Home › How-to guides › Automate research and monitoring

How to automate research and monitoring with an AI agent

I looked through 201 real cases about research and monitoring and pulled out what people actually did, the prompts they used and the places it went sideways. No theory, no invented success stories.

What people actually do

• 50 AI agents in Minecraft built a government, a ration ledger and a no-blame inquiry about a chicken · GPT-6.1 Sol agents · Research & Data
  YC Labs, working with Olam Labs, put 50 GPT-6.1 Sol agents in a survival Minecraft world for three hours with the goal 'Build a thriving…
• Pensive offers personalized AI teaching assistants for higher ed with Claude · Claude · Research & Data
  Pensive — Read the case study to see how they build on the Claude Platform. Pensive aims to build an AI Teaching Assistant (AI TA) that supports a…
• From weeks to a day: Everstar on Microsoft Foundry with Claude · Claude · Research & Data
  Everstar Foundry Models — Everstar built Gordian on Claude in Foundry Models, cutting nuclear permitting timelines 80–96% and completing 4–6 weeks…
• Biomni accelerates biomedical discoveries by 100x with Claude · Claude · Research & Data
  Biomni — Read the case study to see how they build on the Claude Platform. They realized the field needed more than incremental improvements—it…
• Speed up multifamily deal research and property summaries · Custom agent · Research & Data
  Real-world thread: the broker says AI is most useful before outreach and during early diligence: they use it to pull together market research…
• Terra Firma uses Harvey and a PwC UK tax model to cut tax research from days to seconds · Harvey · Research & Data
  Private equity firm Terra Firma uses Harvey in deal work, including a UK tax model co-built with PwC that integrates PwC’s tax domain expertise with…
• Reed Smith has 700 Harvey users a day and firmwide associate AI certification · Harvey · Research & Data
  Reed Smith, a global law firm with about 1,800 lawyers across 34 offices, has made Harvey part of an AI-enabled delivery model. The page reports 700…
• ICL Group used Harvey to complete a cross-border restructuring in-house and estimates 30–40% more lawyer efficiency · Harvey · Research & Data
  ICL Group uses Harvey to scale legal work across five regions. On a cross-border restructuring involving entity transfers, dividend distributions…

Browse all 2754 cases to see the rest.

Prompts to start from

These are the reusable prompts from real cases. Swap in your own details, run it once on something low-stakes, and read what the agent did before you trust it.

I'm designing a workflow with several AI agents. Review it for the failure modes seen in agent colonies. Check for a single hub that everyone messages, so it becomes a bottleneck; for agents spending effort on process, such as approvals, roles and permission requests, rather than the task; for handoffs where one agent passes something and another picks it up by mistake; and for agents becoming overcautious after failures. For each risk, propose a concrete fix, such as a named handoff location, a written protocol, a message budget or a deadline. Keep the plan as simple as it can be.

I want an agent to take on this workflow, show its working, and wait for approval on anything that changes a system or contacts someone. The tool in the source case was Claude.

Tools that show up most

Harvey (52), Custom agent (26), Microsoft 365 Copilot (21), Copilot Studio (20), Claude Code (16), Claude (9).

Guardrails

Start small: run it once on a low-stakes example, read what the agent did, and only then let it run on real work.

Where it went wrong, or nearly did

• Try Devin on research tasks (mostly failed): Devin got stuck parsing HTML on a scraping task.

More mess: the AI agent fails page. Related: the prompt library.