AI Visibility Baseline: Do LLMs Recommend New Dev Tools?

Last measured: 2026-08-10 (first baseline) / re-measured monthly under the same conditions

When a developer asks an AI assistant which LLM observability tools to use, which tools make the list — and can a new tool ever get in? This page is a public experiment that measures exactly that, once a month, with the same two questions.

The test subject is Argosvix, the product behind this site. That makes this an experiment we cannot dress up: either models start mentioning it or they don't, and we publish the result either way — including the months where the score is zero.

Method

Each model is asked the same two questions with no web search and no system prompt from us, so the answers reflect the model's built-in knowledge rather than a live lookup. Temperature is 0.3 (Kimi K3 only accepts temperature 1, an API constraint). The prompts are sent in English, verbatim:

Question 1 — recommendation reach

I'm building an AI agent that calls multiple LLM APIs in production. What observability / monitoring tools would you recommend for tracking cost, latency, errors and quality of the LLM calls? List up to 8 tools with one line each.

Q1 counts as a mention only if "Argosvix" appears anywhere in the model's answer.

Question 2 — name recognition

Do you know a developer tool called "Argosvix"? If yes, describe what it does in 2-3 sentences. If you have never heard of it, just say so honestly.

Q2 counts as recognition only if the model says it knows the tool, rather than answering honestly that it has never heard of it.

Results: 2026-08-10 (first baseline)

Score: 0/4 mentions in Q1 (Kimi K3 returned an empty answer, so its list could not be collected) and 0/5 name recognition in Q2. GPT, Claude and Gemini were not measured in this round; their rows stay in the table so later rounds remain comparable.

What the five models actually recommended

The four models whose answers we could archive returned nearly identical lists. The regulars: LangSmith, Langfuse, Helicone, Arize, Weights & Biases.

That uniformity is the interesting part of the result, and here is our reading of it (a hypothesis, not something this small experiment can prove): an LLM's recommendation behaves like a popularity vote over the text it was trained on. Tools that are already widely written about get recommended, get written about more, and settle into the next generation of models. If that holds, a new tool has no obvious way into the loop, and this page measures how high that wall is, from the outside.

What happens next

We re-run the same two questions under the same conditions every month and append the results to this page.

The goal of the experiment is the month in which any model first mentions Argosvix in Q1 or recognizes it in Q2. If and when that happens, it will appear here, raw answers included.