writing
Notes from the workshop.
Deep dives on full-stack development, AI automation, and agentic systems. No fluff, just practical knowledge you can use.
RAG Isn't Magic: When Retrieval Helps and When It Hurts
RAG gets sold as a cure. It's a tool that's excellent for some jobs and actively harmful for others. Here's how to tell which one you've got before you build it.
Read the post →The Problems AI Can't Fix (No Matter How Good the Model Gets)
A contrarian take: the categories of business problem where reaching for AI is the wrong move, why founders keep doing it anyway, and what to do instead.
Read →If You Can Build It in a Weekend, So Can Your Competitor
Wrapping a model isn't a moat. Where AI actually creates defensibility for a startup — and where it just commoditizes you alongside everyone else.
Read →Should You Pay Your AI by the Job, Not the Seat?
Outcome-based pricing is replacing per-seat SaaS for AI agents. When paying per completed job is a great deal for a founder — and when it quietly costs you more.
Read →Start With One Workflow: How to Pick Your First Automation
The mistake is trying to 'adopt AI' across the business at once. Pick one workflow. Here's the scorecard for choosing one that builds the muscle and proves the value.
Read →Your Engineers Aren't Resisting AI. They're Afraid of It.
Your team isn't resisting AI out of laziness. They're afraid of being blamed for code they didn't write. The 70/30 method draws the line that makes adoption feel safe.
Read →Can You Trust an AI Agent With Your Customer Data?
The data-privacy and security questions every founder should ask before letting an agent touch customers, payments, or PII — in plain language, not a compliance lecture.
Read →The 70/30 Engineering Audit: What to Automate and What Must Stay Human
A practical five-question audit for deciding which parts of an engineering or business workflow to automate, assist, approval-gate, or keep human-owned.
Read →Why We Build on Astro + Directus for AI-Heavy Sites
The stack you choose either gets out of the way of your AI work or quietly becomes the bottleneck. Here's why we reach for Astro and Directus when a site leans on automation.
Read →The Hidden Costs of AI Automation Nobody Budgets For
Teams budget the build cost and nothing else. The build is the cheap part. Here are the recurring costs that decide whether an AI project survives its first year.
Read →40% of AI Projects Get Killed. Here's Why Yours Might.
Gartner says 40%+ of agentic projects are at risk of cancellation. The real reasons agent projects die — unclear success criteria, no data access, eval drift — as a survival checklist.
Read →How to Calculate the ROI of an AI Agent (Without Lying to Yourself)
The ROI math everyone does for AI is wrong in a specific, optimistic way. Here's the honest version, including the costs and the one assumption that quietly inflates every estimate.
Read →Job Descriptions for Agents: A Template
The single highest-leverage thing you can do for an AI agent is write it a real job description before you deploy it. Here's the template we use, with every field explained.
Read →What Happens When Your AI Vendor Triples the Price?
Agent software spend is exploding and pricing is unstable. Here's how to architect so a vendor's price hike or a deprecated model can't hold your product hostage.
Read →What a Manager's Job Becomes When Agents Do the Work
If agents do the producing, what's left for the manager? The job doesn't shrink. It moves up: setting the standard, designing the system, and owning the judgment.
Read →Do You Need to Hire an AI Engineer, or Just Rent the Judgment?
The full-time ML hire everyone reaches for vs. what most early teams actually need. How to tell which problem you really have before you spend the headcount.
Read →How to Spot an AI Expert Who's Faking It
A founder's vetting checklist for AI consultants, agencies, and hires — the questions that separate real operators from prompt-jockeys with a good deck.
Read →The Agentic OS Maturity Model: 5 Stages
Most teams have no idea how far along they actually are with AI. Here are the five stages from one-off prompting to a real Agentic OS, and how to tell which one you're in.
Read →Guardrails: The Boring Work That Keeps AI Out of the Headlines
Every AI failure that makes the news shares a missing guardrail. They're the unglamorous constraints that turn a confident-but-risky agent into one you can actually trust.
Read →Why Your AI Demo Won't Survive Real Users
The demo was flawless. Then real users touched it. The gap between a demo that wows and a system that survives is the unglamorous work that decides whether AI ships.
Read →Evals for People Who Aren't ML Engineers
Evals sound like data-science jargon. They're just a way to test whether your AI is any good, on purpose, on a cadence. Here's how to build your first one with a spreadsheet.
Read →n8n vs. Custom Code: A Founder's Guide to Automation Plumbing
Low-code automation is faster until it isn't. Here's the honest framework for when n8n is the right tool, when custom code pays for itself, and how to combine them.
Read →Who's Accountable When the Agent Is Wrong?
Every leader weighing AI eventually hits the real question: when the agent makes a costly mistake, who owns it? If the honest answer is no one, you're not ready to ship.
Read →Cleaning Up Your Data Before You Automate: The Unsexy Prerequisite
Automating a process built on messy data doesn't fix the mess. It scales it. Here's what 'clean enough' actually means, and why it has to come first.
Read →What Does a Fractional CTO Cost? (And How It Compares to a Full-Time Hire)
What a fractional CTO actually costs — retainers, day rates, and projects — and how it compares to a full-time hire once you count equity, benefits, and ramp.
Read →Fractional CTO vs. Agency vs. Dev Shop: Which One Do You Actually Need?
Fractional CTO, agency, or dev shop? One sells judgment, the others sell hands. The honest comparison — what each is good at, where each fails, and how to choose.
Read →5 Signs It's Time to Bring in a Fractional CTO
Five signals it's time for a fractional CTO — from becoming the technical bottleneck to shipping AI that won't survive production. None are about headcount.
Read →The Context Layer: Why Your Agent Keeps Getting It Wrong
When an agent gets it wrong, it usually didn't reason badly. It answered correctly from incomplete information. The fix isn't a smarter model. It's a better context layer.
Read →Your AI Doesn't Have a Model Problem. It Has a Data Problem.
You upgraded the model and the output is still wrong. That's the tell. Almost every 'the AI isn't good enough' problem is a data problem wearing a model costume.
Read →When to Fire an Agent (and Hand the Work Back to a Human)
Everyone talks about deploying agents. Almost nobody talks about pulling one. Knowing when to take an agent off a job is a core management skill, not an admission of failure.
Read →Giving an Agent a Performance Review
Evals sound like an engineering chore. They're really the management ritual you already run: a regular, honest look at whether the work is good enough. Here's how to run one for an agent.
Read →Your First AI Agent Is a New Hire. Onboard It Like One.
Your AI agent isn't failing because the model is weak. You skipped its onboarding. Give it a job description, access, context, and a feedback loop like any new hire.
Read →The 70/30 Method: Building With AI Agents Without Betting the Company on Them
Why I let AI agents handle about 70% of the work and keep 30% for senior judgment — and how that ratio keeps AI projects out of the ditch.
Read →5 Signs Your Codebase Is Quietly Costing You Customers
The expensive problems in a codebase rarely announce themselves. Here are five symptoms I look for in an architecture audit — and what each one costs if you let it run.
Read →Do You Actually Need a CTO Yet?
A straight answer for founders weighing a full-time CTO, a fractional one, or no CTO at all — and how to tell which stage you're actually in.
Read →How I Build AI Agents That Actually Ship
Most AI agents die in the demo. Here's the process I use to get them into production — and the unglamorous parts that decide whether they survive contact with real users.
Read →