#ai

Three concurrent coding agents taught me the actual bottleneck: not prompting, but assignment, evidence, review, and release control.

How I built a full LittleBird clone with screen context reading, meeting recording, arena mode, and MCP tool support — from scratch to packaged .app in a single coding session.

Five lines of Python and an API key produce a working demo. The gap between that demo and a production system contains failure modes the prototype...

Most enterprise AI transformations are solving problems that spreadsheets handle at 1/50th the cost. The misalignment is driven by career incentives,...

Most AI agent infrastructure is premature. The agents themselves barely work. The industry is selling Formula 1 equipment to people still learning to...

AI coding assistants output shell commands, not GUI instructions. That single fact is reversing a decade of developer tooling trends.

AI review only becomes valuable when it can reason about behavior, blast radius, user impact, and the evidence required to trust a change.

Async code generation is delegated execution. The new work is task design, review, evidence, and deciding what the system is allowed to ship.

AI commoditized the pattern recognition and architectural intuition that made 10x developers valuable. The bottleneck moved from individual output to...

AI-generated code produces different bugs than human-written code. QA built for syntax checking is testing for the wrong failures.

I shipped a voice profile extractor at 60% accuracy. Simple pattern matching outperformed ML for writing voice replication.

Different LLMs have different strengths. Routing tasks to the right model -- like heterogeneous compute -- turns out to be more valuable than using one ...

I built a multi-AI content pipeline combining Gemini and Claude. The failures taught me more than the architecture.

I built a voice replication system by extracting patterns from my blog corpus. Here's what it captures, what it misses, and what that reveals about...

Production AI teams do not win by hand-tuning clever prompts. They version, evaluate, optimize, and observe behavior like software.

Combining Semgrep, CodeQL, SonarQube, and Snyk gets you 44.7% vulnerability detection. Semantic SAST combines Tree-sitter with LLM reasoning to do better.

Multiple agents do not need a shared brain. They need explicit context, durable memory, and a record of why the project works the way it does.

Every web design decision now must serve two audiences: humans who browse visually and AI agents that consume data programmatically. The architectural...

I asked Claude to analyze my writing style across my blog posts. The patterns it found -- and the ones I didn't know I had -- were genuinely surprising.

OCode: Why I Built My Own Claude Code (and Why You Might Too): A few nights ago, I opened my Anthropic invoice.

Autonomous agents need a control plane: identity, policy, secrets, and audit trails that make delegated work governable.

A photographer friend posted a sunset photo after three hours of waiting for the perfect light. Within minutes: 'Obvious Midjourney.' 'Nice prompt, bro.'

AI doesn't make everyone equally skilled. It amplifies existing ability. That changes what technical interviews should test.

Deep dive into RAG architectures: chunking strategies, retrieval methods, embedding optimization, and production patterns with research-backed analysis.

Systematic experiments on temperature and top-p sampling parameters across 1000 real queries with empirical data on creativity, coherence, and...

OpenAI recently rolled back a GPT-4 update due to sycophantic behavior. The word itself--'sycophantic'--feels like a punchline from a _Black Mirror_...

Most AI products are designed to fail. Not because the technology is bad, but because product teams are building for the wrong expectations entirely.

I kept writing terrible JIRA titles during customer calls. So I built a Chrome extension to fix it.