cost per resolution showing why unit economics determine whether an agent ships to production.

Engineering • Jul 19, 2026 •12 min read

AI Agent Cost Per Resolution Decides If It Ships

AI agent cost per resolution, not eval accuracy, decides if your agent ships or dies. Token pricing hides the unit economics of retries and failures.

Netflix's in-house LLM serving platform illustrates what large-scale self-hosted inference infrastructure looks like when workload thresholds justify the investment.

Engineering • Jul 18, 2026 •12 min read

Self-Host LLM Inference, the Netflix Decision Framework

Learn when to self-host LLM inference instead of paying per token. Netflix's production stack reveals the real cost, latency, and control tradeoffs.

A behind-the-scenes look at a film set using digital equipment to highlight AI in media production workflows.

Use Cases • Jul 17, 2026 •11 min read

Where AI in Media Production Workflows Actually Pays Off

AI in media production workflows pays off in narrow post-production tasks, not creative origination. Netflix's 300 productions reveal where real ROI lives.

Illustration of a server network designed to orchestrate open source LLMs efficiently.

Engineering • Jul 16, 2026 •12 min read

Orchestrate Open Source LLMs vs Frontier Models

Should you orchestrate open source LLMs or call a single frontier model? We break down Sakana Fugu's claims and the real cost and latency tradeoffs.

OpenAI Codex subagent encryption conceals inter-agent instructions from system owners, breaking debugging, cost attribution, and compliance workflows across multi-agent pipelines.

News • Jul 15, 2026 •12 min read

OpenAI Codex Subagent Encryption Breaks Agent Observability

OpenAI Codex subagent encryption on GPT-5.6 Sol and Terra hides delegation chains from developers. Learn how it breaks debugging, cost, and compliance.

Implementing ChatGPT sales workflows on messy CRM records can lead to highly confident but entirely fabricated strategic outputs.

Use Cases • Jul 14, 2026 •10 min read

ChatGPT Sales Workflows Fail on Real CRM Data

OpenAI showcased ChatGPT sales workflows for pipeline briefs and forecast reviews. We audit where they break on messy CRM data and attribution gaps.

Real-time AI dental image verification assesses radiograph quality at the point of capture to reduce downstream insurance claim denials.

Use Cases • Jul 13, 2026 •13 min read

Real-Time AI Dental Image Verification Cuts Claim Denials

AI dental image verification cuts claim denials across 10,000 practices. Learn the architecture behind 1.4-second point-of-capture quality scoring.

A decision framework for choosing between fine-tuning, RAG, and prompt engineering when working with small open-weight language models.

Guides • Jul 12, 2026 •13 min read

Fine-Tuning vs RAG vs Prompt Engineering Decision Framework

Fine-tuning vs RAG vs prompt engineering: run this eval-gated framework before training a custom model and inheriting its hidden maintenance tax.

Abstract visualization of llm agent scaffolding showing structured layers that organize and constrain software logic.

Engineering • Jul 11, 2026 •13 min read

How LLM Agent Scaffolding Fixes Failing Code Review Agents

LLM agent scaffolding constrains what models see and call. Learn why GitHub Copilot code review regressed with more tools and how routing helps.

A network of interconnected data nodes representing the context graph foundation for building proactive AI agents.

Guides • Jul 10, 2026 •13 min read

Building Proactive AI Agents With Context Graphs

Proactive AI agents use context graphs to cut time-to-surface from 47 minutes to under 30 seconds, eliminating reactive RAG blind spots in enterprises.

Conceptual visualization of how AI coding benchmarks can be manipulated through memorized public data instead of genuine software engineering ability.

Tools • Jul 09, 2026 •13 min read

AI Coding Benchmarks Are Gameable by Design

AI coding benchmarks like SWE-Bench Pro are structurally gameable. Learn why public test sets fail and what leakage-free evaluation actually requires.

The Poket Dev Dispatch

Engineering notes from Detroit. Python, AI/ML, cloud and DevOps. No spam, unsubscribe anytime.

Join 3 readers. No spam. Unsubscribe anytime.