Agents need more than a chat - Jacob Lauritzen, CTO Legora

2026-04-22 · Source: AI Engineer · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Software Development & Engineering · Depth: Intermediate, long

Summary

Legora, a vertical AI company for law firms with over 1,000 customers, addresses the challenges of complex, long-running AI agents, particularly the "context rot" that leads to agents forgetting prior instructions. The company highlights a shift in the economics of production, where planning and reviewing work, rather than execution, are now the primary bottlenecks. They introduce "the verifier's rule," stating that AI will solve tasks that are both solvable and easy to verify, noting that legal tasks like contract writing are solvable but difficult to verify. Legora proposes strategies for effective human-agent collaboration, focusing on increasing trust through task decomposition, guardrails, and proxy verification, and increasing control through planning, skills, and elicitation. They advocate for high-bandwidth, persistent interfaces beyond traditional chat for complex agent interactions, such as collaborative documents or tabular reviews, to facilitate better human oversight and judgment.

Key takeaway

For AI Architects designing vertical AI solutions, recognize that traditional chat interfaces are insufficient for complex agent collaboration. Prioritize developing high-bandwidth, persistent interfaces like collaborative documents or structured tabular reviews. This approach will enhance human control and trust, mitigate context rot, and enable more effective human-in-the-loop verification, ultimately improving the reliability and adoption of your AI agents.

Key insights

Complex AI agents require high-bandwidth interfaces and structured human collaboration to overcome context rot and verification challenges.

Principles

• If a task is solvable and easy to verify, AI will solve it.
• Planning and reviewing are the new bottlenecks in AI production.
• Agents are not humans; do not constrain them to human language interfaces.

Method

Increase agent trust by decomposing tasks, using proxy verification, and adding guardrails. Increase human control via planning, encoding skills for contingencies, and eliciting user input for unknown scenarios.

In practice

• Decompose complex tasks into verifiable sub-tasks.
• Implement guardrails to limit agent actions and increase trust.
• Use persistent, high-bandwidth interfaces (e.g., documents, tables) for agent collaboration.

Topics

• Complex AI Agents
• Vertical AI
• Agent-Human Collaboration
• Verifier's Rule
• High Bandwidth Interfaces

Best for: AI Engineer, AI Architect, AI Product Manager

Related on AIssential

• When AI Talks to AI: A Practical Guide to Agent-to-Agent Communication — Distributing complex AI tasks across multiple communicating agents overcomes single-agent limitations.
• It’s the Lessons We Learned Along the Way. Or, Is It? — AI agents can efficiently complete complex data tasks, but may diminish human understanding and critical thinking development.
• Why Most Financial Chatbots Are Broken by Design — And What It Actually Takes to Fix Them — Effective conversational AI prioritizes user mental models and robust system design over raw NLP capabilities.
• No AI Agent Orchestration Needed? What? — Modern frontier LLMs can self-orchestrate complex procedural tasks more effectively via in-context prompting than with external agent frameworks.
• Sixteen Claude Agents Built a C Compiler Without Human Intervention... Almost — Autonomous AI agents can collaboratively build complex software like a C compiler with minimal human oversight.
• One on One with Marc Benioff — AI agents and conversational interfaces like Slack are transforming work, making human-agent collaboration central to enterprise operations.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by AI Engineer.