If you’re reading this, you can probably thank a bridge.

The parts that make up the technology that allows you to read these words were shipped using one (or more) of the United States’ 600,000 bridges. Bridges don’t usually get credit for what they make possible; they just quietly help us cross gaps that block our way forward.

A gap that many district leaders are becoming familiar with is the one between an AI use case’s promise and its reality. For example, your professional development (PD) on an AI platform lands well in August. By November, usage has dropped. By March, you’re weighing whether to cut the tool from next year’s budget.

Researchers call it the implementation dip–“a dip in performance and confidence” that happens between rolling out something new and actually getting good at it. Even with a clear, high-value use case, staff buy-in and an AI committee, the dip can still show up. That’s why this article focuses on what gets you across that gap.

Which brings us back to bridges. And to build a bridge, you have to first survey the gap you’re trying to cross.

Survey the Gap 

There are several types of bridges, and the kind you need depends on what’s actually underneath it: the terrain, the distance and the conditions it has to withstand. 

The same goes for AI implementation: gaps open not necessarily from the design of the initial plan, but because structures erode under everyday pressure, just like steel. 

• Uneven capacity-building: Strong PD teaches people how to use a tool, but it rarely has time to cover the dozens of small judgment calls that appear during classroom use.  Training that covers those judgment calls in the fall likely won’t hold by spring, since the tools themselves shift capabilities and staff might even shift platforms they use on their own.
• Fading theory of change: Even a clear theory of change, one that goes beyond “this will make things more efficient” to name what should actually change or improve for students or teachers, doesn’t automatically survive a school year—the original “why” fades into the background amidst many competing priorities.
• Thinning expectations: Standards for effective AI use erode as staff make independent judgment calls across hundreds of classrooms and as Large Language Model (LLM) capabilities shift rapidly. As tool usage evolves, expectations can struggle to keep pace.
• Fraying messages and vision: A unified message at launch doesn’t stay unified without work. Staff turnover, competing initiatives and a principal with different priorities, all pull at the edge of what was once a coherent vision. The line between what’s supposed to stay and what’s supposed to change frays without reinforcement.

Key Question: Which force or forces are most likely to erode even the work you launched well?

These four erosion patterns—and the pillars that counter them—are drawn from our work directly alongside district teams, including through Education First’s AIxCoherence approach to sustainable AI implementation.

Build the Crossing

Crossing the gap means building structures that withstand everyday school-year pressures. Below, we break down the four load-bearing pillars of sustainable AI implementation—People, Processes, Outcomes and Communication—and show how each one addresses the specific vulnerabilities outlined above.

People 

Capacity doesn’t become uneven because a district picked the wrong people to implement an AI use case. Capacity gets built and re-built through ongoing, intentional professional development and regular practice, especially given how AI evolves over the course of a school year. Identify who can drive the change, then build a real structure with regular touchpoints and refreshers to maintain their capacity as conditions shift. 

There are people in any school or district whose main job relates to AI and/or technology; their inclusion is implied. There are also people outside those categories who may have specific AI expertise (e.g., hands-on AI experimentation in math classrooms) or are heavily implicated in AI rollouts (e.g., Multi-tiered Support System leads if focused on AI for differentiation). Make sure to include those personnel in ongoing efforts, not just as part of a group informing the launch. 

Processes

A theory of change rarely dies all at once, but it does stop getting mentioned over time. Build a clear model for how AI is meant to be used, appropriately, effectively and equitably, and put it back in front of users on purpose (e.g., newsletters, posters). Then ask: How will you course-correct when something isn’t working, and how will you keep naming what success is actually for, months after the rollout excitement wears off?

• Anchor the Message with Logic and Purpose: Shift from restrictive rules (e.g., “don’t plagiarize”) to principle-based guidance (e.g., “excellence over efficiency”). Explain the logic behind the use of AI, such as clarifying when AI should serve as a thought partner versus a final editor during independent practice.
• Check for Blind Spots: Before treating any AI output as final, ask what part of the story might be incomplete or inaccurate. A differentiated lesson plan that pulls from a student’s assessment history needs the same—if not more—scrutiny as a rubric or grading suggestion. Because AI can flatten or stereotype outputs based on historical patterns, always ask: Does this output truly serve the student in front of me?

Outcomes

Articulate an observable vision of success, what good AI use sounds like, looks like and even feels like and revisit that definition as things change instead of treating it like a one-time exercise. 

By articulating or re-articulating the vision, you enable your staff to ask themselves: Does how I’m about to use AI align with the vision, rather than with exactly how I was trained, on a use case that may no longer apply? 

• Measure the Moments: Using quantitative measures (e.g., platform logins) alongside qualitative measures (e.g., moments when teachers said ‘aha’ when AI use helped them learn or do something new) can provide a fuller picture of the learning experience and environment.

Communication

With your cross-functional team, co-create a crisp, compelling message about the shared vision for effective AI use, and keep repeating it through the same channels beyond the initial rollout, long enough to withstand the people and circumstances that will change around it.

• Examples Can Do the Heavy Lifting: When the message seems unclear, use examples (e.g., videos, written exemplar scenarios, symbols) to depict the heart of the message. The examples can help staff make critical connections between the message’s intent and its interpretation.

Key Question: Which pillar do you want to focus on right now to strengthen your bridge? 

Test the Load

The only way to know where a bridge’s failure points are is to deliberately apply real weight before the weight arrives on its own. Testing AI implementation works the same way. 

• Make small bets. Rather than testing “Is AI working in our schools?”, isolate one use case, one team, one unit. Work with a 9th-grade English team to test an AI platform on rubric generation for a unit, then regroup to see if the process worked, helped and held up for all students. 
• Assess honestly. At pre-determined intervals, check the distance between hope and reality. If teachers are successfully using AI to generate customized reading passages, look closer: Are those passages maintaining grade-level rigor? If usage numbers are high but text complexity is dropping, your implementation failure point has moved to a less visible location. And be honest about what “assess” can mean: sometimes the gap isn’t a structural weak point to reinforce—it’s a sign that the use case itself isn’t the right one. Sunk costs ( PD hours, rollout momentum, political capital spent announcing it) aren’t a reason to keep crossing a bridge that leads nowhere. 
• Retool and repeat. Talk to those closest to the use case before drawing conclusions. If elementary teachers report that AI-drafted weekly newsletters sound cold, don’t automatically abandon the tool. Take that feedback back to your Communication pillar and strengthen what’s eroded. 

Key Question: What’s the smallest bet you could make right now to test your bridge, and how will you know if it holds for everyone who needs to cross it?

Once you’ve crossed this particular gap, you’ll find another one waiting—a new use case, a new tool, a new cohort of staff. 

The bridge you just built may not transfer, but your new skills as a bridge engineer will.

*** 

Josh Parker is a Senior Consultant at Education First and has spent the last two years coaching district leadership teams on AI exploration and implementation. He works with clients across P-20 schools and the nonprofit sector in the areas of coherent assessment systems, instructional quality and grantmaking strategy. A Maryland Teacher of the Year and former educator, Josh brings a practitioner’s lens to the question of what coherent, high-quality AI implementation actually requires for students who have historically been underserved by systems that weren’t built with them in mind.

Sydney Ganon is the Director of Innovation at Education First, where she leads the firm’s AI strategy and works with school systems to build coherent approaches to AI adoption. Over the past three years, she has partnered with districts nationwide and collaborated with philanthropic organizations, including the Gates Foundation and the Chan Zuckerberg Initiative, on AI in education. She brings a change management and systems perspective to AI implementation, helping leaders move from experimentation to sustained organizational practice.