HS.
← Selected work

Case 03 / Execute

A Meal Full
of Love

Can one person use AI to test a small-business idea from product design through market execution?

Real commercial experimentOutcome still being tested

A business idea becomes useful only when reality gets a chance to answer it.

A Meal Full of Love started with a simple question: could a small local meal service be designed, built, and taken toward market without first building the organization that would traditionally support it?

AI made it possible to investigate, design, test, and build much more with one operator.

But it could not answer the most important question:

Will real customers buy—and come back?

01 / The business hypothesis

Can convenience and simplicity coexist at very small scale?

The initial idea was simple:

Home-style meals + predictable weekly menus + local delivery + simple ordering

But a meal business can fail even when customers like the food.

  • Too many menu variations increase kitchen complexity.
  • Small scattered orders can destroy delivery economics.
  • Separate customer channels can create duplicated work.
  • Too many products can overwhelm a one-person operation.

That question had to be tested through an operating model—not just a menu.

02 / Designing for constraints

Demand should earn complexity.

One core weekday menu

Office and Home customers share the same Monday–Friday core menu.

Different customer positioning. One production system.

Delivery boundaries

Defined local service area. Defined delivery window.

Order and delivery rules prevent individual small orders from silently consuming the economics of the business.

Shared infrastructure

Office and Home use the same underlying ordering operation.

Customer source remains identifiable for analysis. Reuse the mechanism. Separate the evidence.

Capacity protection

Additional lighter dinner options were intentionally limited rather than offered every day.

Demand should earn complexity.

Do not add complexity before the market justifies it.

03 / From idea to operating system

The experiment required more than a customer-facing website.

Product
What exactly is being sold?
Customer
Who is the first realistic buyer?
Pricing
What does the customer pay—and under what delivery conditions?
Production
How much variation can one kitchen actually support?
Ordering
What information is required to fulfill an order correctly?
Delivery
When does delivery remain economically and operationally reasonable?
Data
How can Office and Home demand be distinguished without duplicating the backend?
Controls
When should an order be accepted, confirmed, changed, locked, or rejected?

The website became the visible layer of a larger operating design.

04 / Human × AI execution

AI expanded execution. Ownership stayed Human.

The goal was not to ask AI to “start a business.”

Hongmin / Decision authority

  • Customer and product decisions
  • Pricing and operating constraints
  • Capacity boundaries
  • Risk decisions
  • Acceptance
  • Market release

Bumblebee

Investigate and challenge assumptions across customer behavior, pricing, senior-friendly UX, operating logic, and product design.

Codex

Translate frozen requirements into working implementation and tests.

AI expanded the amount of work one operator could investigate and execute.It did not become the business owner.

05 / From decision toward market

Built far enough to face customer behavior.

Product
Weekly meal structure and customer offer defined.
Operating model
Menu synchronization, capacity constraints, delivery rules, and customer-channel logic defined.
Customer experience
Office and Home customer paths developed for different contexts while preserving a shared operation.
Digital infrastructure
Ordering flow and supporting backend developed and tested.
Market materials
Customer-facing acquisition and pilot materials prepared.
Market entry
Prepared.

The product can now face the evidence that matters:

Customer behavior.

06 / The market gate

Build pass ≠ market pass

A functioning website does not prove demand.

A successful order-system test does not prove willingness to pay.

Positive comments do not prove repeat purchase.

An AI assessment does not prove product-market fit.

The commercial hypothesis has to survive evidence from the market.

Future evidence may include

  • Real inquiries
  • Real orders
  • Order size
  • Repeat purchase
  • Customer-segment response
  • Delivery economics
  • Operational burden

UnknownUntil sufficient evidence exists.

FailIf customers do not buy—or the economics do not work.

Failure cannot be rewritten as validation simply because the experiment was well designed.

07 / Minimum viable organization

How much organization is required before one person can test a real business hypothesis?

The experiment changed the question I was asking about AI.

Not: What business tasks can AI perform?

Instead:

How much organization is required before one person can test a real business hypothesis?

Moving from idea toward market requires capability across product design, research, branding, customer communication, software implementation, testing, operations, and analysis.

AI does not eliminate those capabilities.

It changes how many of them one person can temporarily assemble and direct.That can lower the organizational cost of experimentation. It does not lower the standard of market evidence.

08 / Current status

Real commercial experiment.
Outcome still being tested.

Product / operating model
Developed
Customer-facing experience
Developed
Digital ordering infrastructure
Built / tested
Market entry
Prepared
Sustainable demand
Unknown
Product-market fit
Not claimed

The experiment remains open to PASS / FAIL / UNKNOWN.No outcome will be rewritten afterward.

09 / What this case demonstrates

Build far enough to let reality answer the question.

This case is not primarily about food, websites, or AI tools.

It tests whether I can take an ambiguous commercial idea and move it through:

  1. Hypothesis
  2. Product
  3. Constraints
  4. Operating model
  5. Build
  6. Market
  7. Evidence
  8. Decision

AI increased the execution capacity available to one operator.

But the final authority moved somewhere AI cannot replace:

Reality.

Build far enough to let reality answer the question.Do not rewrite the answer afterward.

Selected receipts

What exists—and what remains unevidenced.

01

Operating Model

Shared weekday production logic, delivery constraints, capacity boundaries, and customer-channel design.

Evidence retained / not publicly displayed
02

Office Meal Experience

Customer-facing Office meal path and ordering model.

Evidence retained / not publicly displayed
03

Home / 60+ Experience

Senior-oriented customer path using the shared production system.

Evidence retained / not publicly displayed
04

Ordering Infrastructure

Shared ordering flow with customer-source separation and operational controls.

Evidence retained / not publicly displayed
05

Market Materials

Customer-facing acquisition and pilot materials prepared.

Evidence retained / not publicly displayed
06

Market Evidence

Outcome still being tested. Not yet evidenced.

Not yet evidenced