A note for Michelle Kariyeva, Esq. · From Denis

Where reasoning AI fits inside an agile PI firm.

Four short stories of what a Claude-class agent does inside the operations of an agile personal-injury firm in 2026. One tool for every laptop. One note on a custom build that used to take six months and now takes three weeks. Three minutes to read.

From Denis About Kari Law Group, Richmond Hill Length Three minutes
01 · Where AI sits

On your team's existing files. Not a new platform.

A reasoning model from Anthropic (Claude) reads incoming material, drafts in your firm's voice, and hands cleaner starting points to your associates and paralegals. No new SaaS reseller between your firm and the model. Direct Anthropic account, your files, your bills.

Reported outcome in similar agile firms: 20 to 30% time back per person on real client work. Across a team, that compounds.

AI helps in two places. Before signing: a Saturday-night call in Russian gets a Russian answer in 90 seconds. After signing: faster drafting, faster records review, faster multilingual comms. Lead-gen brings the flow in (whoever runs that). The AI layer below keeps it from leaking between intake and demand.

02 · Story one · Saturday, 11 PM

A Bukharian family just had a fender-bender on the Belt.

Today

The 11:14 PM call hits whoever's covering after-hours. Notes get retyped Monday. Other firms reply faster, in his language.

With the AI agent

Live pickup in the firm's voice: "Kari Law. Press 1 English, 2 Russian, 3 Spanish, 4 Georgian." Structured intake captured (incident, injuries, treatment, other-party, insurance). Bilingual SMS confirmation in 90 seconds with a Monday slot. By 9 AM your intake team has a sorted, structured queue ready to route, no handwritten forms to retype.

What changes

The first hour after an accident is covered, in the caller's language, without burning out the associate on weekend rotation.

03 · Story two · Demand letters in the firm's voice

An associate's afternoon becomes a senior review of half an hour.

Today

An associate spends most of an afternoon on a soft-tissue auto demand. Generic legal-AI templates sound like LegalTemplate.com, so they rewrite eighty percent. You then read it, mark it up, hand it back. The case sits another day before it ships.

With the AI agent

An agent trained on the firm's prior demands and your edits drafts a NY-PI-specific letter in your voice. Associate edits 30 minutes. You review 15. Demand ships the week the file matures.

What changes

Settlement-stage throughput goes up. Junior associates spend their time on the parts that build judgment, not on plug-the-facts-in-the-template.

04 · Story three · Multilingual comms across the team

The 7 PM Russian text gets a fluent Russian reply by 7:05, regardless of who's on point.

Today

The bilingual paralegal goes home at 5. The Russian-speaking client texts at 7. The English-only associate replies in slow English or waits until morning. Multilingual service depends on which staff member is at their desk that hour.

With the AI agent

Inbound message, English summary, draft reply in the firm's voice in the client's language, all on one screen. Edit one phrase, send. Same for Russian, Spanish, Georgian.

What changes

Every person on the team can hold a fluent conversation in any of your three languages. Word-of-mouth in the Bukharian and Russian-speaking communities compounds.

05 · Story four · 80-page records bundle

Pre-deposition prep stops being read three times by three people.

Today

Paralegal does an afternoon-long first pass. Associate reads all 80 pages. You review the prep notes. The same bundle gets navigated three times.

With the AI agent

Paralegal drops bundle into Claude: "Summarize injuries, timeline, gaps, what hurts the case." Five bullets in 30 seconds. Paralegal verifies against records in 20 minutes. Associate reads the actual records only where bullets flag something.

What changes

Meaningfully fewer person-hours per case. Throughput on the active caseload goes up without hiring. Adjuster questions answered in real time. No third-party reseller between the firm and the model.

06 · One more thing · A tool for every laptop

Claude on every desktop, the same AI sitting next to every person.

Separate from the four agents above. Anthropic's Claude Team plan (around $30/seat/month annual) gives every person a desktop assistant they interrupt whenever, with shared workspace and admin controls. Practical uses on a typical day:

Standalone. Doesn't depend on the four agents above. Roll out next week, let actual usage shape the bigger builds.

07 · One more category · Custom software, weeks not months

Built around how your firm runs, not someone else's workflow.

What used to be a six-month, six-figure project is a three-to-four-week build. Claude does the heavy coding next to me. Firm owns the code and the data. An agile firm is the right size: large enough to pay back fast, small enough to skip the six-month rollout.

Three example seeds, only to make the category concrete:

The actual build is whichever piece of your work the closest off-the-shelf tool gets to 80% of and stops.

08 · If any of this resonates

The smallest first move is half an hour, not a project.

If we start anywhere, we start here

Claude Team on every laptop (Section 06), plus the records summarization piece (Story four). Both run on machines and files your team already has, don't change anything customer-facing without your approval, and don't depend on the bigger build. Safest possible starting line.

If something resonates, something's already solved, or something's not your pain, I'd love to hear it. One-word answers are entirely fine: resonates / already doing / not my pain / didn't get it.

Half an hour whenever it's easy.

No deck, no slides. Founder to founder.