Research & Insights

What we're thinking about

Technical writing, architecture decisions, and honest assessments of AI tooling from engineers who build with it daily.

161 articles published

Featured

Tools

Cursor vs Claude Code in April 2026: What Actually Changed

Our original Cursor vs Claude Code comparison got valid pushback — Cursor has shipped agent mode, subagents, a CLI, cloud agents, and terminal access since we wrote it. Here is the corrected picture as of April 2026, with what actually matters for choosing between them now.

Read articleApr 8, 2026 · 8 min read

161 articlesLatest first

APR292026

AI News4

OpenAI Just Shipped Codex as a Claude Code Plugin — Here's What It Actually Means

OpenAI shipped an official Codex plugin for Claude Code, letting devs run both agents inside one terminal session. Truce, trojan horse, or just developers winning?

APR292026

Industry8 min read

The True Cost of Technical Debt in Legacy Systems

Technical debt is not just slow development — it is compounding interest on every future feature, every hire that takes longer to onboard, and every incident that takes longer to resolve. The AI era has made legacy systems even more expensive by making them harder to integrate with modern AI tooling.

APR282026

Opinion12

The AI Agent ROI Math Is Upside Down and Nobody Wants to Admit It

Companies are spending more on AI agents than the humans those agents replaced. The dirty secret is that almost nobody is measuring this honestly. Here is why the math is upside down — and what to do about it.

APR282026

Tools6

GitHub Copilot vs Claude Code in 2026: After Crunching the Numbers, Here's the Truth

Both vendors moved to usage-based billing the same week. After pricing real engineering workloads against both, here is when each one actually wins — and why most senior teams will end up paying for both.

APR282026

Engineering12

How to Add Idempotency Keys to Your API Without Breaking Existing Clients

A field-tested guide to adding idempotency keys to a live API — Postgres-backed, opt-in, and safe to roll out without breaking the clients you already have in production.

APR282026

AI & Engineering12 min read

The Vendor Lock-In Hidden in Your AI Prompts

Your prompts are structurally coupled to specific models. When Claude prefers XML tags and GPT-4 prefers Markdown, switching providers means rewriting every prompt in your system. This is vendor lock-in hiding in plain text — and research shows format alone can swing accuracy by 78 percentage points.

APR222026

Engineering14 min read

Product-Led Growth Engineering in the AI Era

PLG is evolving into ALG — Agentic-Led Growth. Cursor hit $500M ARR in under 24 months. Lovable hit $100M ARR in 8 months. Per-seat pricing is dying. Per-task and per-outcome models are replacing it. Engineering teams are being asked to build for metrics that did not exist 18 months ago.

APR152026

AI & Engineering15 min read

AI Agent Architecture: Production Patterns from the Field

Building an agent that works in a demo takes a day. Building one that runs reliably in production for six months takes a team. These are the architectural patterns that separate the two.

APR152026

AI Strategy8 min

The Complacency Gap: What Switching AI Coding Tools Taught Me About Vigilance

Switching from Claude Code to Codex felt like cognitive load returning — slower, less comfortable, more effortful. Then I realized that load was not a cost. It was the work. A reflection on automation complacency, context decay, and what the right relationship with AI coding tools actually looks like.

APR152026

Explainer5

Prompt Injection Explained for People Who Don’t Write Code

Prompt injection is the AI version of social engineering: instead of hacking the system directly, an attacker tricks the model into following bad instructions. If your product uses AI to read, search, summarize, or take actions, this is a business risk, not just a technical one.

Showing 10 of 161