> **Approach — how I work & how I earn your trust — Silou**
> Canonical: https://silou.dev/approach
> Generated: 2026-07-19 — markdown twin of the HTML page (the HTML is the source of truth).

# How I earn your trust

*Approach*

I'd rather show competence than claim it. So instead of a wall of metrics, here's exactly how you can judge whether I'm worth your time — before you commit to anything.

## 1 · Live, on your screen

The fastest honest proof is unfakeable: we screen-share and I improve your Claude Code setup while you watch — you keep the keyboard, so there's nothing to grant me and no access to provision. Work on your own repo if you're comfortable, or a representative example if you're not; either way you leave with a real win — a command, an agent, a hook, a model-routing decision — in about an hour. No portfolio required, because the work happens in front of you. That's what the [Power Hour](https://silou.dev/sessions) is for.

And it isn't only for developers — with product, ops, or non-technical teams we do the same thing in your real tools and workflows, not a repo. The proof holds either way: you watch me improve how you actually work.

## 2 · The method, in the open

I write up the thinking so you can check it before we ever speak. The through-line is the **harness thesis**: the model is the engine; the context, the agentic layer (commands, agents, skills), hooks for trust and control, model-offloading for reliability and cost, and **evals plus observability** so you know it actually works — that's the harness. Most teams optimize the engine and ignore the harness.

The loop I work by is **plan → build → review**: scope the work first, let the agent build against that scope, then review — with evals so "it works" is something you measure, not a vibe. Everything else is matching each piece of work to the lightest layer that can carry it — inline, a command, an agent, or a packaged skill. Early working notes live on the [Lab](https://silou.dev/lab) page — read them and decide for yourself whether I think clearly.

The payoff of getting all that right is **loop engineering**: work that runs on a schedule (cron), unattended, without you in the loop — because the hooks, evals, and observability are what make it safe to step away.

## 3 · Production experience, honestly framed

My patterns aren't slide-ware. They come from building and running a full agentic harness in **a scaled, production enterprise context** — review agents, cloud operations, CI debugging, observability, and strictly-typed, functional code at scale. That work is under NDA, so I'll describe the *category* and the transferable lesson, and name zero clients, products, or stacks. If a pattern only works because of something I can't show you, I'll tell you that too.

> **Radical honesty as method.** I credit my sources — a lot of my depth came from **[IndyDevDan](https://indydevdan.com/)** ([YouTube](https://www.youtube.com/@indydevdan), [GitHub](https://github.com/disler)) for agentic engineering and **Matt Pocock's [AI Hero](https://www.aihero.dev/)** for LLM fundamentals — resources I still learn from and recommend. I don't invent metrics, I don't promise 10×, and I name the tradeoffs — including where my own approach has limits and where provider terms bite. That candour is the point: it's the trust the rest of the work rests on.

## Want to see it on your setup?

The shortest path is a live teardown. Bring your setup and your biggest pain; leave with a concrete improvement.

[See sessions & pricing](https://silou.dev/sessions)