Every mid-sized company handles legal through one of three models: an outside law firm, in-house counsel, or AI agents supervised by lawyers. Each model wins at something and loses at something, and anyone who tells you one wins at everything is selling you that one. I have spent years on the selling side of legal services, so I know every pitch by heart; this is the comparison I wish someone had put in front of me from the other side of the table.
Law firm vs in-house counsel vs AI: how each model stacks up
The three models in one sentence each. The law firm bills by the hour or by retainer and serves many clients at once. In-house counsel sits on your payroll and lives in your business full time. Agents produce the routine work while a lawyer supervises, approves, and takes the complex matters; in Rafik’s model, agents resolve most matters end to end.
The table, criterion by criterion:
| Criterion | Outside law firm | In-house counsel | Agents + lawyer |
|---|---|---|---|
| Cost per routine matter | Billable hours or retainer; the simple contract pays partner rates | Fixed payroll; cheap per matter only at constant high volume | a fraction of that per routine matter produced by an agent and approved by a lawyer |
| Response time | Typical SLA of 24 to 72 hours | Hours or days, depending on their load | a fraction below the traditional SLA |
| Coverage hours | The firm’s office hours | One person’s working hours | Agents receive and produce at any hour; the lawyer approves during business hours |
| Business knowledge | Every matter starts with you explaining context | The highest: lives inside your operation | Agents draft citing sources from your own corpus: contracts, policies, minutes |
| Scalability | More matters = more billed hours | More matters = another hire | More matters = the same agents; scales without new payroll |
| Litigation and M&A | Their home turf; they win here | Coordinates, but subcontracts the firm | Out of scope; escalated to lawyers and firms |
Where each model wins (and this is not a diplomatic tie)
The law firm wins at litigation and complex deals. A serious lawsuit, a commercial dispute, a merger or an acquisition demands specialists with real hours in courtrooms and negotiation rooms. There the billable hours earn their keep, and anyone pitching you AI agents for an M&A has never sat through a due diligence.
In-house wins when there is daily regulatory volume. Financial services, pharma, energy: if every week brings filings before regulators and decisions carrying fine risk, you need a full-time person holding that context in their head. The payroll pays for itself with the first mistake it prevents.
Agents win at the routine flow, which is most of it. Service agreements, NDAs, letters, day-to-day employment questions: that flow is where that share end-to-end closure rate applies, at a fraction of that per matter. The part that tends to surprise people is the quality control: in Rafik’s internal eval, agents scored the vast majority correct across 100 real matters, and each agent checks its confidence against a 0.95 threshold; below it, the matter escalates to a lawyer instead of the agent guessing. A rushed junior associate does not ship with that mechanism.
What almost nobody tells you: the three models coexist. The companies that run legal best use agents for routine work, law firms for litigation and specialties, and some add an in-house hire when regulatory volume justifies it. The right question is what percentage of your flow belongs to each model, and only then who to pay for each slice.
Deciding with your own numbers
Take your last three months of legal matters and sort them into three columns: routine, recurring regulatory, and complex or contentious. If routine dominates (in mid-sized companies it usually passes 60%), you are paying specialist rates for supervised-machine work. Look at how an AI-native legal department operates and price your routine column against subscription pricing.
The usual closing note, because it matters: this comparison is general information for choosing a structure, and your specific mix of matters, contracts, and jurisdictions deserves a review with a lawyer who knows your case.
Frequently asked questions
Can AI agents handle a lawsuit?
No, and distrust anyone who promises it. Litigation requires procedural strategy, hearings, and representation that belong to lawyers and specialized firms. Agents cover routine work and prepare inputs; contentious matters escalate to humans.
What happens to the 35% of matters agents don’t close?
A lawyer takes them. The agent checks its confidence against a 0.95 threshold and escalates instead of guessing; every human approval is captured as a learning that enters the agent’s context only after sign-off. The public goal is a higher share closure by mid-2028.
Does in-house counsel become obsolete with agents?
No. The job changes: less repetitive drafting, more supervision, strategy, and regulatory work. In heavily regulated sectors, in-house counsel remains the piece no other model replaces.
How do agents learn my company’s context?
Through retrieval with citations over your own corpus: your contracts, policies, and corporate minutes. When the agent drafts, it cites where each clause comes from, so the reviewing lawyer verifies instead of trusting blindly.