Summer institutes have a long and uneven history in K-12 professional development. At their best, they provide the sustained, immersive learning environment that traditional school-year PD cannot match. At their worst, they are expensive retreats that generate enthusiasm in July and amnesia by September. When I set out to design an AI-focused summer institute for educators this year, I was determined to land on the right side of that divide.

Now that the institute has concluded and the post-program surveys are in, I want to share a transparent accounting of what happened - the design principles that delivered, the decisions that misfired, and the modifications I would make if I were building this from scratch tomorrow. My hope is that other educators and PD designers can use this as a practical blueprint rather than repeating my mistakes.

The Design Framework

The institute ran for four consecutive days, six hours per day, with a cohort of 36 educators ranging from second-year teachers to 25-year veterans. The group included elementary generalists, secondary content specialists, special education teachers, and three instructional coaches. Grade levels spanned K through 12. District contexts ranged from affluent suburban to Title I urban.

I anchored the program around three design principles derived from adult learning research and my own experience as a Google for Education Certified Trainer.

Principle 1: Production over consumption. Participants would spend no more than 30% of any session listening to instruction. The remaining 70% would be spent building artifacts they could deploy in their classrooms in August. This was non-negotiable. The single greatest predictor of PD transfer, in my experience, is whether participants leave with something they made, not something they watched.

Principle 2: Pedagogical framing before tool training. Every tool introduction was preceded by a discussion of the instructional problem it addressed. We did not start with "Here is how to use Diffit." We started with "Differentiation requires producing multiple versions of materials calibrated to individual readiness levels, which is logistically impossible at scale without assistance. Here is how AI changes that calculus." This sequencing meant that tool selection always followed problem identification, which is the order in which effective practitioners actually make decisions.

Principle 3: Structured reflection as a daily practice. Each day ended with 30 minutes of written reflection using a protocol adapted from the National School Reform Faculty's critical friends methodology. Participants documented what they built, what they learned, what confused them, and what they planned to try next. These reflections served dual purposes: they consolidated learning for the participant and provided me with formative data to adjust subsequent sessions.

What Worked

The "Build Your AI Toolkit" portfolio

Each participant maintained a running portfolio throughout the four days, organized by instructional function: planning, assessment, differentiation, communication, and feedback. By the end of the institute, every participant had a minimum of eight tested, customized AI workflows ready for classroom deployment. The portfolio structure transformed what could have been a disconnected series of tool demos into a coherent, personal professional resource.

Post-program survey data indicated that this was the single most valued element of the institute, with 94% of participants rating it "extremely useful" for anticipated classroom application. Follow-up check-ins in the weeks after confirmed that participants were actually using their portfolios, not just filing them.

Cross-grade-level collaboration

I deliberately structured table groups to maximize diversity - an elementary art teacher alongside a high school chemistry teacher alongside a middle school special educator. My hypothesis was that cross-pollination would generate more creative applications than homogeneous grouping. That hypothesis was confirmed emphatically. The most innovative AI use cases I observed emerged from conversations between educators who would never have collaborated in a traditional departmental PD structure.

A seventh-grade ELA teacher adapted a formative assessment workflow designed by a third-grade teacher at her table. A high school physics teacher borrowed a parent communication template developed by a kindergarten teacher and modified it for his AP students' families. These transfers would not have occurred in grade-band silos.

The ethics simulation

On Day 2, I ran a 90-minute simulation in which table groups were presented with realistic AI-related dilemmas drawn from actual incidents reported in K-12 settings: a student submitting AI-generated work, a teacher using AI to write recommendation letters, a district purchasing an AI tool with inadequate data privacy protections, a parent demanding that their child not be exposed to AI in the classroom. Groups had to develop and defend a response using an ethical reasoning framework we had introduced earlier.

This session generated the most intense engagement of the entire institute. Multiple participants cited it in their reflections as the moment when AI stopped feeling like a technology topic and started feeling like a professional responsibility. The simulation format - as opposed to a lecture on AI ethics - allowed participants to practice the judgment calls they will actually face, which is a fundamentally different cognitive task than passively absorbing principles.

What Flopped

The pacing of Day 1

I front-loaded too much conceptual content on the first day, violating my own 30/70 principle. The morning session included a 45-minute overview of how large language models work, which I had intended as essential background knowledge. It was too long, too abstract, and too early. By the time we reached the first hands-on activity, I had already lost a portion of the room.

The corrective is straightforward: reduce the conceptual overview to 15 minutes, deliver it through a guided exploration activity rather than a presentation, and trust that deeper understanding will emerge through use. Teachers do not need to understand transformer architecture to use AI effectively, just as they do not need to understand TCP/IP to use the internet.

The advanced track

I offered an optional "advanced" afternoon strand on Day 3 for participants who wanted to explore custom GPTs, API integrations, and prompt engineering at a deeper level. Twelve participants opted in. The session was too ambitious for the time available and too disconnected from the classroom application focus that had made the rest of the institute effective. Several participants in the advanced track reported feeling overwhelmed, while several who stayed in the standard track reported feeling like they were missing something important.

The lesson: in a four-day institute, depth is better served by extending the core curriculum than by splitting the cohort. If I offer an advanced track in the future, it will be a separate program entirely, not a bolt-on to the foundational experience.

Insufficient attention to district policy contexts

I underestimated the degree to which participants' capacity to implement what they learned would be constrained by their district's AI policies - or lack thereof. At least eight participants returned to districts that had no formal AI use policy, which meant that everything they built in the institute existed in a governance vacuum. I should have included a module on policy advocacy - how to approach administration, how to draft a proposal for AI piloting, how to navigate institutional uncertainty.

What I Would Do Differently

Beyond the specific fixes noted above, three structural changes would strengthen the model.

Extend to five days. Four days created compression that forced trade-offs I did not want to make. A fifth day would allow for a culminating presentation in which participants share their portfolios with peers and receive structured feedback - a capstone that reinforces learning and builds professional community.

Require an administrator co-participant. For every cohort of six to eight teachers from a single district, require at least one building administrator to attend. Administrative buy-in is not optional for sustainable AI integration, and the most effective way to build that buy-in is shared experience, not post-institute reporting.

Build in a 30-day follow-up. The institute's impact will ultimately be measured not by what participants built in July but by what they sustain in October. A structured follow-up session - virtual, 90 minutes, focused on implementation troubleshooting - would extend the institute's half-life and provide data on transfer that I currently lack.

The Broader Imperative

This institute was one program in one summer for 36 educators. The United States employs approximately 3.7 million public school teachers. The gap between what is needed and what exists is staggering. Scaling effective AI professional development requires more than good program design. It requires funding, institutional will, and a fundamental reorientation of how we invest in the teaching profession.

I do not have a solution to that systemic challenge. What I have is a model that worked for 36 people, a candid account of where it fell short, and a commitment to iterating. If you are designing something similar, I hope this postmortem saves you some of the mistakes I made and gives you a foundation to build on.

The teachers who attended this institute are going back to their classrooms in August with tools, confidence, and a professional network that did not exist four days ago. That is not enough. But it is a start.

Dr. Janette Camacho is a Google for Education Certified Trainer and Coach, Google Certified Educator Level 1 and 2, Adobe Creative Educator, Apple Teacher, and FETC 2025 Featured Presenter with 28+ years of K-12 classroom experience. She is the founder of iTeachAI.