Don't replace expertise,
amplify it.

HAPPYCODE_CH

Project management + AI engineering. Kaiserstuhl, Switzerland.

Everybody needs to implement AI and they don't know how to do it. This firm is an example of expertise capture and implementation — translating business needs into developed deliverables by finding expertise it doesn't have and using it. This is the worst version of that capability we will ever see.

WORK

Projects as primary evidence.

Build backward from what the work demands.

01

Ask what the deliverable requires — scrutiny, ethics, attribution, verification.

02

Build the tools that answer each requirement.

03

Verify the tools actually do what they claim. Test them.

04

Run the pipeline and let it produce output.

05

Human reviews what the system cannot encode.

First 24 hours in a domain I've never touched.

01

Ask where the client's expertise moat lies.

02

Synthesize their processes into repeatable steps.

03

Understand goals and quickest-gain areas — triage first.

04

Audit their systems and their competitors'.

05

Fire up parallel agents to absorb the domain.

06

Sit with the team and extract the manual workflow.

Three layers. All load-bearing.

LAYER 1 — IDEATION

The agent builds; the human thinks. Where the agent's training would lead it to extend obvious capabilities, the human's synthesis creates the non-obvious move that differentiates the product.

VIS example — Four Whispers, the language-analysis system that differentiates VIS from any institutional research terminal, exists because the operator reframed filings as disclosure behaviour mid-build. The agent would not have reached that framing independently.

LAYER 2 — ACCESS

Agentic development makes capability-level work in a new domain accessible to a curious, directed operator. Expertise is earned through engagement with the work itself, not through years of credentialing before being allowed to start.

VIS example — the firm is not a quant shop. It is a directed operator using agents to absorb the analytical layer a quant team would otherwise provide.

LAYER 3 — VERIFICATION

The human asks the questions the system cannot encode — what is legal, what is ethical, what is publishable, what is true versus hallucinated. Those questions come from engagement with the pipeline itself, not from external credentialing.

VIS example — every paper the firm publishes passes through a gated pipeline the firm designed by asking, before building anything, what a publication would require to stand the test of scrutiny.

Expertise capture and implementation is domain-agnostic. The firm built its first case study in finance; the method transfers wherever tacit expertise is locked in individual practitioners and research burden is the bottleneck. The following are illustrative stories — examples of where the method could apply, not commitments to a specific direction.

STORY 01

Logistics

Workflow and document automation for mid-sized operational clients.

STORY 02

Construction

Permits, schedules, change orders, compliance — the document surface a Swiss mid-sized general contractor carries.

STORY 03

Farming

Tacit expertise locked in family operations. Regulatory and compliance paperwork is the drag. Swiss-specific opportunity.

If this method could help something you're working on, I'd like to hear about it. anthony@happycode.ch

FOR RECRUITERS AND EMPLOYERS

Anthony Calek. Real-estate portfolio management, 1995–2021. Computer science coursework at DePaul University. Based in Kaiserstuhl, Switzerland since 2020. Full history available on LinkedIn.

WRITING

How the firm thinks about the work.

A mix of firm-level reflection and subject-matter writing from the case studies in flight. The work thinks in public.

Further posts will appear here from the Substack feed.

The publication section pulls live from the connected RSS source. See happycode.ch/ for the full feed.

I know how to find expertise that I don't have and use it.

HAPPYCODE_CH · KAISERSTUHL, SWITZERLAND

anthony@happycode.ch