Lk

Lukas Hartrumpf · Applied AI Architect

Demos are cheap.
Systems are rare.

I lead Engineering Intelligence at nerdware. The work is the distance between an agent that impresses in a demo and a system a company can actually run on.

See the workWrite to me
119LkLukretium

01 · What I do

I don't type much code anymore, and I think that is the point. I run agents, and I build what sits around them: the context they work from, the guardrails they stay inside, the gates that decide whether their work ships. The craft moved up a level. Whether the result holds did not.

02 · How work moves now

01

02

03

04

05

  1. sentry

    TypeError: user is undefined

    sentry · incoming
    + checkout/README · ADR-007
    fix/checkout-null-user
    · tests· evals· policy
    ✓ merged on its own
  2. logs

    OOMKilled worker-3, again

    logs · incoming
    + runbooks/memory.md
    chore/worker-memory
    · tests· evals· policy
    ↺ sent back: memory bump, no root cause
  3. jira

    PROJ-497: clean up old accounts

    jira · incoming
    + ADR-019 · data-retention.md
    feat/account-cleanup
    · tests· evals· policy
    → comes to me: deletes production data

Signals arrive from everywhere: a Sentry error, a log alert, a Jira ticket someone wrote in a hurry.

One merged on its own. One went back with a reason. One needed me.

03 · What I hold to

  1. 01

    Measure before you model.

    A brief asked for 16,659 archived emails to be filed against housing units. Measuring first showed only 16 percent could be mapped at all. The requirement changed before a line was written.

  2. 02

    Engineer the harness, not the feature.

    The shift I ask of a developer: stop writing the feature, build what writes it reliably inside this one codebase.

  3. 03

    One door, not ten.

    Every LLM call in the company goes through one gateway. Budgets, models and guardrails live in one place instead of in every team.

  4. 04

    Old skills, new leverage.

    A decade of event-driven systems and enterprise integration is the reason the agent work reaches production.

04 · Numbers, not adjectives

Four things at once, most of them in production.

The whole picture
2,300
Commits since November 2025 on one agent-built production system, and counting
8
Engineers led at peak
4
Client engagements running in parallel
300,000
Daily users on a platform I have been building since 2023

Lab notes

Case studies drafted by a pipeline I built from my working notes. The patterns are real, the incidents are constructed. How it works

All notes