Long Context Hallucinations: Why the Demo Is Not the System
LLMs can pass a small document demo and still fail when the same task scales. A simple invoice benchmark shows why deterministic workflows still need deterministic software.
Fractional CTO who works forward-deployed: embedded in your team, on the tools, shipping with you.
I'm Norman Holz. I take messy ideas, stalled projects, and AI prototypes and drive them to production, with architecture, hands-on code, and shipping discipline in one seat. Part fractional CTO, part forward-deployed engineer: I turn vision into a technical plan and stay on the tools until it crosses the final mile.
Clear decisions, solid systems, teams that deliver.
Warrior in the Garden
Samurai
Nano Banana Pro 2
✦ AI Generated
Astronaut in the Ocean
Astronaut
Nano Banana Pro 2
✦ AI Generated
Caveman in a Modern City
Caveman
Nano Banana Pro 2
✦ AI GeneratedFractional Realities generated by Nano Banana Pro 2
AI tools, no-code platforms, or a fast freelancer proved the shape of the idea. But it's not ready for real customers, and the last 20% is the hard part.
The agency, contractor, or internal team left behind unclear architecture, brittle code, missing context, and a roadmap nobody fully trusts.
Agencies and freelancers promise a lot. Some proposals are worth it. Some bundle in work you pay for but do not need yet.
Your developers are talented, but nobody's connecting their work to actual business outcomes. Features ship. Value doesn't. And nobody's asking why morale is tanking.
Every new feature takes twice as long as the last. Your codebase is a house of cards and everyone's afraid to touch it.
Investors, customers, and your own team are asking technical questions that need real product, architecture, and delivery judgment, not a guess.
Low-commitment ways in before any retainer. Get value fast, then decide if you want me around longer.
Product Rescue Check
A direct diagnosis of what is stuck and what is worth saving. You walk away with a clear recommendation: repair, rebuild, or a sharper product cut.
Release Sprint
A focused push to turn a stuck project, AI prototype, or urgent product release into something demoable, fundable, sellable, or production-ready. Paid and scoped up front.
The Forward-Deployed CTO Retainer: weekly CTO leadership across product direction, architecture, vendor decisions, team cadence, and the work it takes to get software shipped. Flat monthly, scoped to your situation, easy to stop if it stops being useful.
Advisory
Steady technical judgment without full immersion. Product calls, architecture reviews, vendor decisions, and a filter on the expensive questions.
Embedded
I work inside your team. Hands-on architecture, code reviews, sprint cadence, and direct execution. For when the build is stuck and someone has to get software out the door.
Monthly retainer. Cancel anytime. No notice period.
Notes on AI strategy, technical leadership, and the human side of software teams.
LLMs can pass a small document demo and still fail when the same task scales. A simple invoice benchmark shows why deterministic workflows still need deterministic software.
A malicious dependency can read SSH keys, API tokens, and CI secrets before your code ever runs. Safer uv, npm, and JavaScript package manager defaults reduce the blast radius.
AI vendors promise a lot. This checklist helps founders separate real capability from polished sales material before the invoice lands.
Send the facts as they are: what exists, where it hurts, which constraints matter, and what you want to use an intro call for.