AI Will Make Intelligence Cheap. Your PI Firm's Judgment Is the Moat.

When the owner is not in the room

Most PI firms already possess valuable institutional knowledge. The problem is its form. It lives in a partner's instincts, a senior paralegal's workarounds, old deposition transcripts, model demands, recorded training calls, and unwritten rules about when a case deserves escalation.

That creates founder dependence. The firm's quality changes with the availability of a few people. New employees learn through proximity. Departures remove context. The same facts can receive different treatment depending on who notices them.

The first AI question for an owner is therefore diagnostic: if you disappeared for a week, where would the firm stop thinking like you? Those points are candidates for institutionalization. They are not all candidates for automation.

Intelligence is becoming infrastructure

Models will continue to improve, and access to capable drafting, summarization, classification, and research will spread across vendors. Having AI will become less differentiating for the same reason having email or cloud storage is not a strategy.

The competitive gap will come from what surrounds the model: the quality of the firm's source material, the specificity of its instructions, the systems the agent can use, the tests applied to its work, and the judgment points reserved for people.

A competing firm does not need a magical model to pull ahead. It needs to respond to good leads sooner, preserve more case knowledge, remove more administrative delay, and give lawyers more time for strategy and trust. That is an operating advantage, not a software feature.

What your firm knows that software does not

"Wisdom" can sound mystical. Inside a PI firm, much of it is concrete. It is knowing:

• which inquiries need a lawyer's immediate personal response;
• which facts change case acceptance, urgency, or referral strategy;
• how the firm investigates coverage, collectability, and liability;
• which treatment or documentation gaps require explanation;
• what makes a demand, deposition, or negotiation persuasive in this venue; and
• when a client needs judgment and reassurance rather than another automated message.

Some of that knowledge can become criteria, examples, checklists, and escalation rules. Some remains tacit and must stay with experienced people. A mature AI strategy knows the difference.

Three shifts PI owners should prepare for

1. 01

   The best lawyers become more valuable, not less

   Agents can prepare first drafts, monitor queues, assemble records, and surface exceptions. The lawyer's scarce contribution becomes clearer: judgment, persuasion, responsibility, and the human moment that wins confidence. Firms that use AI well can return time to those activities and to life outside the firm.

2. 02

   Consolidation increases the value of a specific operating edge

   Better-funded firms can buy media, software, and talent. Legal-tech vendors can copy thin AI features. Neither can instantly reproduce a firm's local reputation, referral relationships, trial experience, and tested way of handling a narrow case type. A smaller firm's advantage becomes stronger when that knowledge is explicit enough to scale.

3. 03

   Conversation becomes the interface

   Lawyers will increasingly describe a mission in ordinary language: prepare the status update, assemble the chronology, identify what is missing, or draft the first version. The interface may be voice. The difficult part will still be the system underneath: approved sources, permissions, workflow state, review rules, and a record of what happened.

Build an institutional brain, not another chatbot

Uploading a folder of documents and adding a chat box does not create firm-wide intelligence. A useful system needs three layers.

Knowledge

Approved policies, playbooks, training material, exemplary work product, decision criteria, and carefully selected examples. The source should be identifiable, current, and appropriate for the task.

Workflow

The trigger, sequence, system integrations, assignments, deadlines, and escalation paths through which knowledge becomes action. This is what separates an answer from an operating system.

Governance

Permissions, matter boundaries, citations, evaluations, audit logs, uncertainty thresholds, and human approval. These controls determine what the agent may see, propose, and do.

That architecture is why we distinguish between isolated AI tools and AI systems that run defined workflows. The model is one component. The firm's operating design is the asset.

The lawyer becomes editor-in-chief

"Editor-in-chief" is a useful metaphor only if it does not become an excuse for superficial review. The lawyer defines the mission, standards, sources, and boundaries. The system handles approved execution. A person reviews the result in proportion to its consequence and uncertainty.

In intake, AI can acknowledge an inquiry, collect facts, organize the transcript, and alert the right person. It should not impersonate the relationship or make the final legal judgment. A competitive serious case may need the owner's text or call precisely because the human signal is what distinguishes the firm.

In case development, an agent can identify missing records, draft routine follow-up, and surface a discrepancy. The case manager decides what the discrepancy means. The attorney retains strategy, valuation, settlement authority, and responsibility for work sent to a client, opposing counsel, or a court.

This is consistent with the State Bar of California's practical guidance on generative and agentic AI, which emphasizes that greater autonomy requires stronger supervision and verification, while professional judgment remains with the lawyer.

Ownership means control and governance

A firm does not need to own every server to own the strategic asset. It needs meaningful control over its knowledge and operating layer: the ability to export data, preserve provenance, change vendors, define permissions, set retention, prevent unauthorized training, and inspect what the system did.

Broad data access is not a sign of sophistication. An agent connected to email, documents, calendars, and case files can create more leverage, but it also creates more ways to cross matter boundaries or disclose information. Access should be no broader or longer-lived than the workflow requires.

The ABA's Formal Opinion 512 similarly frames AI use through competence, confidentiality, communication, supervision, and review. Firms should also apply the rules and guidance of their own jurisdictions. Our practical starting point is an explicit AI governance and vendor-risk review before sensitive data or autonomous actions enter scope.

Start with one workflow

Do not begin by trying to reproduce the owner's entire brain. Choose one repeatable workflow where inconsistency or delay has a visible cost. Serious-lead response is often a good candidate because it combines urgency, economics, empathy, qualification, and attorney escalation.

1. 01Choose the failureIdentify where work currently stalls, varies, or depends on one person.
2. 02Observe the best operatorDocument what that person notices, decides, records, and escalates.
3. 03Separate execution from judgmentMark what can be prepared automatically and what still requires a person.
4. 04Curate examplesUse strong, weak, and ambiguous historical examples to define expected behavior.
5. 05Build the narrow systemConnect only the data and tools required for that workflow.
6. 06Test before autonomyEvaluate accuracy, uncertainty, escalation, and failure handling against real scenarios.
7. 07Measure the operating resultTrack response, completion, rework, exceptions, and human time rather than demo quality.

This is the same diagnostic logic behind our approach to AI-supported personal injury intake: make the firm faster around the human conversation, not less human at the moment trust is won.

Protect what should remain scarce

Your moat is not access to AI. Every firm will have access to AI.

The moat is the judgment your firm has earned, the relationships it has built, and the operating system through which both appear consistently. Technology can make a trusted lawyer's standards more available. It cannot manufacture the trust, responsibility, or experience that gave those standards value.

The immediate task is not to automate everything. It is to find where your firm's best judgment lives, decide which parts can guide a system, protect the underlying data, and preserve human control where the client or case needs it.

If you stepped away for a week, which parts of the firm would stop thinking like you?

Start there.