Skip to content
PRTOTYPE.COM · Production studioAudit · Build · Maintain
Field notes

What it costs.
What breaks.

Two subjects. Private LLMs: cost, compliance and model choice for teams deciding whether to self-host. AI-built apps: what fails in due diligence, in a security questionnaire and in production. Written with the numbers on the table, including the ones that argue against us.

16 November 2026

Coding agents in production, 2026: what they leave undone

Coding agents and MCP moved fast in 2026. The incident record shows what they still leave open: auth, secrets, data access rules, tests, deploy and rollback.

Read the note
16 November 2026

Security questionnaires for AI-built apps: a UK guide

SIG, CAIQ, SOC 2, ISO 27001 and Cyber Essentials: which questions an AI-built startup product cannot answer yet, and the evidence that satisfies each one.

Read the note
16 November 2026

Technical due diligence on an AI-built MVP: what fails

What a technical reviewer checks in an AI-built MVP, what usually fails, and the evidence pack a founder should prepare before investors open the repository.

Read the note
6 July 2026

Data That Can't Leave the Building: Private LLMs for Regulated Sectors

For regulated data, the LLM question is whether the tokens may leave, not what they cost. GDPR, the EU AI Act, and on-prem inference rising from 12% to 55%.

Read the note
6 July 2026

Open-Weight Models Are Good Enough Now: Llama, Mistral and Qwen in Production

Open-weight models pass real production acceptance criteria in 2026. What Llama, Mistral and Qwen are each good at, and what 'good enough' actually means.

Read the note
6 July 2026

Private LLM vs API: Where the Break-Even Actually Sits in 2026

Self-hosting an LLM breaks even at roughly 2–5M tokens a day against frontier APIs, and 50M+ against open-model hosts. The real maths, with charts.

Read the note
6 July 2026

The Hidden Costs of Self-Hosting an LLM — and When Not to Do It

The GPU is the cheap part. Engineering hours, the 2.5–3× infrastructure multiplier, and the cases where you should stay on the API, with a decision flowchart.

Read the note
6 July 2026

Vibe Coding Rescue: What It Takes to Get an AI-Built App to Production

Your Lovable or Bolt app works, until real users arrive. What actually breaks in vibe-coded apps, what a rescue involves, and what it should cost.

Read the note
6 July 2026

Your API Bill Became a Line Item: Renting vs Owning Intelligence

When the LLM API bill becomes a board question, the answer isn't always self-hosting. How to think about renting vs owning inference, with the spend curves.

Read the note