Every managing partner I talk to in 2026 has seen the same pitch deck four or five times. A polished legal-AI platform promises research assistance, document review, contract analysis, deposition summarization, and a dashboard that tracks your firm’s “AI maturity.” The logo at the top of the slide changes. The slides do not.
The dirty secret of the legal-AI platform industry is that every firm in your city, your practice area, and your referral network is looking at the same product. If a tool is available to all of you, it cannot, by definition, be a competitive advantage for any of you. It is a new floor, not a new ceiling.
That is the first problem. The second problem is worse.
The platforms are generic. Your practice is not.
Legal-AI platforms are designed to be usable by every practice area at every firm, which means they are optimized for none of them. A mid-market M&A practice, a plaintiff-side class-action boutique, and a regional insurance-defense shop all run very different workflows. They have different document types, different intake patterns, different billing logic, different review standards, different judges and opposing counsel they see repeatedly, different clients who want different things.
A platform built for “legal” cannot meaningfully understand any of this. It defaults to the lowest common denominator: a general-purpose summarizer, a general-purpose drafter, a general-purpose search box pointed at a corpus you don’t control. What you get is a tool that feels useful for the first two weeks and then becomes another login nobody uses.
The “security” story is a misdirection
Every legal-AI vendor leads with security: SOC 2, encryption at rest, zero-retention agreements, privilege protections. That is table stakes, not a differentiator. Your ethical obligations to your clients do not end when a vendor signs a DPA. If the AI platform is handling privileged matter data, you are responsible for what it does with that data, and you are one vendor breach away from a Rule 1.6 problem you didn’t author.
A custom module, built inside your infrastructure on an LLM provider you contract with directly, removes an entire class of intermediary risk. The matter data never leaves your perimeter. The model calls are logged in systems your firm controls. The audit trail is yours, not a screenshot of a vendor’s dashboard.
What bespoke actually looks like in a 20-attorney firm
Consider a boutique real estate practice doing 30 purchase-and-sale transactions a month. Every transaction requires pulling together title reports, survey exceptions, loan documents, corporate authority, and closing statements into a reviewable package for the closing attorney. A paralegal spends, on average, four hours per transaction assembling and cross-checking this package.
A generic legal-AI platform cannot help with this meaningfully. It does not know the firm’s closing binder template. It does not know which title exceptions the firm’s two closing attorneys care about versus which ones they waive. It does not know the firm’s lender and title-agent relationships.
A bespoke module can. Scope it narrowly: one workflow, one set of document types, one firm’s review standards. Train it on two years of the firm’s completed closings. Point it at the firm’s document management system. In four to six weeks of focused work, you have an agent that produces a draft closing package in twenty minutes instead of four hours, flags exactly the exceptions the closing attorneys historically care about, and surfaces the three questions that require a human judgment call.
That is a competitive advantage. It is also, critically, a thing you own. The next time a new real estate practice opens across town, they cannot license this tool and erase your edge.
The economics have inverted
The argument for buying a legal-AI platform rests on a 2015 view of software economics. Custom software was expensive, slow, and risky. A platform was the rational choice even knowing the fit would be approximate.
That is no longer true. A custom AI module for a single well-scoped workflow now costs less, at initial build, than two years of per-seat licensing for most legal-AI platforms at a mid-sized firm. After those two years, the platform continues charging and the custom module continues running. The break-even math is extremely favorable, and that is before you count the hours your associates spend learning a new platform every time your firm switches vendors.
What to build first
Start with the workflow that meets three criteria:
- It is high-volume for your practice. Fifty matters a quarter, not five.
- It has clear, firm-specific review standards. A senior attorney can explain what “good” looks like in under five minutes.
- It currently eats paralegal or junior-associate hours that the firm is not able to bill at full rate.
Common candidates: intake triage for a litigation practice, discovery review in a narrow document type, deal-point extraction in a transactional practice, deposition digest generation, conflict-check pre-screening, engagement-letter drafting against a partner’s historical templates.
Each of these is a 4-to-6-week build. Each one, owned by the firm, compounds. None of them is available to your competitors when they call the next vendor.
The platforms will not disappear. They will get commoditized.
None of this means the platforms go away. Harvey, Legora, Clio Duo, Lexis+ AI, Thomson Reuters CoCounsel: they will continue to exist and continue to improve. Some of them will be genuinely useful for horizontal tasks like research and citation-checking where there is no firm-specific advantage to be had.
But the real leverage for your firm, in 2026 and forward, is in the workflows that are yours specifically. Those workflows are not going to be solved by a product designed for every lawyer in America. They are going to be solved by a small amount of custom code, owned by your firm, that understands exactly how you practice.
Stop buying platforms for the work that defines your practice. Build instead.
