The Deep Dive

You don’t need this page to run a cycle. It is for anyone, at Peninsula or in another district, who wants the reasoning and research behind the design.

Where this came from

AI tools are improving faster than schools are built to change. One widely cited measure, from the research group METR, found that the length of software tasks AI agents can complete on their own has doubled about every seven months since 2019 (METR, 2025). District plans usually run for several years. A district that cannot change at something closer to the pace of the tools ends up letting vendors decide what AI does in its schools.

The Open Adaptive District is how Peninsula is responding. It was built as part of Superintendent Krestin Bahr’s Google & GSV Ed Leader Fellowship (2026–27) and published so other districts can adopt or adapt it. It is a way of working, and it does not replace Peninsula 2030. It is how we keep that plan current as conditions change.

The action plan submitted for the fellowship, with its commitments, measures and timeline, is kept as submitted, apart from small corrections. Read the action plan. Everything else, including these pages, is revised as we learn.

The design choices, and why

Why six weeks? Six weeks is long enough to learn something real, and short enough that a wrong turn costs weeks rather than a year. With two weeks between cycles, a school year holds about four cycles. A one-year pilot would spend most of its time studying tools that had already changed.

Why one build per team? A team running one build usually finishes it. A team running three tends to finish none and report vaguely on all of them. If an idea is too big for six weeks, split it.

Why publish stopped builds? Organizations usually let failures go quiet, and the next team repeats them. A build stopped openly, with a clear reason, saves other teams from trying it again. We expect the stopped builds on What We’re Learning to be the most useful part for other districts, because so few districts publish theirs.

Why can anyone build? For decades, building software was a specialist’s job, so a district could fix only as many problems as its technology department had time for. AI tools that work from plain language changed that: anyone who can describe a problem clearly can now build a working fix. Spreading the building out spreads the district’s capacity to adapt, while the guardrails (student data privacy, and review of anything students will see) stay central. The goal is that every administrator, and eventually every educator, sees “I could build that” as a normal response to a problem that keeps coming back.

Why doesn’t this threaten the instructional core? Because the goals don’t move; only the methods do. The guaranteed and viable curriculum, the Instructional Essentials, MTSS and UDL are what builds aim at. Leadership fixes the goals and leaves teams free to choose the methods, which is Bungay’s idea of directed opportunism (below) applied to instruction. A good build makes an existing commitment easier to keep, such as finding a struggling student earlier, or producing scaffolds that keep work at grade level. A build that undermines a commitment fails its own build plan, which asks every team to name the thing that must not get worse.

Why do kept builds become skills and tools? A finding written up in a document helps the team that wrote it. A skill in everyone’s PSD AI Agent, or a tool anyone can run in AI Studio, helps the whole district from the day it ships. Each cycle adds to a shared set of tools, so a new administrator a few years in inherits working tools instead of a binder of reports.

Why an AI agent in the middle? Ten or more teams produce more weekly notes than any one person can read, connect and summarize. One agent sits in every team’s space and does that reading, so district leaders can spend their time acting on what it finds.

Why three sentences from the superintendent? Three sentences is short enough that a team can read the focus once and think of two or three builds. The focus says what we are after and why; teams decide how. Anything longer turns into a task list, and teams stop choosing their own work.

The research behind it

Five bodies of work shaped the design.

• Peter Senge, systems thinking (The Fifth Discipline, 1990). Organizations often fail to change because their systems are built for stability, whatever the people in them want. So we changed the system itself: the cadence, the documents and the sharing.
• Stephen Bungay, directed opportunism (The Art of Action, 2011). Leaders state the intent, and the people closest to the work choose the method. When conditions keep changing, adapting as you go works better than carrying out a fixed plan. This is why the quarterly focus is three sentences and not a work plan.
• Stanley McChrystal, team of teams (Team of Teams, 2015). Small teams move fast but cannot see each other. Everyone reading everyone’s wrap-ups, plus one agent that reads every team’s space, gives them that view, while decisions stay with the teams.
• Dave Snowden, the Cynefin framework (with Mary Boone, “A Leader’s Framework for Decision Making,” Harvard Business Review, 2007). Most AI questions are not complicated problems an expert can solve; they are complex problems where nobody knows the answer yet. In complex situations you probe, sense and respond, which is what a six-week build does.
• David Teece, dynamic capabilities (“Explicating Dynamic Capabilities,” Strategic Management Journal, 2007). Organizations that do well in turbulent conditions keep sensing change, seizing opportunities and transforming themselves. The cycle puts that on a six-week clock.

Guardrails

• Builds start with how adults work, so early mistakes don’t land on students.
• Anything students will see is reviewed before it reaches them.
• The agents never touch personnel evaluations, discipline, grievances, or anything marked out of bounds.
• Student data privacy and the relationships in our schools come first, every cycle.
• Everything the agents read stays on district systems.