In a significant development for the legal technology sector,
DeepJudge, a Zurich-based enterprise renowned
for its legal AI solutions, has introduced the Agent Handoff Protocol (AHP).
This open protocol aims to enhance interoperability between AI products by allowing users to maintain the full context of their work as they transition between applications.
Highlighting its potential impact, legal AI company Harvey has announced its intention
to implement the protocol, currently rolling out its integration in beta.
Moreover, Thomson Reuters is aligning its CoCounsel for Legal solution with AHP
to support smoother transitions while further specifics on this development are awaited.
Siva Gurumurthy, Chief Technology Officer at Harvey, emphasized
that “The most capable legal workflows will draw on more than one specialized system.
Agent Handoff Protocol lets a user move from Harvey into a partner’s product and back without losing the thread of their work,
while leveraging both products to their full potential.”
Acknowledging the frequent transitions legal professionals make between
different AI interfaces to optimize their work, AHP seeks to eliminate redundant processes.
While existing models like the Model Context Protocol (MCP) enable system information exchanges,
they often fall short in porting comprehensive work context. AHP addresses that limitation,
ensuring users can continue their work seamlessly from one AI application to another, preserving objectives,
materials, and discussions.
In practical terms, should a lawyer research a legal question in DeepJudge,
they can seamlessly move into Harvey, bringing along essential work context, eliminating repetitive tasks.
DeepJudge’s CTO, Yannic Kilcher, affirmed the need for multiple AI experiences, stating,
“People should be able to use the AI tool that’s best for what they need to do and take their context with them.”
For more technical insights, the protocol’s detailed documentation is accessible at
agenthandoffprotocol.org.
As described by Kilcher, users effectuate this transition via a simple click,
while DeepJudge manages the secure transfer of context, upholding user data protection.
Unlike MCP, AHP does not seek to supplant existing systems but instead builds upon them,
creating a complementary layer that allows for a richer user interface experience when engaging directly
with applications. The GitHub page elucidates that while MCP focuses on connecting LLM applications and
external contexts, AHP facilitates the transfer of objectives and continuity data between independent applications.
Further information on this development can be found in the full
report.