AGI data analyst: from spreadsheet to clear insight.

Most analysis is spent before the analysis. The file has three date formats, two spellings of the same customer name, a column of totals that does not add up and a sheet of notes that only the person who made it understands. Cleaning is not a preliminary chore; it is where most of the judgement lives, because every decision about what to keep and what to correct changes the answer that follows.

The agent profiles the data before it touches it. It reads the columns, the types, the ranges and the gaps, and it reports what it found: the missing values, the duplicates, the outliers that look like entry errors and the fields that will cause trouble later. That profile is the first deliverable, because a dataset that is not understood cannot be analysed honestly.

Analysis comes next and it is written in code that can be read. Queries and scripts are saved in the project, not hidden behind a button, so the result can be reproduced by anyone with the file. Charts are built to answer a question, with axes and labels that a person outside the work can follow, rather than to fill space with colour. The report is written last, in plain language, with the findings, the caveats and the next question.

The distinguishing feature is that the working is visible. A number in the summary can be traced to the formula, query or step that produced it, and the assumptions behind each step are written down rather than assumed. That is what turns analysis into evidence: a reader can disagree with the method and check the arithmetic instead of having to trust the conclusion.

The honest limits are worth naming. Data quality in is data quality out, and a clean analysis of the wrong sample is still wrong. The agent can describe what the data shows and what it does not, but it cannot supply business context it was not given, and it will not dress a correlation up as a cause without the evidence to support it.

The profile also sets up the questions. Once the columns are understood, the agent proposes the analysis that the data can actually support and says what it cannot answer with the information available. That step alone prevents the most common failure in reporting: answering a question the dataset was never designed to address, with a chart that looks convincing and a conclusion that cannot be defended.

A report that says revenue grew without saying how revenue was defined is not a finding. It is a sentence with a number in it. The agent writes down the definitions it used, the rows it excluded and the period it measured, so the reader can see the shape of the claim before deciding whether to believe it.

Reproducibility is part of the standard. The cleaned data, the code and the report live together in the project folder, which means the analysis can be run again when the data is refreshed or the method is questioned. That is a higher bar than most spreadsheet reporting clears, and it is the bar that makes the work defensible in a meeting where someone asks how you know.

Automation extends the same discipline over time. A report that runs on a schedule keeps its definitions and its checks, so the second edition is consistent with the first rather than a fresh interpretation of a new export. When something changes, the run says so instead of quietly producing a different kind of number.

The limits stay honest. The agent can analyse what it is given, describe what it shows and write the caveat that belongs with it. It cannot invent the context that makes a number important, and it cannot decide for you what the business should do about the result. The judgement stays with the person who owns the decision.

Delivery matters as much as the calculation. A report is read by people who did not do the analysis, so the agent writes the summary first, the method second and the appendix third, with the numbers that support each claim close to the sentence that makes it. A reader should be able to stop at any point and still know what the document is telling them and how much to trust it.

Data is one of unlimited domains, and it is rarely the last step. The analysis feeds a report, the report feeds a decision, the decision feeds a page or a plan. Zeus AGI carries the context forward, so the definitions used in the analysis are the same ones used in the writing that follows. Good analysis is remembered by the decision it changed, not by the chart.

This is one example. The skill and MCP system means the next domain is a request away. Zeus AGI acquires new skills and connects MCP servers, so a capability it does not ship with today can be added when the work calls for it. Describe the outcome on the anything page, read how invention is handled in beyond, and see the full map in capabilities. The data analyst inherits the definitions instead of re-deriving them.

The list is not a boundary. Unlimited domains are examples of what the intelligence can do now. The real constraint is the quality of the data and the quality of the question, and both are worth more attention than any chart. That is why the profile comes before the picture.