Lovely Can. What's in It?

Everyone is a product designer now, myself included.

I mean.

Look at this glorious website.

I’m off to update my linkedin:

‘AI|Developer|Designer|Marketer|Ops|Fiance|Claude, can you finish this list?‘

A ‘founder’ can generate forty package directions before the manufacturer has answered the first email.

The colours are coherent.

The benefit line is punchy.

The can sweating in expensive-looking light.

There is a launch campaign, a product page and a video of people enjoying a drink that does not exist.

Then the less visual questions arrive.

What is the formula? Claude's mix must be right… Right?

Which ingredient delivers the claim?

Is the claim allowed? I asked Codex, it said yes…

Does it still taste right after twelve weeks? Can the co-manufacturer run it on the available line? What changes when the liquid meets the liner? Which allergens appear in the plant? What is the case configuration? Where is the GTIN? What is a GTIN anyway? Who can trace the affected batch if something goes wrong?

AI has made the front of D2C and CPG development dramatically faster.

It has not removed the proof required behind the pack.

The copy, formula and evidence have to agree.

“Source of protein” creates a compositional threshold.

A digestive-health line creates an evidence question. An ingredient that sounds useful in a prompt may change taste, cost, processing, allergen controls or the claims available on pack.

An AI system can help retrieve the rule and check consistency.

The business still owns the evidence, the specification and the consequences of getting them wrong.

Ingredients arrive within tolerances rather than as perfect database entries. Heat, pressure, mixing order, hold time and filling speed change the product. Packaging can change shelf life. A substitute ingredient can change flavour, label declarations and process behaviour at once.

This is where a concept becomes a specification. The manufacturer needs ingredient identity, quantity, process, tolerances, finished-product tests, packaging, yield and release criteria. The brand needs to know which changes require approval and which invalidate the claim or shelf-life work.

The mock-up does not contain any of that.

I would want AI helping with formulation and testing too. There is expensive work there worth reducing.

It still does not make the bottleneck disappear. It moves the scientist’s effort away from searching every combination and towards setting the constraints, validating the prediction, running the process and deciding what evidence is enough. A model can suggest a formulation that meets the recorded targets. The business has to prove the targets were complete and the manufactured product meets them.

The campaign tries to find every customer. The recall tries to find every unit.

Both depend on identifiers, batches, records and trading partners. Only one appears in the brand deck.

AI can improve exception detection, document review and trace searches. It cannot recover a lot code that was never captured or distinguish two products that share the wrong identifier.

That is what “get it on the shelf” contains: accurate master data, images, dimensions, pack hierarchy, price, retailer validation, lead time and a physical unit that matches the record.

AI can prepare much of the data. It can spot missing fields and compare an artwork file with the product record. The manufacturer and brand still have to supply the truth early enough for the retailer to use it.

Of course, I could choose a mature contract manufacturer, take a stock formulation, use standard packaging and launch direct to consumers. In that model, creative and demand testing may remain the real constraint. AI can make the experiment meaningfully faster and cheaper.

That can be a sensible way to validate demand. It should not be confused with owning a defensible product simply because the render is original.

The render can be generated before lunch. The product exists when the same formula can be made, labelled, traced, recalled and restocked, and when the promise on the front of the can remains true after everything behind it has done its work.