Every AI Agent Has Three Parts.
Lawyers Need to Understand All Three.

A legal opinion depends on who writes it, what rules they operate under, and what facts they were given. An AI agent works exactly the same way — governed by its Model, its Harness, and its Context.

→ Try the Interactive Lab

The Three Components

Understanding each one is as fundamental as knowing the difference between a fact and a legal conclusion.

🧠
Component 1 — Model

The Intelligence

The model is the AI's core brain — its training, reasoning depth, and domain knowledge. Different models have vastly different capabilities. The same question put to a basic model versus a frontier model yields qualitatively different results: one might miss a circuit split that could sink your case; the other anticipates it.

Legal analogy: The model is the attorney's legal education and raw intellect. A first-year associate, a seasoned litigator, and a former SCOTUS clerk are all lawyers — but hand them the same case file and you'll get very different memos.
🔧
Component 2 — Harness

The Rules of Engagement

The harness is the scaffolding built around the model. It defines what tools the AI can access, what actions it is permitted to take, what it is expressly prohibited from doing, and how much human oversight is required before anything goes out the door. This is where control lives.

Legal analogy: The harness is the firm's operating procedures — which databases the attorney can access, whose sign-off is required before a filing goes out, and whether they can send client emails autonomously or only with partner review.
📁
Component 3 — Context

The Case File

Context is the specific information fed to the agent for this task: the facts, documents, instructions, and prior communications. The same model with the same harness will produce entirely different output depending on what's in the file. Context is what makes a general-purpose AI into a matter-specific advisor.

Legal analogy: Context is the case file. The same senior partner advises differently on a slip-and-fall versus a securities fraud matter — not because they changed, but because the facts changed.
Configure the agent below and observe how each component shapes the output

The Agent Configuration Lab

Select one option from each column — then run the agent to see how the combination shapes the legal output.

🧠 Model — The Intelligence
🎓 First-Year Associate Basic
Foundation Model · GPT-3.5 tier
Knows black-letter law; may miss strategic nuances, circuit splits, or procedural traps. Eager but needs supervision.
🏛️ Senior Litigator Advanced
Advanced Model · GPT-4 tier
Deep reasoning, strategic judgment, recent precedent awareness. Anticipates counterarguments and spots procedural landmines.
🔬 Appellate Specialist Expert
Frontier Model · Claude Opus tier
Constitutional depth, cross-jurisdictional insight, long-horizon strategy. Former SCOTUS clerk energy — and the billing rate to match.
🔧 Harness — Rules of Engagement
📚 Research Library Read-Only
Search & Summarize · No External Action
Can search Westlaw/Lexis, read cases, and produce internal memos. Cannot draft filings, send communications, or take any external action.
🏢 Full Firm Toolkit Full Autonomy
Complete Access · No Review Required
Research, draft, email, file documents, schedule, bill — all autonomously. The agent operates end-to-end with no required human checkpoints.
👁️ Supervised Workflow Human-in-Loop
Full Toolkit · Partner Review Gate
Same tools as Full Firm, but every output above a threshold requires partner sign-off before execution. Work product flows; autonomy doesn't.
📁 Context — The Case File
📄 SaaS Contract Dispute Commercial
Meridian Tech v. CloudBase · D. Del.
3-year SaaS agreement. Counterparty unilaterally changed API terms after 14 months via a modification clause, causing $2.3M revenue loss.
🏥 Slip & Fall — Premises Liability Personal Injury
Harrington v. Apex Retail · Cuyahoga Cty.
71-year-old client, fractured hip. Store video: spill unattended 47 minutes. Medicare beneficiary. Ohio modified comparative fault.
👔 Wrongful Termination — Title VII Employment
Chen v. Pinnacle Financial · S.D.N.Y.
Asian-American VP fired 21 days after internal harassment complaint. Employer claims restructuring. EEOC right-to-sue letter received.
Try different combinations — each variable changes the output in a distinct way.

Agent Output

⚖️

Select your configuration above and click Run the Agent to generate a simulated legal memo.

Discover more from Ethan B. Holland

Subscribe now to keep reading and get access to the full archive.

Continue reading