Agents move into the system of record: contracts, databases and the controls that travel with them

Contracts stop leaving the system

The clearest pattern in this batch is that agents are being wired into the system of record instead of working on copies of it. Legora and Conga announced an integration that brings contracts managed in Conga CLM directly into the Legora platform, so legal teams can draft and analyze with AI on documents that stay where they are governed. The integration rolls out in three phases, starting with contract access for legal work in Legora, then allowing edits made in Legora to save back to the original document location in Conga, and then enabling natural-language search for contracts within Conga. Conga access controls automatically apply to contracts pulled into Legora, the companies say, and the integration is intended to reduce manual handoffs between systems. m:…5910

The scale the companies cite explains why the plumbing matters. Legora says it is used by more than 130,000 legal professionals at over 2,000 law firms and in-house teams. Conga says it serves more than 10,000 customers worldwide, and that over 50 percent of the Fortune 100 trust it for commercial operations. Gabriel Quek of Legora says the integration aims to prevent lawyers from having to export contracts to apply AI, and Brian Farr of Conga says it allows customers to apply AI directly to agreements managed in Conga CLM. m:…5910

Review becomes ranking, not reading

Harvey describes a second shift inside the same system: contract review becomes ranking provisions by risk rather than reading every clause with equal attention. Its contract review AI checks documents against standard positions in a contract playbook, and counsel should rank provisions based on data sensitivity, changeability, and recovery difficulty, according to the company blog. AI contract comparison tools can identify edits in execution copies compared to agreed versions, which matters because a signed document that differs from the negotiated version can quietly undo earlier risk work. m:…6151

The blog walks through a hypothetical example to show why ranking matters: a 36-month auto-renewal with 7 percent annual fee increases, alongside a separate cap of 540,000 dollars negotiated to cover breaches of data-protection and security terms. Change-in-terms provisions allow providers to update online terms after signing, the blog warns, and data used in model training is difficult to claw back once signed. A 12-month liability cap may be insufficient for platforms holding customer personal data, the blog adds, and usage data definitions can include prompts, outputs, and interaction logs. The real-world anchor the company cites is Repsol, which reduced contract clause review time by over 90 percent using Harvey. m:…6151

Databases follow the same pattern

Databricks follows the same pattern with data. The company says Lakebase Postgres is generally available on Google Cloud, separating compute from storage to remove resource conflicts and infrastructure overhead, and eliminating the need to build and maintain ETL pipelines by sharing the same storage layer as analytical data. It supports serverless autoscaling and scale to zero, zero-copy branching of production data, point-in-time recovery, and pgvector compatibility, with Unity Catalog providing unified governance. Multi-zone endpoints carry a 99.99 percent SLA and single-zone endpoints 99.9 percent, and loading 1TB of data takes less than five minutes, the company says. m:…2394

The agent angle is explicit. Databricks says KaBuM! uses Lakebase Postgres as the default memory layer for AI agents, and that Lakebase Search provides twice the throughput of pgvector at four times lower cost. The product is available in five Google Cloud regions: asia-southeast1, europe-west3, us-central1, us-east4, and us-west1. The pitch, in other words, is that operational workloads, analytics, and AI run on one platform, so agents read and write the same store analysts query rather than a copy synced through a pipeline. m:…2394

The control question

Taken together, the three items describe one architectural bet: AI works best when it works on the governed record in place, with the record controls traveling with the data. Contracts keep their access controls when pulled into a legal workspace. Reviews check against a playbook and rank by stakes. Databases serve agents from the same store as analytics, under unified governance. The open question is what happens when agents move from reading the record to changing it. The Legora and Conga save-back phase is described as a later phase, not the starting point, and Harvey comparison tooling exists precisely because execution copies can drift from agreed versions. The governance story is strong on access; the write-back story is still being built. m:…5910 m:…6151 m:…2394