Attorney Zack Shapiro just published one of the most detailed, practical accounts of AI in legal practice that we've seen — and it should be required reading for every firm leader, managing partner, and operations director paying attention to where the industry is headed. His thread, titled "The Claude-Native Law Firm," doesn't deal in theory or hype. It's a blow-by-blow look at how a two-person boutique law firm is using AI to compete against firms with hundreds of lawyers — and winning.
At Dyntyx, we work with law firms every day to deploy AI agent systems across their operations. What Shapiro describes from the practitioner's seat maps almost perfectly to the transformation we help firms implement at a systems level. Here's a breakdown of the thread and why the patterns he's surfacing matter far beyond his own practice.
The night before a client's acquisition was set to close, opposing counsel dropped a last-minute demand letter restructuring key deal terms — new escrow conditions, expanded indemnification carve-outs, revised closing deliverables. It was 7 PM and the implicit message was: accept or we walk.
Shapiro uploaded the purchase agreement, disclosure schedules, and the demand letter into Claude. Within minutes, the AI mapped every proposed change against the existing deal terms and found contradictions the buyer's own lawyers had apparently missed — two proposed carve-outs conflicted with representations they'd already confirmed in the disclosure schedules, and a third would have actually weakened the buyer's own post-closing protections. As the negotiation played out over email through the evening, Shapiro fed each new communication into Claude, which tracked how every concession interacted with provisions across the full agreement and helped him build a response that conceded the points worth conceding and held firm where it mattered. By 11 PM he had a clean set of counter-positions grounded in specific cross-references to the buyer's own language. The deal closed the next morning.
The work, Shapiro estimates, would have taken three associates at a mid-size firm until morning. He had the core of it in under two hours.
That kind of speed isn't just impressive — it's the new competitive baseline. We see this with our own clients: contract review times dropping by 75%, with typical 80-page SaaS agreement negotiations going from 20–30 attorney hours down to 3–5 hours. The firms that adopt these workflows aren't just saving time. They're winning business because they can move faster and deliver more thorough work product than the competition.
One of Shapiro's sharpest arguments is against the wave of specialized legal AI products — Harvey, Spellbook, CoCounsel, Luminance, and others. His point: these are wrappers on top of the same foundation models, marketed with the promise of firm-level customization. But the customization they offer — template libraries, clause banks, firm playbooks — solves a problem that barely matters. Every competent firm in a given practice area has roughly the same templates. The real differentiator was never the template; it was the lawyer's judgment about what to do with it.
Shapiro's alternative is to encode his own analytical frameworks, preferences, and decision-making patterns into persistent instruction files — "skills" — that fire automatically based on context. When he uploads a contract, the AI doesn't apply a generic review framework. It applies his framework, the one he's developed over a decade of practice. The result is first-pass work product that's immediately useful, not something that requires extensive revision.
This is a point we emphasize at Dyntyx constantly. The AI solutions that actually deliver ROI aren't generic chatbots bolted onto legacy processes. They're purpose-built systems designed around how a specific firm actually works — its workflows, its standards, its escalation rules. That's why our approach to law firms centers on orchestrated multi-agent systems rather than one-size-fits-all tools. When we deploy a contract lifecycle management solution, for example, we're configuring five coordinated AI agents — automated drafting, intelligent review and risk analysis, surgical redlining, compliance tracking, and portfolio analytics — all tuned to the firm's specific standards, style, and criteria. The agents get more accurate over time because they're learning from your patterns, not some industry average.
Shapiro walks through three modes of working with Claude — chat, autonomous (Cowork), and code — and the thread really takes off when he describes what autonomous mode looks like in practice.
For tracked changes, he uploads a counterparty's 40-page redline and the AI organizes every change by severity, flags where risk was shifted, identifies tensions between modified provisions, checks for missing standard provisions, and produces a summary with specific counter-language. Then — and this is the part he says "drops jaws" — the AI opens the Word document at the XML level, applies tracked changes attributed to his name, preserves every formatting detail, and produces a clean .docx that opposing counsel can open and review normally. He never opens Word. He never opens Litera.
For research, his custom skill instructs the AI to run parallel research across every relevant regulatory angle simultaneously, cross-reference sources, prioritize primary authority, and then run a self-review before delivering anything. That self-review step is critical: the AI must verify that every cited authority actually says what the memo claims, flag anything below high confidence, check for internal contradictions, and guard against hallucinated citations — the problem that got several lawyers sanctioned and made national headlines.
For real-time contract interpretation, a client received a demand letter claiming breach and threatening termination with a 48-hour response window. The AI mapped every allegation against the specific contract provisions cited and discovered that two of the four claimed breaches referenced obligations that had been modified by a side letter the counterparty's own counsel had drafted. As Shapiro prepared his response, the AI pressure-tested each draft paragraph for unintended implications on other provisions — catching a defense that could have been read to concede a point on a separate payment dispute.
These aren't hypothetical use cases. They're Tuesday. And they map directly to the kinds of outcomes we deliver for our law firm clients. Our AI-powered legal research and brief writing solution cuts drafting time by 65–70% while enforcing mandatory citation verification and court compliance — zero fabricated citations, consistent brief quality across attorneys and matters. Our intake and client onboarding solution deploys four coordinated AI agents that handle the entire lifecycle — 24/7 engagement, real-time qualification, instant conflict checks, and automated documentation — reducing response times from hours to under five minutes and capturing 100% of after-hours leads that would otherwise be lost.
Shapiro is candid about the downstream implications. His two-person firm handles the workload of a much larger practice. The work that traditionally justified an associate hire — first-pass document review, research memos, initial drafts, redline summaries, routine correspondence — is now handled by AI under his supervision. Associates aren't obsolete, he says, but the bar for when hiring one makes economic sense has shifted, and what you need them to do has changed: judgment, client relationships, and AI output supervision.
On billing, he's moved to subscription pricing alongside traditional hourly billing. Clients get ongoing advisory, contract review, compliance monitoring, and governance for a flat monthly fee. AI makes that model sustainable because he can deliver comprehensive service within a predictable cost structure. Clients love it because they're never afraid to pick up the phone.
And on judgment — this is where Shapiro is most emphatic — he warns against the temptation to let the AI do too much. Every output still requires attorney review. The AI makes you faster, more thorough, and more consistent, but the judgment — the part where you decide what to fight for, read between the lines, and stake your reputation on a call — that remains yours. Experienced lawyers, he argues, have an enormous advantage in this new world because the expertise they've built over 10 or 20 years is exactly the asset AI makes more valuable, not less.
We couldn't agree more. At Dyntyx, every system we build is designed with transparent oversight — every AI decision is auditable, escalation triggers ensure attorneys maintain control, and there is zero black-box AI. The technology handles the routine execution. Your team handles the high-judgment decisions. That's the operating model that works.
The legal industry is at a transformation point. 64% of in-house legal teams are now building AI capabilities internally, reducing their dependency on outside counsel. 93% of legal leaders are actively implementing generative AI for contract management. Firms that wait aren't standing still — they're falling behind.
Shapiro's thread is proof of concept for the solo practitioner. What Dyntyx builds is the scaled version of that same philosophy: orchestrated, governed AI agent systems that deliver measurable ROI in 6–8 weeks, not quarters. Whether it's intake, contracts, research, eDiscovery, or marketing, the playbook is the same — encode your firm's expertise into intelligent systems, automate the repetitive layers, and keep human judgment exactly where it belongs.
If you're a firm leader wondering what this transformation looks like for your practice, we'd love to show you. Schedule an AI Strategy Call
