The Pedagogy Behind the Studio

6 days ago

There is a growing need for enduring skills like judgment, creativity, and critical thinking to manage AI, but few pedagogical approaches exist to develop them.
Generative AI should be treated as a creative medium rather than a tool, using art and design education methodologies.
The Generative AI Studio pilot applied studio structure, critique, and charrette methods to develop discernment and creative process.
The studio structure involves individually-driven projects with no single right answer, focusing on criteria like intrinsic AI, aesthetics, UI, architecture, data security, and ethics.
The critique (crit) develops discernment through feedback framed as perspective, not truth, with psychological safety and group norms.
The charrette is a time-bound process for rapid prototyping, moving through divergence, sense-making, convergence, and assessment.
Students learned to direct AI by evaluating output, closing the gap between promises and delivered experiences, and moving beyond chat interfaces.
Final projects were more differentiated, used multimedia forms, and included security and ethical considerations.
The studio approach equips students with enduring skills to navigate generative AI, with plans to expand to more students.

Invest in upfront alignment by making implicit assumptions explicit in the initial prompt to avoid unexpected AI outputs.
Restart with better context rather than steering AI when initial outputs are far off; avoid compounding errors through path dependence.
Equip AI agents with the same tools you use (CLI, MCP, browser control) to handle setup, testing, and instrumentation tasks.

The author created a beeswarm chart showing AI model releases over time, with each mark representing a model and its horizontal position indicating the publication date.
They discuss the initial excitement of AI-generated code feeling like magic, but overuse leads to producing low-quality work and overwhelming PR reviews, forcing a choice between being a bottleneck or relying on agents with half-hearted efforts.
Reflecting on the past, the author notes that coding involved problem exploration, decision-making, and team collaboration, with code evolving from fuzzy ideas to clear implementations as a medium for thought.

The JARVIS Challenge at MIT tested if AI can accelerate the design-build-test cycle for a small gas turbine engine, with undergraduate teams building engines in four weeks.
AI tools, accessed via the Parley platform, helped with tasks like summarizing textbooks, software tutorials, vendor sourcing, and design trade studies, but faced limitations like hallucinations and lack of physical understanding.
Teams found that engineering judgment and expertise were crucial; while AI assisted, human oversight was needed to trust, challenge, and translate AI outputs into working hardware.

Traditional tools like Vim and Emacs gain trust through deep customization and muscle memory, making them feel like extensions of the developer.
AI coding agents speed up output but reduce trust; usage rises while trust falls (from 40% to 29% in a recent survey).
New AI tools require a shift in process and culture, not just tool adoption; a broken process remains broken even with better tools.

The author initially doubted AI's coding abilities, treating it like Stack Overflow and manually reviewing code to maintain personal style.
Claude Code demonstrated AI's ability to handle maintenance tasks and reduce the need to learn specific frameworks like Laravel.
Newer AI models (e.g., Fable, GPT 5.6-Sol) now outperform the author in 80% of coding tasks, leading to acceptance and excitement about productivity gains.

Invest in upfront alignment by making implicit assumptions explicit in the initial prompt to avoid unexpected AI outputs.
Restart with better context rather than steering AI when initial outputs are far off; avoid compounding errors through path dependence.
Equip AI agents with the same tools you use (CLI, MCP, browser control) to handle setup, testing, and instrumentation tasks.

The author created a beeswarm chart showing AI model releases over time, with each mark representing a model and its horizontal position indicating the publication date.
They discuss the initial excitement of AI-generated code feeling like magic, but overuse leads to producing low-quality work and overwhelming PR reviews, forcing a choice between being a bottleneck or relying on agents with half-hearted efforts.
Reflecting on the past, the author notes that coding involved problem exploration, decision-making, and team collaboration, with code evolving from fuzzy ideas to clear implementations as a medium for thought.

The JARVIS Challenge at MIT tested if AI can accelerate the design-build-test cycle for a small gas turbine engine, with undergraduate teams building engines in four weeks.
AI tools, accessed via the Parley platform, helped with tasks like summarizing textbooks, software tutorials, vendor sourcing, and design trade studies, but faced limitations like hallucinations and lack of physical understanding.
Teams found that engineering judgment and expertise were crucial; while AI assisted, human oversight was needed to trust, challenge, and translate AI outputs into working hardware.

Traditional tools like Vim and Emacs gain trust through deep customization and muscle memory, making them feel like extensions of the developer.
AI coding agents speed up output but reduce trust; usage rises while trust falls (from 40% to 29% in a recent survey).
New AI tools require a shift in process and culture, not just tool adoption; a broken process remains broken even with better tools.

The author initially doubted AI's coding abilities, treating it like Stack Overflow and manually reviewing code to maintain personal style.
Claude Code demonstrated AI's ability to handle maintenance tasks and reduce the need to learn specific frameworks like Laravel.
Newer AI models (e.g., Fable, GPT 5.6-Sol) now outperform the author in 80% of coding tasks, leading to acceptance and excitement about productivity gains.

Hasty Briefsbeta