Today in Seongnam, at 'Bioprotech', we ran a training on automating hands-on work with AI agents. We worked through 3 cases with real practitioners, and the response was enthusiastic. This is probably how most AI training will run from now on.
Case 1. Contract-generation agentA web app that completes a contract automatically once you enter the client, product, and contact person. Train it on a well-made existing contract, build the web app with Claude Code, and it’s ready for real work.
Case 2. Automating sales data from production resultsA system that, once production results come in, automatically forecasts sales data and generates the report. We designed it all the way through to automatically sending tailored reports to the sales rep, the division head, and the CEO.
Case 3. Sales dashboard + forecast managementThe most complex case, but the one the team wanted most.
A decision-making tool that shows target → actual → weekly forecast at a glance, so you can reallocate team resources when signs of a sales slowdown appear. In Claude Cowork, we pointed to the Excel folder, mapped the sheet names, and built it as a web app. Results came out right away.
Key insight: the IPO (Input-Process-Output) frameworkTo build agentic AI properly, you ultimately have to be able to design the IPO (Input → Processing → Output) clearly.
And of course, designing this takes domain knowledge.
In today’s session, the team lead who knew the whole business process designed agents much faster.
In the AI era, "knowing what to build" matters more than being good with the tools. Have a great holiday ^^b
27 Feb 2026
[ Instructor: John Kim ]