Dr. Matt Rhoads

Co-teaching myth: one teach, one float is a starting point, not a destination. If your model never evolved past October, that is the finding.

Most AI-in-education headlines don't say whether real students were involved. The Evidence Signal will. A free Tuesday brief that grades each study by design and who took part. First issue Nov 10. https://evidencesignal.beehiiv.com/subscribe

AI in the classroom should reduce teacher decision load, not multiply it. If a tool adds more choices, it is not helping.

If your co-teaching supports never fade, your students never own the skill. Plan the exit, not just the support.

An AI tutor with no research backing is a product demo, not pedagogy. Ask vendors for the evidence base before the pilot.

The best instructional coaches I know ask questions they do not have answers to yet. Coaching is not delivering feedback. It is thinking alongside a teacher until the next move becomes obvious. What is a question you could sit with this week?

The hardest part of instructional coaching is the first five minutes of the first classroom visit, when a teacher decides whether you are an evaluator in disguise. Trust is the work. On Kindle now: https://www.amazon.com/dp/B0HKTH4T65

Co-teaching fails quietly when one partner becomes a permanent assistant. We wrote Co-Teaching Evolved because the model has to be intentional or it collapses into proximity. Presence is not partnership. What does intentional look like in your building?

The Kindle edition of Crush It from the Start: 25 Tips for Instructional Coaches and Leaders is out today. For the coach who owns other adults' growth with no authority over them. Get it: https://www.amazon.com/Crush-Start-Instructional-Coaches-Leaders/dp/1962765083?tag=drrhoadseduca-20 #ad

The Kindle edition of my book, Crush It from the Start: 25 Tips for Instructional Coaches and Leaders, is out today. A playbook for the coach responsible for adult growth without authority. https://www.amazon.com/dp/B0HKTH4T65

Teachers tell me the same thing about AI tools: students finish faster and know less. That gap is the whole issue. Speed of completion and depth of learning are not the same outcome. What evidence would convince you a tool actually improved learning?

A district can log a million AI sessions and still not know a single thing about learning. Engagement metrics measure presence, not growth. Before you renew an AI contract this year, ask the vendor one question: what evidence shows students learned more? If they show you a dashboard, ask again.

The model is the access point. If your co-teaching model defaults to one teacher talking and the other managing, you are not co-teaching. You are staffing. The model you run decides what your students learn. What model is actually doing the teaching?

The question is not whether students use AI. It is when. Timed access beat unrestricted access in the study I keep citing. The norm you set in week one sets the year. When will you set the AI norm in your classroom?

New instructional coaches start the year with a portfolio full of frameworks and no clear first move. I keep telling them the framework is not the plan. The plan is the relationship, then the observation, then the feedback. What is your first move as a coach this year?

The most common co-teaching model in American schools is one teach, one observe. It is the safest model and the least useful one. We built Co-Teaching Evolved to move past it. The model you default to is the one that decides what your students learn. What model are you actually running?

A superintendent told me her AI dashboard showed login frequency up 40 percent this year. That is a good number. It is also the wrong number. I keep asking what the 40 percent bought in learning. The dashboard does not say. What would your district measure if it could only pick one learning outcome?

New coaches don't need more frameworks. They need a sequence: trust first, then instructional range, then systemic impact. My 25 Tips gives first-year coaches a 90-day plan from listening to leading change. amazon.com/dp/B0DZQRL7YC?tag=drrhoadseduca-20 #ad

We wrote 50 Tips for New Teachers with Dr. Fecich, Dr. Jakubowski, and Dr. Leichtman because the first year needs a sequence, not more theory. First 50 days, first grading period, first parent conference. Five min to read, one class period to apply. amazon.com/dp/1737864347?tag=drrhoadseduca-20 #ad

44% of new teachers leave within five years. The first year is the highest-risk period (Ingersoll et al., 2018). Our tips won't fix the structural problems behind that number. They help a new teacher stay long enough to benefit from the policy work above them. 50teachertips.com

New paper: AI safety alignment is often superficial. Adversarial framing (creative writing, role play) bypasses refusal mechanisms by hiding harmful intent in benign contexts. For K-12: surface refusal is not a safety policy. https://arxiv.org/abs/2608.20378

The first year of teaching is overwhelming because no one gives you a sequence. 50 Tips for New Teachers: 50 short, specific tips from four authors (Rhoads, Fecich, Jakubowski, Leichtman). Five minutes to read, apply the next day. 50teachertips.com

New paper proposes 'Slow AI' design principles (slower, lower-throughput, more durable) as a counterweight to maximalist generative AI. For districts: slow AI is a defensible procurement category. Ask vendors per-task energy cost. https://arxiv.org/abs/2608.20398

New paper: TTS dialogue lesson formats interact with learner style. No universally best format, concrete-experience and active-experimentation learners respond differently. AI as a portfolio of formats per lesson, not one format per learner. https://arxiv.org/abs/2608.20822

New paper: a diagnostic taxonomy of six LLM misconceptions teachers carry into AI classrooms. Deflationary vs. anthropomorphic folk theories. Useful for AI professional development. https://arxiv.org/abs/2608.20421

The transition from teacher to coach is harder than most districts admit. You move from a classroom where you control outcomes to a role where you influence them through other adults. If you are onboarding new coaches this fall, start with the relationship domain. Everything else depends on it.

New instructional coaches start the year with frameworks and no clear sequence for what to do first. 25 Tips for Instructional Coaches and Leaders is organized into four domains: relationships, coaching cycle, technology integration, and professional learning. matthewrhoads.com/books

Before August ends, every co-teaching pair should answer: which model are we using, and why? "One teach, one assist" underuses the second teacher (Friend & Cook, 2017). Dr. Karge and I lay out a three-stage progression matched to lesson type and student needs. coteachingevolved.com

Cognitive offloading supports the teacher's thinking. Cognitive outsourcing replaces it. The tool is the same. The professional judgment is the difference. A four-check audit (cognitive science, curriculum, pacing, slop detection) is what separates the two. matthewrhoads.com

134 bills in 31 states regulate AI in education because data leaves the device. Local inference changes that. Models can run on school laptops, free, with zero data transmission. The privacy compliance problem disappears. https://matthewrhoads.com/2026/07/18/local-llms-school-districts-lm-studio/

200 laptops. One afternoon. Zero vendor calls. LM Studio + an open-source model, pushed through MDM. No contracts, no licensing, no servers. The laptops are the infrastructure. https://matthewrhoads.com/2026/07/18/local-llms-school-districts-lm-studio/

Claude for Teachers can build personalized instruction overnight. But every AI-generated material should pass four checks before it reaches students: cognitive science, curriculum, pacing, and slop detection. https://matthewrhoads.com/2026/07/24/the-offloading-trap-audit-ai-instructional-output/

A school pays $14,400/year for AI tools. The open-weight equivalent is free, runs on laptops they already own, and offers better privacy by design. A 500-teacher district spends six figures for what's available at no cost. https://matthewrhoads.com/2026/07/18/local-llms-school-districts-lm-studio/

China has a binding K-9 AI curriculum standard and prescriptive classroom guardrails. The U.S. has a state patchwork and federal encouragement. Both bets are uncertain. https://matthewrhoads.com/2026/07/20/two-roads-china-us-ai-education-schools/

Chinese policymakers settled whether AI belongs in schools in 2017. U.S. educators are still arguing. China is now asking how, at what scale, with what guardrails, and how to fund the rural gap. https://matthewrhoads.com/2026/07/20/two-roads-china-us-ai-education-schools/

134 bills in 31 states regulate AI in education — all because data leaves the device. Local inference changes that. Models run on school laptops, free, zero data transmission. The privacy compliance problem disappears. https://matthewrhoads.com/2026/07/18/local-llms-school-districts-lm-studio/

California just shipped the first state-level model AI policy for K-12 schools. One state now has a template that 49 others can adopt, adapt, or ignore. The districts that move first will define the national pattern. cde.ca.gov/nr/el/le/yr26ltr0706.asp

25 Tips for Instructional Coaches and Leaders — coaching cycles, feedback conversations, teacher partnerships. Real growth, not checklists. 5/5 on Amazon. matthewrhoads.com/books

Four steps for districts building AI literacy: 1. Start with the learning problem, not the technology 2. Does the tool offload or outsource cognition? 3. PD: cognitive science first, tool second 4. Demand impact evidence, not usage data

"If a vendor cannot provide evidence of impact on student learning, that is the answer." Evidence-first AI literacy means demanding proof, not promises. matthewrhoads.com/2026/07/07/beyond-the-hype-an-evidence-first-framework-for-ai-literacy-in-education/

10 books for educators this summer — organized by your role. Coaches, co-teachers, leaders, specialists. Every one rated 5/5 on Amazon. Plus 22 free coaching templates. matthewrhoads.com/books