Resources Get in touch

Build your agent

Check out the resources available under "Building agents" on the Cogniti site for how to build your agent. This pages goes through what you learned in the introductory workshop.

2 Test your agent and ask for feedback

Before releasing your agent widely, trial it with a small group of colleagues or students. Encourage specific feedback on the clarity of prompts, the relevance of answers, and the tone of interaction. Ask them to use the agent as intended, and note where responses may be accurate, useful, unclear, incomplete, or off-topic. Use these insights to refine your system message and improve the overall learning experience.

You can also involve students in co-creating and refining your agent. Invite them to trial the agent, suggest improvements, and propose new prompts or features that make it more relevant to their learning needs. This collaborative process not only improves the agent’s effectiveness, but also fosters student autonomy, competence, and critical engagement with the learning design process.

3 Review and refine

Creating an effective Cogniti agent is an iterative process. Even well-designed agents can miss the mark if prompts, resources, or instructions aren’t perfectly aligned with the learning goal(s). Regularly review your agent’s responses to ensure accuracy, clarity, and relevance. Use feedback from users to identify gaps, adjust the system message, update resources, and fine-tune instructions. Continuous refinement helps your agent remain effective, engaging, and fit for purpose. See resources available under "Controlling agents well" on the Cogniti site for more information on reviewing and refining your agent.

Some common problems with agents include:

Unfocused system messages, causing inconsistent or irrelevant responses.
Overly complex or overloaded instructions, hindering task prioritisation.
Lack of testing, leaving issues unnoticed until agent deployment.
No clear target user, leading to misaligned responses.
Excess or poorly curated resources, confusing the agent and reducing accuracy.
Missing key terms in resources, causing misunderstanding.
No refinement over time, leading to outdated content and function.
Unrealistic expectations, including expecting the agent to replace, rather than support, educator expertise.