About

Hi, I'm Devguy.Haagsoft is me.

I'm a full-stack solopreneur building agentic AI software in TypeScript. My speciality is the unglamorous part that decides whether AI features survive contact with users: reliability, evals and the harness around the model.

Sunlight through a young forest
haag — Dutch for hedge, and the first half of Den Haag, where I'm based. Small things, planted deliberately, that grow.

01Why Haagsoft

Haagsoft is a one-person studio on purpose. Instead of one big platform, it is a hedge of small, independently deployable apps — each on its own haagsoft.xyz subdomain — that stand alone or compound. Every one is sized to ship fast and prove its value before it earns more investment.

The same approach applies to client work: write the north star, cut it to the smallest honest slice, get it live, then let real usage and evals decide what comes next.

02What I'm good at

Agentic software you can actually rely on.

  • Agentic reliability

    Agents that follow the north star on the hundredth run as on the first. A written goal, typed tool contracts, bounded loops, and retries and fallbacks that are designed, not bolted on.

  • Evals

    Fixed datasets, graded runs and regression checks that score behaviour against the north star, so a prompt or model change is a measured decision instead of a hopeful one.

  • Harness engineering

    The layer around the model — context assembly, tools, memory, tracing, cost control — that turns an impressive demo into a product people can depend on.

  • Type-safe TypeScript, end to end

    tRPC, Drizzle and Zod from the database to the tool schema the model sees, with LangChain where it earns its place. One language, fewer seams for bugs to hide in.

03Working together

Spec, slice, measure.

  1. 01

    Spec

    A short written spec: the north star, the smallest slice worth shipping, and how we will know it works (including the evals).

  2. 02

    Slice

    A live URL early — usually within the first week or two — on hot-swappable infrastructure you own.

  3. 03

    Measure

    Evals and real usage decide the next slice, and I review every change. Autonomy and complexity only arrive when something real asks for them.

The principles behind this, and the stack I use, are on the home page. How the apps were actually built is in the blog.

Got an agent that needs to be boringly reliable?

devguy@duck.comgithub.com/oreillyross