The modern developer's workflow has become a symbiotic loop of prompting and refining. For most, this means using LLMs to squash a stubborn bug, refactor a legacy function, or generate boilerplate for a new API. It is a utility of efficiency. However, a new case study suggests that when these agentic workflows are applied to unstructured historical data, the result is not just faster coding, but the resolution of academic cold cases that have remained frozen for over a hundred years. This is the territory Tom Di Mino, a self-taught AI engineer and amateur linguist, has entered by attempting to crack Linear A, the enigmatic Bronze Age script of the Minoan civilization.
The Agentic Pipeline for Ancient Corpora
Deciphering a dead language is traditionally a grueling process of manual comparison and intuitive leaps. Tom Di Mino bypassed this bottleneck by deploying Claude Code, Anthropic's terminal-based coding agent, to build a sophisticated research infrastructure. Rather than writing a single script, Di Mino used the agent to develop a suite of Python tools designed to ingest and manipulate two massive digitalized corpora of Linear A: the GORILA and SigLA databases. These databases contain the fragmented remains of a civilization's record-keeping, but their sheer volume and lack of structure make manual cross-referencing nearly impossible for a human researcher.
By leveraging Claude Code to automate the querying and systematic organization of these datasets, Di Mino was able to perform hypothesis testing at a scale previously reserved for institutional research teams. The output of this AI-driven pipeline is substantial. Di Mino has proposed the readings for 40 distinct symbols, 13 of which were previously unknown to the linguistic community. Furthermore, the process yielded solutions for five symbols in Linear B, the successor script that was deciphered decades ago but still contained lingering ambiguities. The tangible result of this work is a vocabulary list of 408 Linear A terms translated into English and a nine-page draft paper titled Ya Diktu: Grammar of the Minoan Peak Sanctuary Libation Formula. This work is currently undergoing peer review by linguistics experts at Rutgers University and the University of Cambridge.
From Code Generation to Research Automation
The true shift in this story is not that an AI wrote Python code, but that the AI agent functioned as a bridge between a raw dataset and a theoretical breakthrough. For a century, the academic consensus treated Linear A as an isolated system, a linguistic island with no known relatives. Di Mino used his AI-powered toolkit to challenge this premise, proposing instead that Linear A is an extinct Semitic language, making it a distant ancestor to Biblical Hebrew, similar to how Latin serves as the root for the Romance languages.
The breakthrough occurred when the agentic workflow identified a specific pattern in the data that matched the morphology of Semitic languages. The critical clue was the verb root nawaya, meaning to dwell. By combining the symbol *301 with the symbol na, Di Mino identified a pattern that aligns perfectly with the N-W-Y tri-consonantal root system characteristic of Hebrew and Akkadian. This specific linguistic marker allowed him to determine that the inscriptions were not mere accounting ledgers, but prayers dedicated to a goddess.
This represents a fundamental pivot in how we perceive LLM-based coding agents. We are moving past the era of the AI as a digital scribe and into the era of the AI as a research collaborator. The agent does not possess the linguistic intuition to guess the root of a dead language, but it possesses the computational stamina to cross-reference thousands of data points across multiple databases to see if a specific hypothesis holds water. The tension has shifted from the ability to write the code to the ability to ask the right academic question. The engineer's role is no longer to manage the syntax of the Python script, but to interpret the patterns the AI extracts from the noise of history.
This convergence of agentic AI and domain expertise suggests a future where the barrier to entry for complex academic research is no longer the size of one's research team, but the precision of one's hypotheses.
