Agent Builder v2

What's Coming

Agent Builder v2 will introduce a more flexible and intuitive way to create custom agents in Harvey by replacing rigid, block-by-block construction with simple text instructions and a conversational interface. Users will be able to describe what they want to build in natural language and collaborate with Harvey to iteratively refine their agent in real time, making it easier to create, test, and improve agents without managing underlying structures.

Users will be able to start from existing agents by importing templates from the Harvey Library, making it faster to customize proven workflows for specific use cases. A visual diagram will provide a clear view of how each agent runs, while still allowing for quick edits to individual steps.

In addition, Agent Builder v2 will streamline how agents collect and use inputs. Compared to Workflow agents, which require step-by-step user inputs, custom agents will be able to gather necessary information upfront or request it dynamically when needed—resulting in a smoother, more efficient experience when running agents.

Feature Highlights

• Create and edit agents using simple, flexible text instructions
• Build agents conversationally by describing requirements in natural language
• Upload Markdown files to generate agents instantly
• Start from existing templates in the Harvey Library and customize them
• Visualize agent logic with a flowchart and make targeted edits
• Improve run experience with smarter input collection and fewer interruptions
• Conversational w/ Magic Builder: Simply describe what you want to build to Harvey, and Harvey will walk you through building and testing it in a back-and-forth conversation.

Use Case Examples

• M&A Due Diligence (Corporate/Transactional): A senior associate at a mid-size firm oversees due diligence on nearly every deal, but the scope of due diligence shifts depending on the transaction type. With the revamped Agent Builder, she starts with a due diligence template from the Library, tweaks the text instructions to incorporate her firm's standard diligence checklist and flag categories, and saves it as a custom agent. When a junior associate runs the agent, the agent asks for the data room link and deal type, then works through the documents without stopping for input at every step. What used to require a walkthrough with the junior now runs on its own, and she shares the agent across her deal team in minutes.
• Regulatory Response Drafting (Litigation/Regulatory): A partner at a large firm regularly responds to SEC comment letters, and his team has developed a reliable internal process for how to structure those responses. He builds a custom agent using text instructions that encode his preferred approach. When a new comment letter comes in, an associate uploads it and the agent handles the initial structuring and drafting in one seamless run, asking for any needed context (like the prior filing and client tone preferences) at the start. The partner can update the agent himself when SEC expectations shift, without needing to understand underlying block logic.
• Employment Agreement Review (In-House): A general counsel at a mid-market tech company reviews dozens of offer letters and employment agreements each quarter, checking them against the company's current template and flagging deviations. She modifies an existing contract review agent with text instructions specific to her priorities. The agent asks for the agreement and the state of employment upfront, then produces a redline summary focused on her flagged issues. When the company expands into a new state, she edits the instructions to account for different enforceability standards and shares the updated agent with her deputy, no rebuild required.
• Public Company Benchmarking for Deal Pricing (Corporate/Transactional): A mid-level associate at a large firm regularly pulls comparable financial data to help partners advise on public company acquisitions. She builds a custom agent with text instructions encoding the firm's standard comps approach—which SEC filings to pull, what financial metrics to extract, and how to benchmark governance and compensation terms—and shares the agent with a junior associate. The junior enters the target's ticker, and the agent asks whether to use a provided list of comparables or identify a peer set based on industry, market cap, and revenue profile. From there, the agent pulls the relevant filings from EDGAR, extracts the data, and produces a structured comps table and benchmarking memo ready for the partner to review. When the firm starts advising in a new sector, the associate updates the peer criteria and benchmarks in the instructions—no rebuild required.

FAQs

Q: How is Agent Builder v2 different from the current Agent Builder?
Agent Builder v2 will replace rigid, block-based construction with a more flexible text-based approach. Users will be able to define and refine agent behavior through direct instruction editing, making the creation process faster and more intuitive.

Q: Do I need to start from scratch when creating an agent?
No. Users will be able to import agents from the Harvey Library and use them as a starting point, making it easy to build on proven templates and adapt them to specific workflows. The ability to convert block-based workflows into the new agent format will be coming in a future update.

Q: How will this improve the experience of running agents?
Agent Builder v2 will enable agents to collect required inputs more intelligently—either upfront or dynamically during execution. This will reduce the need for step-by-step prompting and create a more seamless experience compared to Workflow agents.

Q: Can I use Magic Builder with the legacy block-based Builder?

Yes — Magic Builder is available in both the new Agent Builder v2 and the legacy block-based Builder experience.