--- title: "Which of your land leases expire in the next 60 days?" description: "Land teams lose acreage to expirations and Pugh clauses nobody had time to read. Here's a Decision Agent that reads county records and cites the source." url: "https://www.datarobot.com/blog/oil-gas-landman-agent/" authors: - "DataRobot, Value-Driven AI" published: "2026-11-09" updated: "2026-11-09" category: "AI Across Industries" language: "en-US" --- # Which of your land leases expire in the next 60 days? Name the leases on your books that reach the end of their primary term in the next 60 days. Now name the ones carrying a Pugh clause. If your lease database holds the dates, you can probably build the first list. The second often means a landman pulling instruments, because a lease database only knows what someone keyed in. That gap is where acreage can slip away. Picture a lease reaching the end of its primary term. In the original instrument, typed decades ago and assigned twice since, there’s a Pugh clause that releases the depths below the producing formation. Nobody on your team caught it, because nobody had the hours to pull that instrument this month. The deeper rights went back to the mineral owner, and another operator’s landman leased them first. You find out when their filing shows up. Nobody was careless. The work is built that way. ## Your database only knows what someone keyed in Your lease database is the system your team trusts, and it should be. The trouble is coverage. It holds what someone had time to enter, and the details that decide who holds what aren’t always among them. When the question gets specific, like whether a lease has passed its primary term or which depths a Pugh clause has released, the database points to an instrument and a person goes and reads it. That reading is where a landman’s week goes. A lease is almost never one document. It’s the original instrument, the assignments that moved it, and the deeds underneath the ownership, each recorded separately, often decades apart, in whatever format the county used at the time. Much of it was never born digital. It’s scanned paper, and the legal descriptions take a trained eye to read correctly. So the questions that matter most are the slowest to answer. Which leases expire in the next 60 days? Which depths are open? What drilling does it take to hold what we already have? Each one turns into days of pulling records by hand. Covering a new county makes it worse: you either hire more landmen, who are hard to find, or accept that the work moves more slowly. Meanwhile the clock on every lease keeps running. ## What the landman agent does The landman agent reads the county instruments your team already relies on and turns them into a record you can ask questions of. It works in three steps. It reads the record. Scanned leases, deeds, and assignments come in from your own document storage, including instruments going back to the 1950s. The agent pulls the fields a landman would: ownership, primary term, start and expiration dates, Pugh clauses, and the legal description. Those fields land in a database that sits alongside your existing lease database, so the record draws from the instruments themselves. The agent also maps each lease to the wells on it, so the question of what’s holding a lease starts from one connected record instead of two files someone has to line up. It answers with the source cited. A landman asks, in plain English, which leases expire in the next 60 days or what a specific lease grants and to whom. The answer comes back with the source instrument cited. Your landman checks the page instead of rereading the whole file, which means the time goes to confirming and acting instead of hunting. It keeps your landmen’s corrections. When a landman fixes an extracted field, that correction permanently overrides the extraction. Regression tests run across extraction runs, so a fix stays fixed as new instruments come in. The record gets more reliable every time someone experienced touches it. The judgment stays with your people. The agent doesn’t decide which lease to extend, which depth to pursue, or where to drill. It doesn’t talk to mineral owners or sign off on title. It makes sure your land team is deciding from the record, with the source in hand, instead of waiting days for someone to pull it. ## A Decision Agent, built for one hard problem The landman agent is what DataRobot calls a Decision Agent: an agent built around a decision that moves a number the business already reports, working across the systems, records, and people that decision depends on. Here the decision is which expiring leases and open depth rights to move on first, and what drilling it takes to hold the leases you have. The numbers are how many counties one landman can cover and how fast your team finds what’s expiring. Land work turns on clause language, recording conventions that changed over the decades, and answers your team will act on with acreage at stake. Decision Agents are built for complex, specialized problems like that one, on your own documents and systems, by people who know the domain. The landman agent is one of many. DataRobot’s Agent Workforce is our collection of agents to start from, organized by use case and industry. In oil and gas, it includes agents for drilling operations, artificial lift, subsurface data, and waterflood injection. Manufacturing, telecommunications, financial services, and federal government have agents in the collection too. A new Decision Agent starts from that collection rather than a blank page, and DataRobot’s team builds the rest around how your business actually runs. ## Two weeks on real county records We built this with a leading domestic upstream oil and gas operator. In a two-week proof of value, the agent ingested more than 700 scanned county instruments going back to the 1950s and answered lease questions with the source document cited. The agent’s logic was validated on that document set, and it covered the full working core: reading the scans, extracting the lease fields, answering in plain English with citations, keeping landman corrections, mapping leases to wells, and regression testing across extraction runs. That was a proof of value, and we’ll be plain about what it covered. A full deployment builds on that core for a land team running it every day: proactive alerts on upcoming expirations and open acreage, live document feeds, state regulatory production and permit data next to the lease record, maps of leases, units, and wells, and scale across multiple counties and states. Which of those comes first depends on your acreage. A team that lives by the expiration calendar starts in a different place than one hunting for open depth rights. ## Built for files you can’t let wander Lease files show where you’re leased, where you’re thin, and where you plan to drill. The agent reads your scanned instruments and well data from local document storage and works alongside the lease database you already have, so the documents stay where you keep them instead of moving to a new system. It runs on the DataRobot Agent Workforce Platform, which deploys as managed SaaS, in your VPC, self-managed in your own environment, or fully air-gapped. The platform traces every output back to its inputs, tools, and decisions. At the agent level, every answer cites its source instrument and a landman’s correction permanently overrides the extraction, so your land team can check what it says and fix what it misses. ## What it takes to get started DataRobot’s oil and gas team builds the agent on your documents, then runs and governs it for you. You don’t need to stand up an AI team. Setup for a first county takes about two weeks on a sample of your scanned instruments. Your side of it is the instruments and a landman to review the answers. ## Pick one county Tell us a county you care about, and we’ll show you what the landman agent finds in its public records: the instruments it reads, the terms and dates it pulls, and answers to your lease questions with the source cited. It’s the fastest way to see whether it reads your counties the way your best landman does.