Legal Services & Law Firms

AI Use Cases in Legal Services and Law Firms:
The Five Jobs This Sector Actually Described

Five recurring jobs, the people who run them, the constraint that binds each one — and the largest of the five, which our own on-device assistant does not cover.

Built from real buyer questions in our sales meetings

No law firm asks for AI. It asks for the deposition transcript before the week ends, for the clause buried in an agreement signed in 2008, for the demand letter a partner can put a name to tonight. Legal buyers arrive naming a document and a deadline, and laid side by side those requests resolve into a small, stubborn set: the same five pieces of work, described again and again by attorneys and paralegals who have never met each other.

That job set belongs to one sector. The cross-industry view lives in the cross-industry catalog.

Direct Answer

Five jobs, and an attorney or paralegal runs every one of them. Legal teams described extracting clauses, dates and obligations across the whole contract base; transcribing the deposition on the device and jumping back to the moment; querying the whole case file from a courtroom with no wi-fi; turning the case file into a demand letter or brief; and organizing the matter so the whole firm can work and report on it. A different constraint binds each one, which is why one answer cannot cover all five.

The limit is ours, and it lands on the biggest job of the five. A legal document analysis use case is outside AirgapAI scope, because the models it requires are bigger than a laptop can handle unless it is an Apple laptop. AirgapAI chat may not be able to do what a contract-review buyer needs. Laptop-local deployment is not powerful enough for legal work without a mini desktop supercomputer behind it, and AirgapAI does no bulk document processing. The contract-base job therefore belongs to server-class deployment or to another tool entirely. We would rather you read that here than discover it in a pilot.

A catalog is not a demonstration. Every job below is work somebody asked for, not a result anyone has published: our record carries no published outcome for most of these five, and a sector match on a list is not evidence that anything has been built for that sector. The deposition job carries specific recorded failures of its own — a transcribed word does not jump back to its moment in the audio, speaker detection is unsupported, and a long recording has failed to open. Treat the set as demand, then test it on one of your own matters.

These are jobs, not a tool shortlist. Which products to compare, whether material put into an AI tool stays protected, and which machine a document-heavy job needs are three questions with three homes. For more information visit the legal tooling comparison, the privilege and records page, and the sizing page.

The Five Jobs, as This Sector Described Them

The legal work in our record is not one firm telling one story. It runs across litigation and personal-injury practices, in-house legal departments, a legal publisher and a legal-workflow software company, and firms small enough that a solo practitioner works a very large document set alone — several of those relationships across repeat engagements rather than a single conversation.

The table below comes from what those buyers told us in our own sales and customer conversations rather than from an independent survey of the profession, so a job in it is a job somebody asked for. Each row carries the job as it was described to us, the role, the binding constraint, and what our record says about running it on a device. Where the record says nothing, the cell says nothing:

The job Who runs it What binds it What our record says about running it on the device
Extract clauses, dates and obligations across the whole contract base Attorney or paralegal Document volume exceeds what any human team can read Outside AirgapAI scope. The models it needs are bigger than a laptop can handle unless it is an Apple laptop, and AirgapAI does no bulk document processing.
Transcribe the deposition on the device and jump back to the moment Attorney or paralegal Data may not leave the device or the organization On-device capture is the job itself, and the record holds named failures against it: no jump back from a transcribed word to its moment, no speaker detection, a three-hour recording that would not play, a large audio file rejected at a size ceiling, and a recording format that cannot be imported.
Query the whole case file from a courtroom with no wi-fi Attorney or paralegal Air-gapped or fully disconnected environment Not recorded. No statement in our conversations either way about this job on the device.
Turn the case file into a demand letter or brief Attorney or paralegal No in-house AI skills and nobody to stand it up Not recorded. No statement in our conversations either way about this job on the device.
Organize the matter so the whole firm can work and report on it Attorney or paralegal Not recorded. Nobody described a binding constraint on this one. AirgapAI does not directly create a project management system, so the system of record stays where it already is.

Three cells are empty on purpose. Filling them would mean writing a sentence nobody in this sector ever said to us, and an invented answer about your courtroom is worth less than an admission that we do not have one.

What Makes Each of These Legal Rather Than General

Industries differ on one axis that survives scrutiny: the document in front of the person, and what it costs when nobody reaches it in time. Strip the document away and these five jobs read like anyone else's. Put it back and they are legal work with a deadline. Five documents, five consequences:

  • The contract archive. Get this wrong and an agreement auto-renews on terms nobody has read since it was signed. Buyers described paper from 2008 still renewing unchanged and sitting outside the regimes that arrived after it, entitlements such as rebates never claimed because the text was never machine-readable, and compliance flagging left to luck. The deep workflow page for this one is the contract-portfolio page.
  • The deposition recording. Get this wrong and the transcript arrives after the moment it was needed, which in a deposition means after the chance to use it has closed. None of our manual-work pages owns this job; the closest thing we publish is the on-device product it belongs to, AirgapAI Transcribe.
  • The loaded case corpus. Get this wrong and a question the file could have settled goes unanswered while the attorney stands in the hearing room with no signal. No manual-work page owns this one either; the condition it depends on is settled on the disconnected-operation page.
  • The medical history and the records. Get this wrong and a partner signs work product assembled from a file nobody read end to end. The nearest deep workflow page covers assembling and producing the material rather than drafting from it — the records and evidence page — and nothing we publish owns the drafting half.
  • The matter record. Get this wrong and a firm cannot report across its own cases: custom objects for injuries, treatments and pleadings, unified case views, hidden reporting fields, workflow triggers. No deep workflow page owns it, and that absence is itself the finding — the next section says why.

Where Our Own Product Stops on These Jobs

The most useful page for a legal buyer is the one naming the jobs we do not carry. Three of the five have a recorded boundary, and each has a mechanism behind it rather than a shrug. For more information visit the AI for law firms page.

The contract-base job is a scope statement, not a shortfall to argue about. The answer block carries our limit in full; the mechanism behind it is arithmetic. Extraction across a whole contract portfolio needs models and passes a portable machine cannot carry, and the on-device product has no bulk processing step to carry them with. The largest legal job in our record therefore belongs to server-class deployment or to a tool built for that shape of work. Where the machine sits is settled on the placement page. To put hours against that portfolio before scoping it, for more information visit the AI Contract Review Calculator page.

The deposition job runs on the device and still has sharp edges. Clicking a transcribed word does not jump to that point in the audio — which is precisely the jump-back the job name promises. Speaker detection is not supported, so the output is one undifferentiated transcript of everything said, and the mechanism is hardware: no laptop today can run the speaker detection model. A three-hour deposition recording would not open or play back and showed an error in the app. A large deposition audio file fails with a maximum-file-size error. AirgapAI does not support the AudioNote file extension, so recordings already held that way cannot be imported, and there are no promises that they ever will be. A firm recording depositions in that format today owns a conversion step and should price it before signing anything.

The matter record is a case management job wearing an assistant's clothes. AirgapAI does not directly create a project management system. Custom objects, landing pages and workflow triggers belong to the system of record a firm already runs; an assistant reads what that system holds and never becomes it. That distinction saves a procurement cycle. For the wider picture of law firm automation and where it sits beside a case management system, that page covers the ground.

Pin it down: questions for your evaluation
  • Which of these five jobs runs on the devices our people already carry, and which needs a desktop-class machine behind it?
    Where the volume line falls for your matter sizes rather than for a datasheet.
  • What is the longest deposition recording the build we receive has handled end to end, and what happens at the file-size ceiling?
    Whether the recorded playback and import failures are fixed in the build you get, in writing.
  • Can the recordings we already hold be imported in their current format, and if not, who owns the conversion step?
    Whether the format gap sits inside your workflow or inside ours, and what it costs to close.
  • If the matter stays in the case management system we already run, what does Iternal do and what stays with that system?
    The boundary between an assistant and a system of record, agreed before implementation.

Three Jobs This Sector Runs That Are Not Legal at All

Legal teams also run work every sector runs, and calling it legal would inflate the set with borrowed material. Three such jobs sit behind the five above, owned elsewhere in full, so they are named here and explained where they belong:

What Would Have to Be True Before Any of This Runs

Each job carries a precondition that decides whether it is buildable at your firm at all. The preconditions are structural rather than legal, and each is settled on the page that owns it:

  • Document volume beyond what any human team can read. The reason the contract job exists, and the reason it outgrows a laptop. Settled on the manual document work pillar.
  • Material that may not leave the device or the organization. A deposition recording is client material from the second it is captured. Settled on the confidential-data pillar.
  • An air-gapped or fully disconnected environment. A courtroom with no wi-fi is a sealed facility described by a different profession. Settled on the disconnected-operation page.
  • No in-house AI skills and nobody to stand it up. The constraint that quietly kills the drafting job: a firm that cannot staff the build never reaches the draft. Settled on the no-expertise page.

The Objection This Sector Raises Before Any Use Case

No legal buyer reaches a use case before raising this, and it deserves a straight handover rather than a reassurance. In their own words: material put into AI tools is discoverable and not protected; boutique firms are getting into hot water over their obligations; firms have lost arguments after using a public chatbot with attorney and client material; case documents sit under a protective order; and firms are simply unsure whether any of this breaks the protection they owe a client.

Privilege is settled on its own page rather than inside a catalog of use cases. For more information visit the privilege and records page.

What a Catalog Does Not Prove

A catalog is not a demonstration. Seeing your sector on a list is no evidence that anything has been built for it, and our record carries no published outcome for most of the five jobs above. What it carries is demand: attorneys and paralegals describing work that hurts, in enough detail to name the document and the deadline.

Our own failures on this exact point sit in the same record. Iternal has not always been able to offer genuinely different vertical demonstrations, only different wording around the same one. A use-case list grew long enough that its sub-filtering stopped working for the partners it was built for. And there are conversations where no demonstration of the prospect’s exact use case existed, only a prior one to show instead. Assume that gap sits behind any sector list a software company publishes.

Answered elsewhere
Continue Reading

More from The AI Strategy Blueprint

The Contract Portfolio Workflow

The deep workflow behind the contract-base job: executed agreements as a governed corpus, queried by duty and date.

AirgapAI Transcribe

The on-device capture and transcription product behind the deposition job, including what it does not do yet.

Records, Privilege and AI Regulation

Where privilege and discoverability questions are actually settled, rather than reassured away.
FAQ

FAQ: AI in Legal Services and Law Firms

Five recur across our own conversations with legal buyers: extracting clauses, dates and obligations across the whole contract base; transcribing the deposition on the device and jumping back to the moment; querying the whole case file from a courtroom with no wi-fi; turning the case file into a demand letter or brief; and organizing the matter so the whole firm can work and report on it. They are jobs somebody asked for, not results anyone has published.

The contract-base job, and Iternal says so directly: a legal document analysis use case is outside AirgapAI scope because the models it requires are bigger than a laptop can handle unless it is an Apple laptop. Laptop-local deployment is not powerful enough for legal work without a mini desktop supercomputer behind it, and AirgapAI does no bulk document processing. That job belongs to server-class deployment or to a tool built for that shape of work.

An attorney or a paralegal, on all five. The work stays with the person carrying the matter rather than moving to a data team, so whatever runs these jobs has to survive being used by someone whose day is already full. That is why having nobody in-house to stand it up binds harder here than the technology does.

Not today, and the job name promises exactly that, so we state it plainly: clicking a transcribed word does not jump to that point in the audio. Speaker detection is not supported either, so the output is one undifferentiated transcript of everything said, and the reason is hardware — no laptop today can run the speaker detection model. A three-hour recording would not open or play back, a large audio file failed at a maximum-file-size error, and the AudioNote file extension cannot be imported.

The documents change and the jobs do not. A personal-injury practice describes medical histories and records assembled into a demand letter, plus matter objects for injuries, treatments and pleadings. A transactional or in-house team describes the contract archive instead. The deposition and courtroom jobs belong to anyone who litigates. Match your document type to the row rather than your practice area to the sector.

Not most of them, and the distinction matters before a pilot. A catalog is not a demonstration: our record carries no published outcome for most of these five, and a sector match on a list is not evidence that anything has been built for that sector. Iternal has also not always been able to offer genuinely different vertical demonstrations. Test the job on one of your own matters.

Bring Us the Matter, Not the Category

Sector pages persuade nobody who practices law. One matter does. Pick a file you know cold, name the document and the deadline that hurts, and hold the list against it: if the job is on the list, we have told you what binds it and where we stop; if it is not, the list is incomplete, and we would rather hear that from you than publish around it.

Explore AirgapAI
John Byron Hanby IV
About the Author

John Byron Hanby IV

CEO & Founder, Iternal Technologies

John Byron Hanby IV is the founder and CEO of Iternal Technologies, a leading AI platform and consulting firm. He is the author of The AI Strategy Blueprint and The AI Partner Blueprint, the definitive playbooks for enterprise AI transformation and channel go-to-market. He advises Fortune 500 executives, federal agencies, and the world's largest systems integrators on AI strategy, governance, and deployment.

G Grokipedia LinkedIn X Leadership Team