AI for a Well Drilling Company: Drilling Logs and Permit Records

An AI agent for a well drilling company can keep drilling logs, permit status and customer service history in one connected record, so the office can answer a permit question or a warranty call without rebuilding the job from memory. The agent works from the documents the crew already produces, tags each entry with its source, author and date, and flags gaps for a person to resolve. Its real value depends less on writing speed than on whether the records stay complete, correctable and easy to retrieve years after the well was finished.

Key takeaways

• Treat each well as a record with a lifecycle: inquiry, permit, drilling, completion, inspection and every service call after that.
• Keep the driller's original log as the source of truth, and let the agent structure it rather than rewrite it.
• Track permit status separately from permit documents, and give every open step a named owner and a due date.
• Correct records by appending a dated note, never by overwriting the earlier entry, so the history of changes survives an audit.
• Start with one role, one approval boundary and one success criterion before connecting more tools.

What does a well drilling office need to keep on record?

A well is a long chain of decisions made by different people in different places. The sales contact takes the first call, the office prepares the permit application, the crew drills and completes the well, an inspector may visit, and the homeowner or property manager calls back years later about a pump, a pressure problem or a warranty question. Each stage produces its own documents, and most of the trouble starts at the handoffs between them.

Before you deploy anything, map the lifecycle on one page. For each stage, write down the document that proves it happened, the person who creates it, where it is stored and who needs to read it later. This map becomes the memory the agent is allowed to use. Without it, an agent will summarize whatever it finds, and what it finds is often a folder of photographed pages, a text thread and a spreadsheet that disagrees with both.

How should drilling logs be captured and structured?

The driller's log is the most valuable record in the file and the easiest one to damage. It typically describes depth intervals, formation, water-bearing zones, casing depth, grout placement, static water level, yield and any bit or equipment changes, with times attached to some entries. Many crews still write these by hand on site, which means the raw record is an image and the structured record does not exist yet.

An agent can turn that image into a fixed set of fields without changing what the driller wrote. A reasonable workflow looks like this:

• Attach the original photograph or scan to the well record before any extraction happens.
• Extract depth intervals into a table with a start depth, an end depth and a formation description copied as written.
• Check that intervals are contiguous and that depths do not decrease, and flag any mismatch for the driller instead of correcting it silently.
• Record the static level, yield and test duration with the test date and the person who measured them.
• Mark every field the agent could not read clearly, so a reviewer sees the gap instead of a guess.

The agent should not infer geology that the driller did not record. If a formation is unclear in the image, the entry stays unclear until the driller confirms it. That rule is what makes the structured log trustworthy when someone reads it years later.

How do you track permits without a spreadsheet nobody trusts?

Permit tracking fails quietly. An application is submitted, a reference number arrives by email, an inspection is requested by phone, and six weeks later nobody can say whether the completion report was filed. The fix is a status record for each permit with a small number of states, such as drafted, submitted, approved, inspected and closed, and a rule that a state change requires a document or a named confirmation.

A practical setup for the permit side:

• Store the permitting authority, the reference number, the submission date and the contact name for each permit in one record.
• Link every permit to its well record and its drilling log, so the two cannot drift apart.
• Let the agent list open permits with the next required step, the responsible person and any step that has passed its expected date.
• Require a person to approve every filing, inspection request or correspondence sent to an authority. The agent drafts, the office sends.
• Keep the rules for each jurisdiction in a short reference note maintained by the office, and have the agent cite that note rather than relying on general knowledge.

The last point matters because permit requirements vary by state, county and sometimes by well type, and they change. An agent that answers from memory will eventually give an outdated answer with full confidence. An agent that cites the office's own reference note, and shows the date that note was last reviewed, gives the reviewer something to check.

How should customer service history stay attached to the well?

A service call is only useful if it can be found from the well. Owners change, pumps are replaced, and a technician who visited in one season may have left the company by the next. The customer record and the well record need to point at each other, so that a call about low pressure can show the original completion report, the pump test, the installed equipment and any previous repair in one view.

The agent can summarize that history for the person answering the phone, but it should keep three things separate: what the customer reported, what the technician observed and what the office decided. When these blur together, a warranty decision looks arbitrary later. Use the same discipline for warranty terms. The agent may show the terms attached to the job, but a person decides whether a claim falls inside them.

Keep personal data to what the task needs. A service thread does not need the owner's full history of unrelated contacts, and the agent does not need access to billing details to answer a question about a pump pressure reading.

What makes a record usable for a future audit?

An audit does not ask whether the records exist. It asks whether they can be trusted: who wrote them, when, from what source, and whether anything changed afterward. Records that were convenient to create can fail that test easily, especially when an assistant has been editing them quietly.

Use this checklist for every well file:

• The original source, such as a photograph, email or signed form, is attached to the entry that depends on it.
• Each entry shows the author, the date and the tool or person that created the structured version.
• Corrections are appended as new notes with a reason, and the earlier value remains visible.
• Agent actions are logged with the input it received, the output it produced and the person who approved it.
• The export format can be read without the tool that produced it, such as a dated PDF bundle plus a plain spreadsheet.
• A retention rule states how long each record type is kept, and a person owns that rule.

The logging item deserves attention. Observability practice in software is built on the idea that you can reconstruct what a system did from its recorded signals, and the same idea applies to an agent. If you cannot say which document an agent read before it drafted a completion summary, the summary cannot be defended later. The guidance in the [NIST AI Risk Management Framework](https://www.nist.gov/itl/ai-risk-management-framework) is a useful reference for setting up that kind of governance and documentation. Teams that already run formal quality processes will recognize this as the same discipline described in AI for Quality Management: Nonconformances and Audit Prep, applied to wells instead of production lines.

Who decides what the agent may change?

Permissions are where a well records system either stays safe or becomes a liability. The agent should read the logs, permit records and service history it needs for its role. It should write only to the drafts, status notes and summaries that the office has defined, and it should never change a signed log, submit a filing or promise a warranty outcome on its own. Authorization guidance such as the [OWASP Authorization Cheat Sheet](https://cheatsheetseries.owasp.org/cheatsheets/Authorization_Cheat_Sheet.html) describes the general principle: grant each identity only the access its function requires, and check that access on every action.

An illustrative scenario: a crew completes a well on a Thursday afternoon. The driller uploads a photograph of the handwritten log and a pump test sheet. The agent structures the intervals, flags one depth that does not match the total, and drafts a completion summary for the office. The office manager confirms the flagged depth with the driller, approves the summary and marks the permit as ready for the completion report. Several months later, the customer calls about reduced flow. The service coordinator opens the well record, sees the original log image, the confirmed depths, the pump test and the note that explains the flagged depth, and can answer the call within minutes.

Be Human's managed approach begins by defining that kind of role in a first engagement: one useful job, the memory and tools it needs, the approval boundaries and an observable success criterion, such as whether a reviewer can locate the source of any completion summary within a short, agreed time. Ongoing operations can then include monitoring of the agent's outputs, recurring evaluations against real records, permission reviews when people or tools change, and controlled updates to the workflow. The aim is to keep the agent's memory and its records under the company's control, not inside a tool that the company cannot inspect.

If you want to find where this kind of record work would pay off first in your own operation, take the free Be Human AI diagnostic to identify the first task worth handing to an agent.

Frequently asked questions

Can an AI agent read handwritten driller logs? It can extract text from photographs of handwritten logs, but accuracy depends on the handwriting, the lighting and the layout of the form. The safe pattern is to keep the original image attached, extract fields into a structured table, and have the driller confirm any entry the agent marks as unclear. The agent should not fill gaps with plausible geology.

Does this replace the filings the company must submit to the authority? No. The agent can track which filings are due, draft the documents and remind the office of open steps, but a person reviews and submits every filing. Requirements differ by state and county, so the office should maintain its own reference notes and treat any agent answer about a requirement as a draft to verify.

Who owns the records the agent creates? The company that holds the well files should own the records, the source documents and the change history. Ownership should be written into the engagement, along with the export format and the retention rule. A record that can only be read inside one vendor's tool is a weak record for an audit.

How long should well records and agent logs be kept? The answer depends on the record type and the rules that apply in your jurisdiction, so it should be set by the company with its legal or compliance advisor, not by the agent. Once the retention rule is decided, the system should apply it consistently and show who owns it. The rule should cover the source documents, the structured entries, the correction notes and the logs of agent actions.

Be Human editorial guidance. Examples are illustrative unless explicitly attributed. Sources provide background; implementation choices depend on your workflow.