Encode the playbook, not just the contract

Over a single year, the same limitation-of-liability clause crossed five different desks, and it got five different treatments. One associate pushed back hard and won a cap at twelve months of fees. Another accepted uncapped liability because it was buried on page forty and she was moving fast. A third flagged it but deferred to the counterparty's wording to keep the deal moving. None of them were wrong by the lights of what they knew. The firm simply did not have one position on that clause that everybody applied. It had a dozen positions, scattered across the memories of whoever happened to be reviewing.

That is the quiet inconsistency inside most contract review. The standard, what we accept, what we push on, what we never sign, lives in senior lawyers' heads and travels by apprenticeship and luck. Two good people review the same paragraph and reach different answers, because the thing they should be comparing it against was never written down anywhere a system, or a junior, could apply it.

A playbook that lives only in people's heads gets renegotiated, by accident, every time someone new opens the document.

Most "AI contract review" reads a contract and hands you a summary, which is the easy half and the wrong half. A summary tells you what the clause says. It does not tell you what your firm does about it, and that is the entire job. The method that matters is encoding the playbook: for each type of clause you write down the firm's preferred position, the fallback you will accept, and the red line you will not cross. Limitation of liability: prefer a cap at fees paid, accept a cap at twelve months, never accept uncapped. Those positions become structured rules a system can actually apply.

Now review changes shape. The system classifies each incoming clause by type, compares it against your encoded position, and tells you not just "here is a liability clause" but "this one sits below your fallback, here is the redline that moves it to your standard." It is measuring the contract against your playbook, not against a generic notion of normal.

Here is why this matters more than any single review. When the playbook is encoded, every contract gets measured against the same standard, by the most junior associate and the most senior partner alike. The clause that one person would have missed at page forty gets caught, because the system is not relying on attention or experience, it is applying a rule. The firm's hardest-won judgment, the positions the senior partners fought to learn, stops walking out the door each time someone leaves and starts compounding instead.

The value is not that the AI read the contract. It is that the AI redlines toward your position instead of a stranger's.

Contract volume keeps rising and review headcount does not, so more agreements get a fast skim than a real defense, and the expensive clauses are exactly the ones that hide in long documents nobody has time to fight line by line. An encoded playbook is leverage against that math: it lets a small team hold a consistent line across far more contracts than they could ever read by hand, with the standard applied the same way every time.

We built this for a team where contract positions lived entirely in tribal knowledge, and the redlines stopped depending on who picked up the file. The playbook reviewed the contract, and the associates spent their time on the genuinely novel terms instead of re-deciding the settled ones.

Encode the playbook, not just the contract, and review stops being a test of who is reading. It becomes the firm's actual standard, applied in full, on every agreement, by everyone.

TensorLabs builds the contract-intelligence infrastructure behind that kind of playbook-driven review.