Mike on Legal Geek: open source as a third path, and good-enough models
November 18, 2026 · By Will Chen
Legal Geek has come back to Mike twice this year. The first piece placed open source alongside buying a platform and building in-house as one of three ways firms are approaching legal AI. The second picked up a different thread: large legal AI companies turning to open models to bring their costs down.
The coverage
A legal AI trinity?
Legal Geek sets out three routes for law firms: roll out one of the legal AI unicorns, invest in proprietary technology, or build with open-source platforms such as MikeOSS. It cites Bloomberg's Jason Wilson on what a two-week build says about the economics of legal tech, and Harvey co-founder Winston Weinberg's reply that open source does not mean free.
Identity crises?
A roundup of AI developments that includes Thomson Reuters building its Thomson model on the open-source Qwen, and Harvey's research preview on the open-weight Kimi. It quotes Will Chen's view that the aim is a model good enough for particular legal workflows, which makes it a cost-control measure.
Two threads in the coverage
The first is choice. Open source is now discussed as a real option next to commercial platforms and proprietary builds, not as an experiment on the side. Each route trades off control, effort, support and cost differently, and the debate in Legal Geek's piece, including the point that open source still carries infrastructure, governance and support costs, is the one firms are having internally.
The second is cost. When the largest vendors build on open models, the question shifts from which model is best to which model is good enough for a given task. Summarisation and similar workflows rarely need the most expensive model available, and choosing the right one for each job is how legal AI becomes affordable to run at scale.
For earlier coverage, see Mike in the press, June 2026.