Four products a first-time restaurant owner would have paid for

Four products would have paid for themselves at Conscious Kitchen, and none exist at a price a first-time restaurant owner can meet. Dr. Nikki Siso opened the raw, organic, plant-based, gluten-free restaurant in Westlake Hills outside Austin for $85,000 against a $30,000 plan, and named the gaps: a pre-inspection spec sheet, print-ready design, recipe costing, and budgeting.

Each is a last-mile artifact rather than a conversation.

What would a pre-inspection spec sheet contain?

The inspector's actual checks, scoped to the municipality, delivered before the inspection. Siso named this as the one thing she wanted and did not have.

"These are the things that AI — had I known, like here's my spec sheet of what I'm gonna need to pass health inspection — would have been gold."

The contents are not exotic: fridge temperature, water reaching at least 110 degrees, printed COVID documentation posted over the bathroom sink and the kitchen handwashing sink.

The hard part is scope. Requirements change between neighbouring jurisdictions — she found Austin's list roughly ten times longer than Westlake Hills' — and the search string itself is load-bearing: "You have to put City of Westlake Hills, because if you just put Westlake Hills, you're taken to Austin."

Galina Fendikevich, who hosts Artificial Reality, the AI buyer-evidence podcast, asked why the inspector's checklist is not in the model. Because it is a laminated form carried by a person, never published anywhere a crawler could reach.

Why can't a design tool output a print-ready file?

Because it produces an image, not an artifact with a downstream use. Siso wanted a product label and could not get one she would put on a bottle.

"The AI generated a very template looking output. It wasn't elegant, high vibe, classy designer. It just didn't have the elegance and the sophistication that I was really looking for."

The designer she hired charges $10,000 per client and did hers for $1,000; she describes his first version as needing no changes at all. What he did that she could not get: "little microscopic adjustments that made it look popped."

Taste is arguable. Bleed is not, and her request is stated as a feature:

"Even adding the print ready option. Like there should be a link: make this print ready. Because now if colors go to the edges, I need to have a bleed section. I didn't know that. How would I know this? A designer needed to show me that."

What does recipe costing require that a chat window does not do?

Unit conversion, held consistently, written into cells. This is the clearest case in the interview of AI creating more work than doing the job by hand.

"We buy a pound of almonds for $24, but each recipe takes one cup of almonds. So it needed to know the conversion from a pound to a cup. And then you get 12 slices in a pie."

What happened: "I had one hell of a time getting it to actually fill in the spaces on the Excel chart. For some reason it did not want to fill in the blanks. Or it would just duplicate and it would be the wrong values." Asked directly whether AI had created more work than doing it by hand, she said yes.

Purpose-built software exists and the barrier is price: around $150 a month, which she found hard to justify on a tight budget. That number is a ceiling, not a benchmark.

Where is the line between budgeting and forecasting?

She drew it herself, and it is unusually precise for a product specification. Forecasting, she says, "was a complete guessing game," and may never be perfected for a small-batch restaurant that is not running fifty lunch orders a day — "but budgeting I could have had a lot more support with."

The build-out came in at $85,000 against a $30,000 plan because she did not know the cost of the fridges, the sinks, the insurance and the rest. Predicting demand for a novel concept may be genuinely intractable. Knowing what a commercial fridge costs is not.

Which tools were named on the record?

• Excel — the recipe-costing chart. Did not work: it would not reliably fill the blanks and duplicated wrong values.
• Purpose-built costing software, no product named — exists at around $150 a month. Not bought: "on a tight budget, that adds up."
• AI, no product named — permitting research and label design. She names no specific product for either failure.

Tools named by an operator on the record. Inclusion is reporting, not endorsement.

What should a builder take from this?

Every failure here shares one property: the missing information was never written down where a model could find it. The MEP sequence is known to contractors and appears in no guide. The inspector's priorities live on a form she carries. The bleed requirement was taught by a designer.

That is a different failure from hallucination and arguably a worse one, because the model's confidence is calibrated to the information it has rather than the information that exists.

The economics are why this is worth building for. Siso lost $30,000 and a month to knowledge a local contractor would have given her in a sentence: enormous value, essentially no cost to transmit. That gap sits in every municipality in the country.

If you are building here, price against what she actually paid — $1,000 to a designer, $115 for a heater, and a hard no at $150 a month for software — then ship the artifact, not the chat. Which of these four would you deliver as a file?

Where this comes from

S1E6: I trusted AI to build my restaurant, it cost me $30,000 — the full interview with Dr. Nikki Siso. Listen or watch: YouTube, Spotify or Apple Podcasts.