San Diego County Office of Education

It is not the what.
It is the how.

A cross-divisional task force for the responsible adoption of artificial intelligence, and the first ninety days that decide whether it works.

See the 90-day plan Find your gap
Dr. Marie Martin, Ed.D.
Candidate, Coordinator of Artificial Intelligence · September 2026

Nine divisions. Four outside voices. One table.

Dr. Marie Martin
Why I do this work

I want students to leave us as producers of this technology, not only consumers of it.

I wrote my dissertation on what it takes for a school system to move past the factory model, and the finding was not about technology. It was that leadership has to build the conditions before the practice can change. Everything on this page follows from that.

01

Community-based accountability. I design with the people who will use what I build, and I report back to them on what worked. The team is my rubric, not the dashboard.

02

Problem-based learning. The best learning starts from a real problem the learner already cares about, not from a list of features.

03

Human-centered design. Design starts with the human, not the content. Who is the learner, and what do they have to be able to do afterwards?

Dr. Marie Martin, Ed.D. Ed.D., USC Rossier · Founder and Chief Learning Officer, Alexandria's Design · Twenty years across K-12 systems, higher education, and industry
Why it exists

Every leader asks me the same question.

“What do we teach?”

The tool question. It changes every six months.

→

“How do we work now?”

The practice question. It changes the organization.

A task force organized around tools will be obsolete before it finishes its charter. A task force organized around practice will not. That is the whole design decision, and everything below follows from it.

Why now

Other systems started before the moment arrived.

1996
Estonia
Tiger Leap
2014
Singapore
Smart Nation
2018
Finland
Elements of AI
2022
ChatGPT
released
2024
UNESCO AI
competencies
2026
SDCOE stands
one up

The systems that moved fastest after 2022 were not the ones with an AI plan. They were the ones that already had a delivery system. Estonia laid digital public infrastructure in 1996, so it could stand up a national AI program in seven months. South Korea moved early, funded teacher training at roughly 740 million dollars, mandated it onto teachers anyway, and in August 2025 the program lost its legal standing. Early is not the same as done well. The difference was whether educators helped build it.

The divide, honestly

We solved the first level. We never funded the third.

Level 3

The Educator

Mindset, time, capability, and permission to redesign the work. This is where learning actually happens.

Barely funded
Level 2

The Recurring Cost

Peripherals, consumables, subscriptions, licenses, and the hardware refresh cycle. A device is not a one-time purchase.

Under-funded
Level 1

Connection and Devices

Broadband, hotspots, one-to-one devices. Largely closed through the pandemic response.

Largely funded
20–40%of students in this county under-connected at home (SANDAG, 2021)
100,000+San Diego households lost the federal broadband subsidy in 2024
42% / 97%broadband access, unincorporated county vs urban
46% / 3%English learners, National Elementary vs San Dieguito. Same county.

Targeted Universalism, already named in SDCOE's Strategic Plan, is the right instrument. One universal goal for the county, pursued through different strategies, because Borrego Springs and Del Mar are not standing on the same level of this stack.

The capability gap

Everyone has been trained at 101. The middle is empty.

Pick a role. Then move through the levels. The gap is not abstract once you see what it means for a person who actually works here.

Where the professional learning actually went

This is not an SDCOE problem. It is in nearly every organization. Introductory sessions went wide, technical teams went deep, and almost no one built the middle. RAND found district-provided AI training rose from 23 percent to 48 percent between fall 2023 and fall 2024, but 67 percent in low-poverty districts against 39 percent in high-poverty districts, and nearly all of it optional and one-off.

Three colleagues working together around a laptop

The middle is where the work actually changes.

Not in an awareness session, and not in the server room. At the desk where somebody does the task differently on a Tuesday.

Photograph licensed via Pixabay
The sequence
Technology Pedagogical Practice Relationships and SEL

Cultural being and equity, present at every level

Every failed rollout I have studied inverted this.

Devices arrive first. Practice is assumed. Adult culture is never addressed. Then the tool underperforms and the conclusion is that the tool failed.

The tool did not fail. The base failed.

This task force works bottom up. That is the whole reason the first thirty days are listening and not procurement.

What it is

Who sits at the table decides what the table can see.

Select a seat to see why it is there.

Named in the classification for this role
Two I would add
North Star AI Task Force · advisory to the Senior Leadership Team
The seats most task forces seat last
Select any seat above

Twelve to fifteen members. A two-year charter with staggered terms, so it survives a change of leadership. It advises. It does not purchase, and it does not police.

The precedent

We have done this before

SDCOE already leads a countywide risk domain it has no authority over. Cybersecurity. One accountable owner, shared free tooling, countywide training, and a 2024 Grand Jury commendation for it. AI is the same shape. This is not a new competency, it is an existing one pointed at a new risk.

The risk

Sequence, not technology

The county's largest district built a 65-member AI task force, held a town hall, and still had not adopted a policy. Its superintendent has said publicly that the cause was insufficient engagement with teachers and labor. That is why labor partners and families come in at week two, not week ten.

The first 90 days

Ninety days. Three moves. No procurement.

The phase names are SDCOE's own, borrowed from the LCAP Development Series on purpose. This window sits inside the 2026-27 LCAP development cycle, so districts can write AI access, AI literacy, and professional learning into three-year plans with funding attached. Miss that window and the next clean one is three years out.

Days 1–30

Reflect and Learn

  • Labor partners and families in week two
  • Listening sessions in all nine divisions
  • An honest inventory of the AI already in use here
  • Readiness baseline across the three levels of the divide
  • Student and classified staff voice, not only management
Readiness Baseline
Days 31–60

Design and Align

  • Co-design studios, not briefings, with each division
  • Guardrails built on the CDE model policy, adapted for this county
  • Role-based capability map that fills 201 and 301
  • Industry roundtable on what the work now requires
  • Spanish-language and interpreted sessions throughout
Draft Guidance and Capability Map
Days 61–90

Develop and Refine

  • Three small pilots in three different divisions
  • Measure against the baseline and publish what did not work
  • Board-ready recommendation to Senior Leadership
  • Year-one plan with named owners and honest cost
  • Existing staff time. No new headcount.
Board-Ready Recommendation
4outcomes SDCOE already reports that this body moves
9divisions at the table, seven named plus two added
0dollars of new spend proposed in the first 90 days
1page a superintendent can actually use
Students celebrating graduation

Ninety days is not a long time. It is one cohort closer.

Every semester we spend deciding is a semester a student spends on the wrong side of the third level of the divide.

Photograph licensed via Pixabay
The brief, in three minutes

The task force, summarized.

What other systems did differently

They were earlier, not faster. And early is not the same as done well.

Estonia1996 → 2025

Tiger Leap laid digital public infrastructure in 1996. AI Leap was announced February 2025 and went live that September, seven months later. Teachers were trained before students got access. Deployed on ChatGPT Edu, tailored with University of Tartu researchers to behave as a Socratic tutor. No outcome data yet. It is a conditions story, not a results story.

Singapore1997 → today

Unbroken five-year ICT masterplan cycles since 1997. A national learning platform in every school from 2018, a device in every secondary student's hands by 2021, and 100 hours of professional development entitlement per teacher per year. AI did not need a new channel. It slotted into one that already existed.

South KoreaThe cautionary tale

Earliest and best funded, at roughly 740 million dollars for teacher training over 2024-26. Still failed. Adoption was mandated onto teachers rather than built with them; 80 percent of teachers reported being negative on it. In August 2025 the National Assembly removed AI digital textbooks' legal status as textbooks, ending guaranteed funding and mandatory use. Adoption fell to about 19 percent. Any initiative that cannot survive a change of leadership is not an initiative. It is a term of office.

Finland2018

Elements of AI reached hundreds of thousands of adults voluntarily and about 40 percent of enrollees were women. It also hit roughly 0.055 percent of the EU population against a 1 percent target. Voluntary access is not a delivery system.

The United StatesAhead and behind

Behind on delivery: roughly 13,000 districts, no national platform, no national teacher-development architecture. Ahead on substance: American work wrote the curriculum framework other systems borrowed, and 60 percent of US teachers adopted AI without any mandate, with weekly users reporting about six hours a week saved. No ministry produced that by decree.

The five conditions that separated the systems that worked: a platform that already reached every school; teachers trained before students got access; continuity across political terms; pedagogy stated before tools; and a named owner with authority over curriculum, platform and teacher development at once. That last one is the argument for why a county office is the right layer for this work in California.

Supporting material

The work behind the argument.

Sources

The evidence base

Every figure on this page, with its source and its date. Including what could not be verified and is therefore not claimed.
Detail

The 90-day plan in full

Week by week, with deliverables, the people in each session, and the measures each phase reports against.
Portfolio

Industry and district work

Learning design and AI-enabled programs across biotechnology, healthcare, defense, publishing, and K-12 systems.
Professional learning

The 101 to 401 ladder

The AI capability curriculum this argument comes from, including the 201 and 301 material most organizations never build.
White paper

The Teaching Underneath the Technology

A LEVER Framework white paper on skill formation, the digital divide, and appropriate use of technology in the AI era.
Research

The Fourth Industrial Superintendent

Doctoral research, USC Rossier, on how a superintendent moves a system past the factory model. Leadership creates the conditions. The practice follows.

The factory model of schooling was built for an economy that no longer exists.

This is the first moment in a century when we have both the tools and the reason to change it. I would rather this county spend ninety days deciding how we work than deciding what to buy.

Our students should leave us as producers of this technology, not only consumers of it.