Insights
RSSTechnical perspectives, industry commentary, and things we've learned building software.
Boring Languages as an AI Leverage Strategy
Mainstream languages produce dramatically better results from LLM-assisted coding, which turns language conservatism into a deliberate engineering advantage.
By John Jansen · 29 May 2026 · 7 min read
Reading 12-Factor Agents Against the Reality of Shipping LLM Software
A close reading of the 12-factor-agents principles: what they get right, what they restate from web ops, and where the framing hides the hard parts.
By John Jansen · 28 May 2026 · 7 min read
Personal vs Canonical Skill Libraries: Where Should Coding Agents Get Their Instructions?
Matt Pocock's personal .claude/skills repo crossing 3,800 stars in a day raises a real question for engineering orgs: whose curation do you trust?
By John Jansen · 25 May 2026 · 6 min read
What Specs Own When Agents Write the Code
Spec-driven development is becoming a real discipline. The hard part is the boundary: when agents write the code, what should the spec own, and what shouldn't?
By John Jansen · 24 May 2026 · 7 min read
Where Multi-Provider LLM Routers Fit in Production Stacks
Provider routers solve real problems, but the layer they occupy in your stack determines whether they help or quietly undermine retry, governance, and quality.
By John Jansen · 20 May 2026 · 7 min read
What Bun's Rewrite in Rust Says About Runtime Engineering in 2026
Bun moving its core from Zig to Rust is less a language preference and more a verdict on what runtime authors now need from their toolchain.
By John Jansen · 19 May 2026 · 7 min read
AI in Government: What Adoption Looks Like, and What It Doesn't
The NZ government wants to replace 9,000 public servants with AI but has no plan for how. There's a better way to think about this.
By John Jansen · 19 May 2026 · 7 min read
What Spec-Driven Development Asks of Coding Agents
GitHub's spec-kit puts specs at the center of agent workflows, but the review, versioning, and authority questions need answers before adoption.
By John Jansen · 19 May 2026 · 7 min read
What's already inside the model?
Guardrails gate what a model will say, but the weights themselves carry knowledge of every exploit ever published. That asymmetry deserves more attention.
By John Jansen · 29 April 2026 · 7 min read
Where GitHub Actions Sits in the AI-Augmented Delivery Pipeline
As Copilot review and agent runners consume Actions minutes, the runner becomes a concentration point for secrets, write access, and AI-generated code.
By John Jansen · 29 April 2026 · 7 min read
Who Owns the Diff When an Agent Writes It?
IP, licensing, and CLAs assume human authorship. As agent-authored commits scale, procurement teams and OSS maintainers need new defaults.
By John Jansen · 29 April 2026 · 7 min read
Skills as the Portable Packaging Format for Coding Agents
Skills are quietly becoming the unit of capability for coding agents — a packaging format that moves cleanly across vendors, runtimes, and teams.
By John Jansen · 29 April 2026 · 7 min read
AGENTS.md as a Force Multiplier: What High-Leverage Agent Docs Actually Contain
A well-structured AGENTS.md behaves like a model upgrade; a sloppy one drops agent output below the no-context baseline. Here's what the difference looks like.
By John Jansen · 29 April 2026 · 6 min read
Phantom-Patch Attacks and the Apply-Step Problem
Auto-apply coding agents pull diffs from commit metadata and PRs — that's a new untrusted-input surface, and the apply-step needs hardening before this scales.
By John Jansen · 29 April 2026 · 7 min read
What Automated ML Engineer Agents Actually Do in Practice
Hugging Face's ml-intern is a useful lens for seeing where autonomous ML agents genuinely work and where the seams still show.
By John Jansen · 29 April 2026 · 7 min read
Why AI Tools Are Making Some Teams More Tired, Not Less
AI was supposed to reduce cognitive load. For many knowledge workers, it has quietly increased it. Here's what we think is actually happening.
By John Jansen · 21 April 2026 · 7 min read
What Public-Market AI Labs Mean for Enterprise Procurement
As OpenAI and Anthropic approach public-market scale, enterprises need to rethink vendor commitments, pricing assumptions, and exit paths.
By John Jansen · 20 April 2026 · 6 min read
Prompt Injection in Agentic Systems: A Threat Model
How attackers exploit LLM agents that read untrusted input and hold real tools, and how to architect systems that survive it.
By John Jansen · 20 April 2026 · 7 min read
Where AI Tooling Sits in the Supply Chain Threat Model
The Context.ai–Vercel and LiteLLM–Mercor incidents suggest AI-tool dependencies deserve their own category in vendor review and SBOM practice.
By John Jansen · 20 April 2026 · 7 min read
Where AI ROI Actually Comes From in Infrastructure and Operations
Gartner says only 28% of AI use cases in I&O fully return value. The lever isn't model sophistication — it's integration discipline and governance.
By John Jansen · 20 April 2026 · 6 min read
MCP as Public Infrastructure: What Changes When the Protocol Isn't Anyone's
With MCP at 97M installs and moving to the Linux Foundation, agent architecture stops being about protocol bets and starts being about composition.
By John Jansen · 20 April 2026 · 6 min read
From IDEs to Spec Compilers: What the Next Two Years of Software Engineering Look Like
A prediction: the IDE becomes secondary, specs become the source of truth, and the bottleneck shifts from typing code to deciding what to build.
By John Jansen · 20 April 2026 · 6 min read
What Claude Design Gets Right: The Codebase Is the Design System
Anthropic's Claude Design reads your codebase to derive a design system and feeds finished designs back to Claude Code — the real bet is the closed loop.
By John Jansen · 20 April 2026 · 7 min read
The Datacenter Buildout Is Real. So Is the Pendulum.
Hyperscaler AI infrastructure is genuinely load-bearing — but history suggests personal compute and open models will reshape where that load sits.
By John Jansen · 18 April 2026 · 8 min read
What Anthropic's Constitutional AI Gets Right
Claude's architecture isn't just safety guardrails — it's a coherent engineering philosophy for composing values, uncertainty, and behaviour into an LLM.
By John Jansen · 17 April 2026 · 7 min read
Every AI Agent Benchmark Is Broken
Berkeley researchers hit near-perfect scores on all eight major AI agent benchmarks without solving a single task. Evaluation is broken — teams need a new plan.
By John Jansen · 13 April 2026 · 4 min read
The Death of Progressive Enhancement
JavaScript-first development has killed the web's foundational principle of graceful degradation, and the performance costs are staggering.
By John Jansen · 13 April 2026 · 6 min read
The Container Orchestration Overhead Crisis
Kubernetes and Docker add 3-5x resource overhead to simple applications, but most teams accept this waste because deployment complexity seems solved.
By John Jansen · 12 April 2026 · 7 min read
How Twill.ai Cloud Coding Agents Are Transforming Software Development
Exploring Twill.ai's approach to cloud-based coding agents that automatically generate pull requests and execute development tasks in isolated sandboxes.
By John Jansen · 11 April 2026 · 4 min read
The Great SQLite Revival Is Just Beginning
SQLite evolves from embedded database to application backbone through extensions and edge computing. Most systems don't need PostgreSQL's complexity.
By John Jansen · 11 April 2026 · 7 min read
The Great API Consolidation Is Coming
AI agents need fewer, smarter APIs rather than thousands of specialized endpoints. Learn how this consolidation is reshaping service interface design.
By John Jansen · 5 April 2026 · 5 min read
We Built a Clustered Graph Database
Open-source distributed graph database with Raft consensus and Bloom filter routing. Learn why we built Loveliness and how it works differently.
By John Jansen · 2 April 2026 · 4 min read
Real-Time AI Is Eating Database Buffers
AI inference demands are incompatible with traditional database buffers. Find out how this forces a rethink of memory management for real-time applications.
By John Jansen · 2 April 2026 · 7 min read
The AI Agent Memory War Is Here
Vector databases vs graph stores vs SQL: Which memory architecture enables the most capable AI agents? The battle for the foundation of autonomous systems.
By John Jansen · 2 April 2026 · 7 min read
Securing AI Agents: Best Practices for Enterprise Adoption
AI agents are transforming business operations but introduce new security risks. Here's how to adopt them safely with proper safeguards.
By John Jansen · 30 March 2026 · 5 min read
AI-Powered Development Tools Are Reshaping Software Engineering
AI coding assistants are transforming software development. Here's how they're changing workflows, productivity, and what it means for engineering teams.
By John Jansen · 30 March 2026 · 4 min read
The Jevons Paradox Came for Knowledge Work
Eight-month Berkeley study: AI tools scale work volume, not efficiency — the Jevons paradox has come for knowledge work. Serious implications for engineers.
By John Jansen · 23 March 2026 · 6 min read
AI Is Quietly Picking Your Programming Language
TypeScript just overtook Python on GitHub, and not by accident. AI coding assistants are quietly reshaping which language your next project gets built in.
By John Jansen · 22 March 2026 · 5 min read
The AI Coding Stack Is Splitting in Two
AI coding agents are splitting into two architectures: vertically integrated vs model-agnostic. Your choice determines more than you think.
By John Jansen · 21 March 2026 · 7 min read
AI Companies Are Buying Your Developer Tools
OpenAI acquired Astral. Anthropic already owns Bun. The AI coding war is now a platform war — and the battleground is the toolchain you use every day.
By John Jansen · 20 March 2026 · 7 min read
Code Generation Just Broke Your Engineering Pipeline
AI writes 14,000 lines in three hours. The bottleneck shifts to both ends of the pipeline — specification and review — where most teams are weakest.
By John Jansen · 20 March 2026 · 9 min read
Agent Skills Are the New npm Packages
AI coding agents now have shareable skill ecosystems — and they're the fastest-growing category on GitHub. The platform war for agent distribution has started.
By John Jansen · 19 March 2026 · 8 min read
HTTP 402 Finally Gets a Job
AI agents need to pay for things. Stripe, Coinbase, and Google are racing to build the protocol — and all three chose HTTP 402, dormant since 1997.
By John Jansen · 19 March 2026 · 7 min read
A Smart Firewall for AI Tool Calls
What if every tool call from an AI agent had to pass through an independent process that decided whether it looked safe — before anything executed?
By John Jansen · 18 March 2026 · 12 min read
Where AI Agents Actually Break
Most AI security discussion focuses on the model. The real vulnerabilities are in the loop around it — the tool-calling scaffolding that nobody audits.
By John Jansen · 18 March 2026 · 7 min read
AI Agents Are Learning to Hack Without Being Told To
AI agents autonomously discover exploits, bypass DLP via steganography, and pressure each other past safety checks — no adversarial prompts needed.
By John Jansen · 18 March 2026 · 8 min read
AI Is Writing Code Faster Than You Can Ship It
AI coding tools generate code faster than teams can review it. Three major reports confirm: the bottleneck has moved from writing to everything after.
By John Jansen · 17 March 2026 · 4 min read
What a Million Tokens of Context Actually Changes
A million-token context window at standard pricing eliminates RAG, summarisation, and context engineering. Here's what actually changes — and what doesn't.
By John Jansen · 14 March 2026 · 6 min read
The Uncomfortable Truth About AI Agents That Can Read Your Email
OpenClaw has 312K stars and 30K exposed instances. Autonomous agents with email and API access have no workable security model — and adoption outpaces safety.
By John Jansen · 14 March 2026 · 10 min read
Local AI Is About to Hit a Complexity Wall
Tools like canirun reveal a hard truth: local AI deployment is becoming an enterprise-grade problem most teams can't handle.
By John Jansen · 14 March 2026 · 5 min read
How We Found the Fake News King: An OSINT Investigation for NPR
OSINT, domain forensics, and network analysis: how we traced a fake news story with 1.6M views to its source for NPR Planet Money.
By John Jansen · 10 March 2026 · 10 min read
Scraping 24,000 TEDx Talks Across 147 Countries: A Data Engineering Story
How we scraped, normalised, and deduplicated 24,000 TEDx talks across 147 countries and 50 languages — and the data engineering decisions that made it work.
By John Jansen · 28 February 2026 · 7 min read
How We Built a COVID-19 Research Dashboard That Doctors Actually Used
How an on-demand NLP tool became a pandemic-scale research dashboard overnight — the architecture decisions that made it possible.
By John Jansen · 15 February 2026 · 6 min read
Anthropic's Claude Mythos: AI Cybersecurity Tool Kept Private
Anthropic's Claude Mythos AI model exposes software vulnerabilities but is withheld due to security concerns. Explore responsible AI development implications.
By John Jansen · 4 min read
Research-Driven Coding Agents: Next Evolution in AI Development
AI coding agents that research papers before coding achieve 15% performance gains. Discover how this approach outperforms traditional agents.
By John Jansen · 4 min read
AI Regulation and Cybersecurity: Navigating the New Frontier
How AI regulation and cybersecurity intersect as critical challenges for technology leaders, with insights from Anthropic's Claude Mythos.
By John Jansen · 3 min read
Want to discuss any of this?
We write about what we're actually doing. If something resonates, we'd love to hear about your context.