Tokenmaxxing treated AI usage as evidence of productivity. Now “vibe coding” risks becoming a catch-all for projects involving AI, regardless of how much engineering or research went into them.

In both cases, the label gets ahead of the evidence.

Not everything done with AI is vibe coding

On November 10, Polymarket called AI researcher Pietro Miotti a “vibecoder” after he reported using 48 agents and NASA lidar data to identify a possible Maya archaeological site in Mexico. [1]

Miotti describes giving agents defined visual-search criteria, reviewing possible structures, checking existing research and inspecting the underlying lidar data. That is a directed research workflow, not simply prompting an app into existence. The site still needs archaeological confirmation, and the agent execution has not been independently audited.

“Vibe coding” originally referred to generating software while largely disregarding the code itself. Martin Fowler has distinguished that from agentic engineering, where people still inspect and take responsibility for the work. The terms should not be interchangeable just because both involve AI. [2]

The interesting part of Miotti's experiment is not the number 48. It is what a person can investigate by organizing existing tools and data into a working process.

I have seen a version of this in ODIN

When the war involving Iran, Israel and the United States began in early 2026, I used ODIN to track the conflict through public reporting as events unfolded. I connected reports on military operations, diplomacy, shipping and energy markets into an evolving timeline, compared competing accounts, and retained unresolved questions.

My background is in business operations, not military intelligence. AI made that kind of investigation more accessible. But I was not collecting battlefield intelligence, and a source appearing in ODIN did not make every reported claim independently true.

The distinction matters for both examples. A possible structure in lidar data is not yet an established archaeological discovery. A source-linked conflict timeline is not independent confirmation of every event. Each can be valuable research without claiming more than its evidence supports.

Capability still needs a standard

I have written before about why an agent should not be sufficient authority for judging its own work, and why completed work should leave useful knowledge behind. This is the same underlying problem: new capability is valuable when the evidence and decisions survive the initial demonstration.

AI can help people investigate problems far outside their original disciplines. That is worth recognizing. The next question is what they can substantiate, what remains unknown, and whether the result can be examined and improved.

The measure is not how many tokens or agents were consumed. It is what the work actually established.

Sources

  1. Polymarket, November 10, 2026. Original X post describing Miotti as a “vibecoder”. The caption is verified as Polymarket’s wording, not confirmation of an archaeological discovery. Pietro Miotti, October 2026. First-person Maya lidar investigation. Archaeological confirmation and exact agent-runtime costs remain unverified.
  2. Martin Fowler, May 21, 2026. Vibe Coding. Distinguishes vibe coding and inspected agentic engineering.