Prototype, early access or available? A clearer way to report AI progress
Product status is not a badge. It is a promise about what another person can do today.
AI products are unusually easy to demonstrate before they are ready to use.
A model can produce an impressive answer in a controlled example. A prototype can run on the developer's machine. A private pilot can solve one customer's workflow. A landing page can show the final product before onboarding, support, billing or recovery exists.
All of those can be real progress. They are not the same stage.
At Kvantim we build three different AI directions, which makes status clarity more important. Kvantia AI, Kvantia Harness and Kvantim LLM should not each invent a convenient meaning for "live". We use one practical vocabulary across the ecosystem.
Available now
Available now means the intended user can access the product and complete the core flow described on the page.
The conditions around that access still matter. Supported operating systems, hardware requirements, price, licence, onboarding and known limits must be visible. If the page says the product creates a report, a real user should be able to reach that result through the offered path.
"The code exists" is not enough.
Early access
Early access means the product works, but access is staged and the team remains involved.
An early-access user may need an invitation, a supported device or direct setup help. Some functions may be incomplete. The invitation should state exactly which capability is being tested, what data is appropriate and how feedback is handled.
Early access is not a polite synonym for a waitlist. If nobody outside the team can use the core flow, the product is not in early access yet.
Working prototype
A working prototype proves that a core idea can run.
It may work on one machine, with manual setup, on a narrow test. It can still be valuable evidence. The correct next question is: what exactly can be demonstrated, and under which conditions?
For Kvantia AI, for example, a working local conversation tells us something important. It does not automatically prove persistent memory, simple onboarding or wide device support. Those remain separate milestones.
In development
In development means the team is actively building the capability but a user cannot rely on it yet.
This label should be attached to a specific outcome, not used as a permanent escape hatch. A useful development item names its owner, current work, known dependency and the gate that will move it to prototype or early access.
Planned
Planned means the direction exists, but it is not an executable product claim.
Concepts, marketplace cards, future integrations and intended model specialisations belong here until work has produced a stronger status. Planned items should not use the same purchase buttons and visual weight as available features.
"Tested" and "validated" need an object
Two other words appear often in AI updates: tested and validated.
They are incomplete without more information.
Tested on what? Against which baseline? With how many cases? On which hardware? By the team or by an external user? Did the test measure quality, safety, latency, cost or only whether the program ran?
Instead of "the model is validated", a useful update says: "The training pipeline completed a small run on a named data scope, and the resulting model improved a named evaluation from a stated baseline to a stated result. The full training run has not started."
The narrower sentence is stronger because someone can evaluate it.
One source, many pages
Status also creates a content-engineering problem.
If the home page says persistent memory is available, the roadmap says it is next and the pricing page charges for it today, better copy will not fix the conflict. The status has to live in one shared source used by every page.
For each capability, that source should contain:
- product and capability name;
- current status;
- public meaning;
- evidence link;
- limitation;
- next milestone;
- date updated.
When the milestone is reached, one change can update the home page, product page, roadmap and ecosystem overview together.
Honest status is part of the product
Clear status may feel less exciting than a page full of present-tense promises. For an AI company, it is part of trust.
The user needs to know whether they can start. A pilot partner needs to know what they are testing. An investor needs to distinguish evidence from an intention. The team needs a shared language that resists accidental hype.
Progress is still worth celebrating. We simply want each claim to carry the right weight.
Available now. Early access. Working prototype. In development. Planned.
Five labels, one practical question: what can another person use or verify today?