×the-lab.case
REDACTEDreal work, anonymized — category + problem naming · relative metrics only

Experiments · always running

The lab: trust, autonomy, and an AI content factory

Side experiments that keep my product instincts honest — premium e-commerce, an autonomous trading system, and a media pipeline that makes real videos.

role · Builder — nights, weekends, curiositystatus · Rotating cast; lessons harvested continuously

§01The problem

A PM's instincts decay without skin in the game. The day job supplies depth in one domain; the lab supplies breadth — each experiment picked because it stresses a muscle the day job doesn't: consumer trust, autonomous decision-making under real losses, production media at volume.

[ interactive diagram loading… — three experiments, three different muscles ]
fig — three experiments, three different muscles

§02The thinking

Premium jewellery e-commerce is a trust problem wearing a storefront: high-consideration purchases from an unknown brand, where every pixel either builds or burns credibility. It taught me what 'premium' costs in performance budgets, photography, and copy discipline.

The autonomous trading system is the purest autonomy lab available to an individual: real money, unforgiving feedback, and every safety-rail question AI products face — position limits, kill switches, drawdown honesty — with none of the abstraction. It is where I learned viscerally why 'the system was confident' is not a defence.

The content pipeline — scripts to voiceovers to rendered video — is a study in AI production quality at volume: where automation genuinely holds up, where human taste must intervene, and how quality degrades quietly when nobody owns the last mile.

+ Experiments with real stakes

toy projects and tutorials

# instincts only calibrate against consequences — a demo that can't lose money or customers teaches nothing about either.

+ Harvesting patterns into the day job

keeping the lab separate

# the kill-switch thinking from trading and the trust discipline from e-commerce show up directly in how I spec AI features now.

§03What shipped

The lab runs as a rotating portfolio: build until the lesson is extracted, write it down, move on. The artifacts are real — a live storefront, a trading system with actual P&L, published videos — but the product is the calibration.

[ interactive demo loading… — the rotating portfolio ]
fig — the rotating portfolio

§04Outcomes

Autonomy, calibrated

hard-won intuition for when AI systems deserve independence — and when they must ask

Trust, priced

what premium actually costs to signal, in pixels and patience

Taste, defended

where human judgment must stay in AI production pipelines

§05Reflection

The lab is why I can disagree with an engineer about retrieval design or with a designer about trust signals without borrowing anyone's authority — I've paid tuition in all three currencies.

$ cat takeaways.txt

  • Breadth is a discipline, not a hobby — pick experiments that stress unused muscles.
  • Real stakes or no lesson.
  • Extract the pattern, then let the project go.