Joanna Leung

Product Leader | Passionate in solving problems valuable to people with technology |

AI Agent Generator

π—–π—’π—‘π—§π—˜π—«π—§
Pragmatico delivers a suite of 10 AI agents to beginner learners at mid-market clients. Each agent runs as a Claude or ChatGPT Project with guided onboarding, read-only access and a human in the loop. The first agent required the most iteration. Its patterns became the standard for every agent that followed.

π—§π—›π—˜ 𝗣π—₯π—’π—•π—Ÿπ—˜π— 
Each new agent started from scratch. The design patterns lived with one builder and in the first agent’s instructions. Scaling the suite meant rebuilding the same structure by hand, and teammates with their own use cases had no repeatable method to follow.

π—›π—¬π—£π—’π—§π—›π—˜π—¦π—œπ—¦
If the first agent’s patterns become a reusable generator, then non-technical Pragmatico Trainer or Teammates can be empowered to build new, high quality, working agents for any use cases so they can adopt AI in their line of work without relying on a tech resource.

π—–π—’π—‘π—–π—˜π—£π—§
Codifying the first agent’s patterns into a reusable generator would give each new agent a tested starting draft. The primary metric was time to create a new agent. The guardrails were no loss in agent quality and no skipped testing.

π—¦π—’π—Ÿπ—¨π—§π—œπ—’π—‘
Agent Creator is a Claude Project that converts a use case and context into a complete agent draft in the beginner-learner format. Each draft includes the seed prompt, project instructions and an optional setup guide. Patch Mode and a handoff block ship with every agent, so quality does not depend on who builds it.

π—˜π—«π—˜π—–π—¨π—§π—œπ—’π—‘

  1. Pattern codification: converted the first agent’s proven patterns into a fixed, step-by-step build process.

  2. Connector-first discovery: the agent reads active connectors before asking anything, then asks at most two to three questions to understand the use case it is building against.

  3. Confirmation gates: the Claude version is confirmed before the ChatGPT version starts.

  4. Platform constraints: instructions are compressed to fit 10,000 characters on Claude and 8,000 on ChatGPT.

  5. Human quality control: every draft goes through outcome testing and system prompt refinement by the PM.

π—œπ— π—£π—”π—–π—§
New agent creation time fell by 70%, from 6-8 hours per agent to 3 hours per agent. The remaining 30% covers outcome testing and system prompt refinement. That step stays human by design, because it determines agent quality.

The team can now template agents in one shared structure.

𝗣𝗒𝗦𝗧-π—Ÿπ—”π—¨π—‘π—–π—›
Agent Creator was shared with the Pragmatico team. Teammates now build agents for their own use cases in the same structure. This spreads AI adoption across the organization instead of concentrating it with one builder.

The next iteration adds one generator per AI competency level. The same engine raises agent complexity for advanced learners and for other use cases.

Role
AI Adoption Consultant
For
Mid-Market Companies
Date
April – May '26
Type
AI Agents, Claude Team, ChatGPT Business, AI Enablement, AI Adoption
URL
www.pragmatico.ai/
AI Agent Generator
←
Next:
Personal Project: Dental Expander
→